THE DATA-DRIVEN SCIENCE OF PREDICTING

AND CHANGING HUMAN BEHAVIOR

MINDMASTERS

SANDRA MATZ
HARVARD BUSINESS REVIEW PRESS
BOSTON, MASSACHUSETTS
OceanofPDF.com
 HBR Press Quantity Sales Discounts
Harvard Business Review Press titles are available at significant quantity discounts when
purchased in bulk for client gifts, sales promotions, and premiums. Special editions,
including books with corporate logos, customized covers, and letters from the company or
CEO printed in the front matter, as well as excerpts of existing books, can also be created in
large quantities for special needs.
For details and discount information for both print and ebook formats, contact
[email protected], tel. 800-988-0886, or www.hbr.org/bulksales.

Copyright 2025 Sandra Matz

All rights reserved
No part of this publication may be reproduced, stored in or introduced into a retrieval
system, or transmitted, in any form, or by any means (electronic, mechanical, photocopying,
recording, or otherwise), without the prior permission of the publisher. Requests for
permission should be directed to [email protected], or mailed to
Permissions, Harvard Business School Publishing, 60 Harvard Way, Boston, Massachusetts
02163.
The web addresses referenced in this book were live and correct at the time of the book’s
publication but may be subject to change.
Library of Congress Cataloging-in-Publication Data
Names: Matz, Sandra C., author.
Title: Mindmasters : the data-driven science of predicting and changing human behavior /
Sandra Matz.
Description: Boston, Massachusetts : Harvard Business Review Press, [2025] | Includes
index.
Identifiers: LCCN 2024023257 (print) | LCCN 2024023258 (ebook) | ISBN 9781647826314
(hardcover) | ISBN 9781647826321 (epub)
Subjects: LCSH: Psychology—Research—Methodology. | Psychology—Technological
innovations. | Technological innovations—Psychological aspects. | Human behavior—
Research. | Data mining—Social aspects.
Classification: LCC BF76.5 .M373 2025 (print) | LCC BF76.5 (ebook) | DDC 150.72—
dc23/eng/20240905
LC record available at https://lccn.loc.gov/2024023257
LC ebook record available at https://lccn.loc.gov/2024023258
ISBN: 978-1-64782-631-4
eISBN: 978-1-64782-632-1
The paper used in this publication meets the requirements of the American National
Standard for Permanence of Paper for Publications and Documents in Libraries and Archives
Z39.48-1992.

OceanofPDF.com
 To Moran and Ben, whose love and laughter
fill my life with magic every day.
OceanofPDF.com
 CONTENTS

Introduction: The Digital Village

The origins of psychological targeting

PART ONE

DATA IS A WINDOW INTO OUR PSYCHOLOGY

1 Decoding Our Psychology
How algorithms can get to know us better than our closest friends do

2 The Identities We Craft Online

Tell me what you like and I’ll tell you who you are

3 The Digital Breadcrumbs of Our Existence

Everything we do is data

4 You Are Not Yourself When You’re Hungry

How context shapes who we are

PART TWO

THE BRIGHT AND DARK SIDES OF PSYCHOLOGICAL TARGETING

5 Psychological Insights in Action
An inside look at psychological targeting—from Cambridge Analytica to beauty products
(and back)

6 Finding the Good

Improving our finances, mental health, and maybe even our political climate

7 When Things Go Wrong

Why we stand to lose more than just our privacy

PART THREE
 MAKING OUR DATA WORK FOR US
8 We Need More Than Control
Why being in charge of our own data isn’t always a blessing

9 Creating a Better Data Ecosystem

How to make it harder for third parties to exploit our data (and easier for them to serve
us)

10 Coming Together
What wine co-ops can teach us about new forms of collective data management

Epilogue: The Moral Imperative to Shape Our Future

Appendix A
Appendix B
Notes
Index
Acknowledgments
About the Author

OceanofPDF.com
 INTRODUCTION

The Digital Village

I had begged my boyfriend for weeks to let me ride his motorcycle. A

Suzuki Bandit 600. In a deep, shimmering red. It was beautiful.
We loved to take the bike on adventures together. Up the
serpentine roads of the mountains surrounding the small village I
grew up in, and through the curvy roads of the countryside. But I was
tired of being in the back. I wanted to be in the driver’s seat.
When he finally agreed to let me try, I found an abandoned
military airfield nearby.
His instructions were simple: “Let’s sit on the bike first. You in the
front and me in the back. And then I’ll explain.” I was enthralled but
nervous, too. I was only fifteen, and I didn’t have a license. “Don’t
worry,” he assured me, “I’m sitting right behind you.”
I don’t know exactly what happened next. I remember that we
somehow rolled into the grass on the side of the airfield. When I tried
to pull the bike back out, I must have accidentally twisted the throttle
and let the clutch snatch.
A few seconds later, the front wheel of the motorcycle rose into the
air like a rearing horse. My boyfriend was thrown off the back (so
much for “I’m sitting right behind you”), and I sped off without the
slightest idea of how to control the bike.
Without thinking, I pulled and pushed the bike’s handlebars, trying
to keep my balance. For what felt like an eternity, I swerved left, right,
left again, before I came crashing down on one side, sliding for a few
feet before coming to a stop.
 We were lucky. Neither of us got injured. And nobody had
witnessed the accident. But there was a problem: we were in the
middle of nowhere and the bike wouldn’t start.
After weighing our options, I sat down on the grass to calm my
nerves, took a deep breath, and hit the call button on my phone. Part
of me was hoping no one would pick up. With every ring, my heart
started beating louder, my mind racing. When I was almost ready to
hang up, my dad answered the phone.
“Hi, Dad, … uh … we had a motorcycle accident. But don’t worry,
we were going slow, and we’re both fine.”
“It was you driving, right?”
In addition to a serious conversation with my parents (who were
actually surprisingly cool about it), I had to pay for the repairs. A full
year’s salary of tutoring gone. Painful, but not the worst part.
The minute we dropped the bike at the local repair shop, the news
about my misfortune spread through my village like wildfire. It was
the perfect story. Not just because I was fifteen and didn’t have a
license. I also happened to be the daughter of a local police officer.
There was no place to hide. The next day, on my way to the school
bus, Mr. Werner from across the street waved me over to inquire if I
was doing OK. He had heard about the crash. I got trapped in a ten-
minute recounting of his own teenage offenses.
A few houses down, Ms. Bauer looked up from weeding the little
garden in the front of her house, shaking her head. How could I be so
irresponsible? She had always thought of me as a smart girl.
The motorcycle crash was no longer my own private
embarrassment. Everyone—and I mean everyone—knew my business.

The Village Paradox

Welcome to Vögisheim! A tiny village in the southwest corner of
Germany surrounded by pretty vineyards, fields, and rolling hills.
Population: five hundred. Restaurants: two. Churches: one. Shops:
zero.
 I was born and raised there. Just like my mom, her mom, and her
mom’s mom. I lived there for the first eighteen years of my life—and
for what felt like an eternity as a teenager craving the stimulation of
busier places. Vögisheim is where I spoke my first words, took my first
steps, fell in love for the first time, had my first heartbreak, decided to
travel the world, and eventually embarked on a journey to study
psychology.
As in any small village, the other 499 residents of Vögisheim didn’t
just know about my motorcycle crash. They knew every little detail of
my life. They knew that I loved listening to the Ramones, that my
favorite place for weekend nights was the local pirate bar, and that I
simply couldn’t stand my geography teacher.
Those details alone might not have felt that intrusive. But, as if I
were a human puzzle, my neighbors put the pieces of my existence
together to construct an intimate picture of my inner mental life: my
hopes, fears, dreams, and aspirations. They seemed to truly know me.
Which allowed them to do what village neighbors do best … offer
(un)solicited advice and interfere with my personal life.
For me, this meant two things. On the one hand, I felt supported
by a community of people who understood me. They knew I was
ambitious and longed for a life outside the village. So, when the time
came for me to figure out what to do after high school, they were there
to offer advice and opportunities; they passed on my curriculum vitae
to friends and helped me decide whether a gap year was the right
choice for me.
On the other hand, I felt exposed and manipulated by the same
community. It was a poorly kept secret that I had a hard time saying
no to people. This made me an easy target for anyone who needed a
favor. Moving apartments? Ask Sandra. In need of a ride home from
the club? Ask Sandra (if the vehicle of choice wasn’t a motorcycle, of
course).
Growing up being seen by others was a blessing and a curse at the
same time.
 From the Village to the World
I left Vögisheim after graduating from high school and today live in
New York City where I am a professor at Columbia University.
A difference like day and night. I barely know my neighbors. And
they barely know me. We say hi to each other when we meet in the
corridor. But they don’t know what I do for work. They don’t know my
friends and family. And they certainly don’t know anything about my
deepest fears or aspirations.
But as it turns out, you don’t have to live in a small, rural
community to have someone watch and influence every step you take
and choice you make. That’s because we all have digital neighbors.
Think of it this way: the data-crawling digital equivalent to my
sixty-year-old neighbor Klaus reads my Facebook messages, observes
which news I read and share on X/Twitter, collects my credit card
purchases, tracks my whereabouts via my smartphone’s GPS sensor,
and records my facial expressions and casual encounters using some
50 million public cameras across the United States.
In the same way my neighbors became expert snoopers and
puppeteers over time, computers can translate seemingly mundane,
innocuous information about what we do into highly intimate insights
about who we are and ultimately prescriptions of what we should do.
I call this process of influencing people’s thoughts, feelings, and
behaviors based on their predicted psychological characteristics
psychological targeting. And I’ve been studying it—and practicing it—
for over a decade now.
My colleagues and I have published numerous articles showing
how computers—powered by machine learning and AI—can get to
know you intimately. It doesn’t matter which psychological trait or
data source you pick. For example, algorithms can tell whether you
are excited, sad, sociable, or anxious by tapping into your phone’s
microphone or camera. They can predict your income from your
social media posts. And they can tell whether you are likely to develop
depression or suffer from schizophrenia by tracking your GPS
location.
 But that’s only half the story. I’ve spent most of my career tackling
the glaring “So what?” question. What does it mean that computers
can peek into our psychology and understand what lies below the
surface of the behaviors they can observe? What does it mean for you
and me? And for society at large? It doesn’t take much imagination to
understand that psychological targeting, in the wrong hands, could be
a powerful weapon.
When I was a teenager, I struggled with low self-esteem. I wanted
nothing more than to belong and be liked. But it was my best friend
who was popular, not me. I became very good at hiding my self-
doubts from the other people in the village, putting on a facade that
bordered on arrogance. On the outside, I was strong and confident.
Inside, I doubted myself. I shared these feelings in my diary.
If I were a teenager today, I would probably ask Google for advice.
“How can I become more popular?” “How do I feel better about
myself?” These questions would build up in my search history. And
the resulting profile could easily be used against me. In 2017,
Facebook was accused of predicting depression among teenagers and
selling this information to advertisers.1 No easier target than insecure,
struggling teens. Pretty gloomy.
But let’s look at this in a more positive light. What if we could use
psychological targeting to help millions of people lead healthier and
happier lives? My research, for example, has been used to predict and
prevent college dropouts, guide low-income individuals toward better
financial decisions, and detect early signs of depression.
Yes, that’s right. The very thing I accused Facebook of doing in the
“gloomy” section could also be a real opportunity. Depression affects
approximately 280 million people around the world. Every year,
about 1 million of them commit suicide. That’s more people dying
from the consequences of depression than from homicide, terror
attacks, and natural disasters together.
What makes these numbers particularly upsetting is that
depression is treatable. The problem is that many people are never
diagnosed. Even if they are, the diagnosis often arrives too late. It is
much harder to fight your way back from the bottom of the valley
than from the initial descent.
 What if, instead of selling you out to advertisers, we used the
insights into your mental health profile to build an early warning
system? GPS records or tweets could alert you to changes in your
behavior that resemble patterns observed in other people suffering
from depression. It’s not only a chance to detect depressive symptoms
early (before they develop into a full, clinical depression) but also to
offer personalized advice or resources.
We might observe that you are not interacting with your friends as
much anymore, or you’re spending a lot more time at home than
usual. Why not encourage you to reach out to a few of your friends or
spend some time in the park nearby? And, if necessary, provide you
with contact details of a few therapists in the area that might be of
help.
Predicting and influencing mental health outcomes is merely one
of many examples demonstrating the power of psychological
targeting. What if we could make education more engaging, help
people achieve their fitness goals, or facilitate a more constructive
dialogue across the political divide?
For the better part of my academic career, I’ve felt somewhat
helpless and lost in the tension between the perilous and promising
sides of analyzing personal data. Was I in the camp of techno
pessimists arguing that technology fails to deliver on its promises and
actively harms humanity? Or was I in the camp of techno optimists
who believe in a bright future where technology helps us become
better versions of ourselves?
I often felt like a hypocrite—excited about new findings, with this
nagging feeling that, in the wrong hands, those findings could have
horrible consequences. Or vice versa, talking to media about the
dangers of psychological profiling, while fearing I was backstabbing
my students and industry partners who saw the potential promises of
psychological profiling.
It wasn’t until a Christmas trip back home (and after multiple
rounds of mulled wine) that I realized how similar my current
struggle was to my experience in the village—constantly torn between
the desire to break free and the appreciation for what my community
had to offer. The more I thought about this analogy (in a sober state),
the more glaringly obvious it became.
I was dealing with a new manifestation of a tension that has been
part of the human experience for centuries. How much of our private
lives are we willing (or even happy) to disclose to those around us?
How much of our privacy and autonomy are we willing to give up for
the security and strength provided by the collective?
What this all comes down to is power. In the same way my
neighbors had an easy time convincing me to do chores for them
because they knew I was a crowd pleaser, understanding your
psychological needs, preferences, and motivations gives others power
over you. Power to influence your opinions, emotions, and ultimately
behavior. Sometimes this is good; sometimes it’s bad.
But life in the village taught me that whether we win or lose is—at
least in part—up to us. Even though I never had full control over my
life, I still managed to navigate the ups and downs. As a kid, I had no
idea how the village operated. But over time, I learned more about the
system I was embedded in. I understood people’s motivations, figured
out who was talking to whom, and learned who could be trusted with
information.
Once I understood the game that was played and had a clear sense
of what I wanted out of it, I learned to play it to my advantage.
Suddenly, I was winning more than I was losing.
We need to do the same—and more—for the digital village. We
need to understand the players that control the current data
ecosystem, figure out how they use our personal data for and against
us, and identify the leverage we have (or need) to come out on top.
But merely becoming better at playing the game won’t be enough.
We need to redesign it.

Redesigning the Data Game

I have argued that the tension we experience in today’s data-driven
world is the same as the one our ancestors experienced two thousand
years ago, or the one I struggled with in my village. But that’s not
entirely true.
While I was exposed to the prying eyes of villagers (and maybe
their friends in the adjacent villages looking for juicy gossip), our
digital behavior today exposes us to the entire world. The rules of the
game have changed.
Growing up in a village, my neighbors knew a lot about me. But
trust me, I also knew a lot about them. I knew who was struggling
with alcoholism, who was unhappy in their marriage, and who was
evading taxes. We all played the game as equals. We all paid the price,
and we all benefited.
Today’s data game looks nothing like this. Its rules are opaque, and
its playing field is highly tilted. There are a few people and
organizations that know an awful lot about many of us and that
benefit greatly from this knowledge. We may get certain things from
the exchange (free-of-charge search engines and social media, for
instance), but we don’t get reciprocal knowledge about the people and
organizations that track us so zealously.
Even though the game we play in the digital village might have
slightly different rules, the best starting point to stack the deck in our
favor remains the same. We need to understand what novel predictive
technologies such as psychological targeting are capable of and
collectively decide which applications are conducive to a thriving
society, and which aren’t. Once we do, it is within our control to
design a system that amplifies the positive sides of psychological
targeting and makes it work for us, instead of against us.
Mindmasters is an invitation to do just that—an invitation to join
an informed, nuanced discussion of psychological targeting. Cases like
Cambridge Analytica’s alleged interference in the 2016 US
presidential election (a story I helped break and will talk more about
in chapter 5) have caught the public eye and informed much of the
public debate on the topic.
My goal for Mindmasters is to pull back the curtain, separate
narrative from facts, and offer a science-based account of
psychological targeting. I will do so across three parts, each touching
on an important puzzle piece of the overall process.
 Part 1 takes you on a journey through how computers learn to
translate your digital footprints into intimate predictions of who you
are: your personality, sexual orientation, political ideology, mental
health, moral values, and more. We’ll open a digital window into our
psyche by entering the worlds of deliberately shared identity claims
(e.g., Facebook likes, social media posts, and pictures) and innocuous
behavioral residue (e.g., Google searches, credit card data, and GPS
records). And we’ll explore the role of contextual cues in giving away
information about who we are at any given point in time.
As we open the black box and look under the hood of some of the
AI-powered predictive models that my colleagues and I have built
over the years, you will realize that it doesn’t take rocket science to
translate what we do online to who we are on the inside.
To be clear, computers don’t necessarily need to translate your
behavior into psychological profiles to know you intimately and
interfere with your choices. Many of the potential benefits and
dangers I discuss in later parts of the book apply to the use of
personal data more broadly. But cases like Cambridge Analytica have
caught the public imagination because they allow us to relate to our
data in a fundamentally human way. You don’t think of yourself as a
combination of spending records, GPS-tracked longitude and latitude
coordinates, and Google searches. You think of yourself as extroverted
or introverted, liberal or conversative, and cooperative or competitive.
Part 2 directly builds on the insights from part 1 to discuss the
glaring “So what?” question. Why should we care about the
proliferation of technologies like psychological targeting? What does
it mean for us—and society at large—that algorithms can decode the
inner mental lives of millions of people and alter the way they think,
feel and behave? Should we be scared or elated?
I will argue that we should be both. As the tech historian Melvin
Kranzberg famously said: “Technology is neither good nor bad, nor is
it neutral.”2 The exact same mechanisms can be used to accomplish
diametrically opposed goals. By tapping into your psychology, I can
get you to buy products you might not need, but also to save more
money for a rainy day. I can exploit your emotional vulnerabilities but
also help you overcome them. And I can reinforce your existing
worldviews but also encourage and enable you to expand them.
The impact of psychological targeting ultimately depends on us
and the choices we make. At its worst, psychological targeting
manipulates, exploits, and discriminates. At its best, it engages,
educates, and empowers.
As advanced AI technology—including generative AI—makes the
creation and targeting of hyperpersonalized content easier than ever
before, we need a clear vision for how to amplify the opportunities
afforded by psychological targeting while mitigating its risks. That’s
what part 3 of the book is all about. How do we redesign the data
game to create a better future for all of us?
I will argue that creating this future requires us to return to the
village. Not literally. No need to pack your bags, take the kids, and
move to your version of Vögisheim. I’m talking about small village-
style communities designed to help you manage your personal data.
Entities that are legally obliged (e.g., through fiduciary
responsibilities) to act in the best interest of their members. A data
trust or data co-op.
Today’s data landscape is simply too complex to fight this fight
alone. I might have been able to look after myself in the game we
played in the village, but I don’t stand a chance in today’s global
arena. And neither do you. No one has the knowledge, time, and
energy to manage their personal data all by themselves.
We need allies. Like-minded people who have similar interests and
share the same goals. Expectant mothers, for example, sharing their
medical and biometric data with one another to figure out the best
nutrition for a safe pregnancy. Or educators pooling performance data
from their classrooms to develop more effective teaching strategies.
Unlike village neighbors, your digital allies don’t have to live in the
same place as you. Technology solves that problem (and many more,
as I will describe in more detail later). With about eight billion people
around the world, you will eventually find someone with the same
problems and values as you.
What I am suggesting isn’t simply a return to the old ways of the
village. It’s not just a one-to-one translation of tried-and-tested
solutions to a new problem. The game we play online today is
different from the game I played in Vögisheim, and so are the
solutions. The good news is that if we get this right, we might be able
to have it all: the benefits that come with letting others into our lives
without the costs of losing our privacy and self-determination.
Although Mindmasters centers around data and technology, it is,
at its core, an exploration of the human experience: how we want to
both reveal and conceal, how we gain and lose by letting others into
our lives, and how new technologies like psychological targeting
require us to rethink the social contract. It’s as much an attempt at
sharing my learnings with you as it is an invitation for you to join the
conversation. To become part of redesigning the game.
OceanofPDF.com
 PART ONE
DATA IS A WINDOW INTO OUR PSYCHOLOGY
OceanofPDF.com
 1
Decoding Our Psychology

It’s 8 a.m. on a Monday morning. As usual, it’s not my alarm clock

that wakes me up, but my dog Milou licking my face (a bit gross, I
know). I give her a squeeze, push her to the side, and reach for my
phone.
I check my WhatsApp and Facebook, scan my emails, and catch up
on the latest news on CNN. Before taking Milou out, I put on my
Fitbit to make sure that every step I take counts toward my daily goal
of ten thousand steps. I’m tempted to tie my Fitbit to her collar, but
resist. Don’t judge me. According to my friend Alice Moon’s research,
I’m not the only one who cheats on their step count.
Milou and I go for a walk across campus where she can run around
off leash and harass the early-riser students, who love the free
cuddles.
After a quick shower at home, I grab my phone and wallet and
head out to work. On the way, I pick up a matcha latte and a croissant
from the deli around the corner. I arrive at the office shortly before 9
a.m.
In less than one hour, I have generated millions of digital
footprints. On a server somewhere around the world, there are now
digital records of the messages I have sent and received (incoming:
cute pictures of my nieces; outgoing: cute pictures of Milou), the fact
that I checked in with Facebook from my home location, spent about
ten minutes reading five different articles on CNN’s website, I took
about two thousand steps walking across the Columbia University
campus in Manhattan, and got an unhealthy deli breakfast.
In addition, the sensors in my phone have registered that there was
physical activity starting from 8 a.m. and have tracked my GPS
location continuously. The cameras on the corners of streets, outside
the deli, and inside the elevators have collected visuals of me, telling
them exactly where I have been at any point in time, whether I was
alone or accompanied, and whether I looked happy or not (at 8 a.m., I
rarely do).
Like the average person, you and I generate about six gigabytes of
data every hour.1 Every hour! Just imagine how many USB drives you
would need to save all the data that accumulates over the course of
your lifespan. And that is just your data.
Worldwide, there are now an estimated 149 zettabytes of data (that
is 149,000,000,000,000,000,000,000 bytes), with numbers
doubling every year. If you were to store all this data on CD-ROMs
and stack them on top of each other, you would reach far beyond the
moon. In fact, some have suggested that today there are almost as
many digital pieces of data as there are stars in the vast universe.
Romantic, isn’t it?
Each of these data points represents a little puzzle piece of who we
are. My Spotify playlist, for example, reveals that I love techno and
Taylor Swift. My credit card history, that I enjoy traveling. And my
GPS records, that I love to go for long walks in the park.
Individually, these pieces aren’t that meaningful. Just like in a
puzzle, you start with a pile of disconnected chaos. But once you put
the pieces together, you gradually begin to see the full picture and
understand its meaning. The same is true for data. Once connected,
our digital traces provide a rich picture of our personal habits,
preferences, needs, and motivations. In short: our psychology.

The Internet Knows You Better Than

 Your Spouse Does
In 2015, the Financial Times ran an article with the provocative
headline, “Facebook understands you better than your spouse.”
Sounds like the opening to a dystopian science fiction novel? Nope.
It’s the result of a real scientific study published by my former
colleagues at the University of Cambridge.2
The research team led by Youyou Wu had built a series of machine
learning models that could translate a person’s Facebook likes into
personality profiles. The results were astonishing: after observing just
ten likes from someone’s Facebook profile, their model was able to
judge a user’s personality better than their work colleagues. Sixty-five
likes? It knew users better than someone’s friends. A hundred-twenty
likes? Better than family members. And three hundred? Better than
their spouse.
When my colleagues first told me about their findings, I was sure
they had made a mistake (and so were they, initially). Clearly, there
was a bug in the code.
But there wasn’t. My colleagues were right. Seven years later, I am
still amazed by their findings. We call our spouses our “other halves”
for a reason. They often have years of data on us. They plan,
experience, and live life with us every day. And yet, with access to just
about three hundred of your Facebook likes, a computer can know
you just as well or even better.
This puts the snooping skills of my village neighbors to shame.
Even the most curious among them probably knew far less about me
than any semiskilled computer scientist or engineer with access to the
right data. Today, a fifteen-year-old kid in the basement of their
parents’ home could figure out more about me than all my village
neighbors together.
But how do computers become such master snoopers? How do
they make sense of the vast, unstructured sea of digital footprints to
paint a picture of the person behind it? The simple answer is: they
observe and learn (yup, big shocker, that’s why it’s called machine
learning).
 Let me illustrate this process with an example that has absolutely
nothing to do with computers or algorithms. My main protagonists
are chickens. Baby chicks, to be precise.

A Villager’s Guide to Machine Learning

Have you ever heard of sexing? Don’t worry: this topic is safe for
work.
Chick sexing refers to the practice of distinguishing between
female chicks (pullets) and male chicks (cockerels). Large commercial
hatcheries use it to separate the high-value female chicks from the
male chicks almost immediately after birth.
While the female chicks are used for egg production, the male
chicks are usually killed to reduce unnecessary cost for the hatchery
(and suddenly becoming a vegetarian doesn’t sound so bad anymore).
The act of sexing is done by experienced personnel—sexers—who
have to decide within a few seconds whether a chick is female or male
by examining the vent in the chick’s rear (known as “vent sexing”).
As it turns out, this is not an easy task. The genitals of newborn
chicken are almost indistinguishable by eye, and there are so many
exceptions that it’s practically impossible for even the most
experienced sexers to explain their decision-making process. After
years of training, they simply know. But how do they learn to
distinguish between male and female chicks in the first place? Trial
and error.
Imagine you have just started as a sexer in a major hatchery. It’s
day one and you are excited to start your new job. But there is no
instruction manual, no fifty-page report or PowerPoint deck to
introduce you to the wonders of chick sexing. Instead, you are teamed
up with an experienced chick sexer who stands right next to you,
quietly observing.
You pick up the first chick and examine its rear. Of course, you have
no idea; it’s your first day at work and your experience with chick
vents has been, um, limited.
 You shrug and put the chick into the pullets bin. Your mentor says,
“Yes.” Success. You pick up the next one and after a short examination
put it in the cockerels bin. Your mentor says, “No.”
Your first day at work won’t feel very satisfying—your chance of
making the right choice will likely hover just above 50 percent (that’s
a coin flip). But after a couple of weeks of running through the trial-
and-error game with your mentor, your brain will have been trained
to accurately distinguish between male and female chicks. You have
become a sexing master! Just like your mentor, the rules guiding your
decision-making might be too complex to articulate, but you have
nevertheless internalized them.
Computers learn in the same way: trial and error. You throw a lot
of examples at them and give feedback on whether their predictions
are right or wrong. Doing so will gradually allow the algorithm to
learn how the input (a chick’s rear or a set of Facebook likes) is related
to the output (a chick’s gender or a user’s personality traits).
Do your Facebook likes include content about Oscar Wilde,
Leonardo da Vinci, and Plato? You’re probably intellectually curious
and open-minded. Accounting, MyCalendar, and national law
enforcement? Most likely organized and reliable.
The more data you have to run through the trial-and-error game,
the better the computer will become at turning educated guesses into
highly accurate predictions. That’s exactly what my colleagues did
when they conducted their man-versus-machine experiment. They
collected a large dataset of over eighteen thousand Facebook users,
combined their likes with self-reported personality profiles, and wrote
a few lines of code to automate the trial-and-error learning process (a
trivial task that can be done with user-friendly commercial software
today).
In the next few chapters, I will take you on a journey through
different types of digital footprints and show you how much they can
reveal about who you are. Some of these footprints, such as your social
media profiles, are created intentionally (chapter 2). Others, such as
the GPS records extracted from your smartphone, are mere by-
products of your interactions with technology (chapter 3).
 But they all have one thing in common: they offer a fascinating
window into your psychology—the aspects of your identity that define
who you are beyond what is visible to the naked eye. We’ll venture
into the worlds of political ideology, sexual orientation, socioeconomic
status, mental health, cognitive ability, personal values, and more.
But most of our time together will be spent in the world of
personality traits. It’s where most of the existing research lives,
including my own. And it’s also the world that has received the most
public attention and scrutiny (think Cambridge Analytica).
Because personality is such a popular destination, I want to get us
all on the same page about what to expect there—a crash course for
aspiring master snoopers, if you want (if you know all about
personality, feel free to skip ahead to chapter 2).

The Big Five Personality Traits

Like most people, you probably have an intuitive concept of
personality that guides your everyday behavior and social
interactions.
In my school, it was clear that Vera was the party animal, while I
was, well, the nerdy geek who went home at 11 p.m. when everybody
else was still out dancing.
In my village, we all explained the butcher’s frequent outbursts of
anger with his impulsive and irritable character. And at university, we
all predicted that Anne would become a successful lawyer as a result
of her competitive nature.
Although such lay theories of personality help us navigate our
social world, they are often implicit and only loosely defined. You
might not be able to fully explain why you think a particular person is
irritable and you might not be consistent in your terminology.
Sometimes you might label the same behavior impulsive, other times
grumpy or angry.
In contrast to lay theories, scientific models of personality provide
a structured approach to describing how people differ from one
another in the ways they think, feel, and behave. Rather than
accounting for the full complexity of someone’s identity, they provide
pragmatic approximations of what most people are like.
The results of a personality test, for example, will tell you that my
school friend Vera is highly extroverted. But you won’t know if Vera is
the kind of person who goes to parties to talk to other people, or if she
mostly goes there to dance. Or maybe both.
Scientific personality models sacrifice a high level of granularity for
a high level of consistency and comparability. You won’t be able to
understand all the nuances of who Vera is. But you will be able to
directly compare her character to that of others.
The most popular scientific model of personality is the Big Five.3
You might also know it as OCEAN model named after the five
personality traits it measures: openness to experience,
conscientiousness, extroversion, agreeableness, and neuroticism. I
will give you the chance to take a short personality test in a minute. It
will allow you to learn about your own personality profile before we
embark on our master snooping tour.
But let me first add a little bit more color to the five traits. You can
also see a summary in table 1-1.

TABLE 1-1

The Big Five in a nutshell

Personality trait Low High

Openness Practical; down-to-earth; Imaginative; curious;

traditional; conservative; original/creative; appreciation for art,
preference for the familiar beauty, and aesthetics; open-minded
Conscientiousness Flexible, carefree, Organized; dependable; goal-
disorganized; unreliable; oriented; detail-oriented; self-
spontaneous disciplined
Extroversion Reserved; quiet; Sociable; talkative; energetic;
introspective; deliberate; enthusiastic; assertive
solitary
Agreeableness Critical; blunt; skeptical; Compassionate; cooperative;
competitive; independent- trusting; altruistic; diplomatic
minded
Personality trait Low High

Neuroticism Calm; emotionally stable; Anxious; emotional; sensitive;

resilient; confident; grounded nervous; easily stressed

Source: Adapted from Gerald Matthews, Ian J. Deary, and Martha C. Whiteman, Personality Traits (Cambridge, UK:
Cambridge University Press, 2003).

Openness to experience: The Picasso trait

Openness to experience (or openness) refers to the extent to which
people prefer novelty over convention. People scoring high on
openness are intellectually curious, sensitive to beauty, individualistic,
imaginative, and unconventional. You might find them engaged in
philosophical discussions, traveling the world, exploring new
restaurants, visiting a museum, writing poetry, or painting.
People scoring low on openness, on the other hand, are down-to-
earth and more conservative in the values they hold (including
politics). They might not get excited by the idea of traveling to new
and unknown places, but instead prefer to return to their all-time
favorite all-inclusive hotel on the Riviera.
The Spanish painter, sculptor, printmaker, ceramicist, stage
designer, poet, and playwright Pablo Picasso is an excellent example
of an open-minded personality. Regarded as one of the most talented
artists of his time, and one of the most inspiring and influential
figures of the twentieth century, Picasso experimented with a wide
variety of artistic styles over the course of his career and gave birth to
more novel forms of artistic expression than any other artist at the
time (e.g., the collage or the Cubist movement). The quote “Others
have seen what is and asked why. I have seen what could be and asked
why not” perfectly captures Picasso’s mix of intellectual curiosity,
preference for novelty, and artistic interest, which also characterizes
the personality trait of openness.
 Conscientiousness: The Angela Merkel trait
Conscientiousness refers to the extent to which people prefer an
organized or a flexible approach in life. It captures how we control,
regulate, and direct our impulses.
People scoring high on conscientiousness are organized, reliable,
perfectionists, and efficient. They tend to be good at following rules,
resisting temptation, and sticking to schedules. And they love order.
Everything needs to be in the right place. Everything needs to be
perfect.
In contrast, people scoring low on conscientiousness are more
spontaneous, impulsive, careless, absent-minded, or disorganized.
They don’t care as much about achievements and instead take a much
more relaxed and spontaneous approach to life. They might wait until
the last minute to study for an exam or plan their holiday on the way
to the airport. And yes, they are the ones who regularly forget about
their friends’ birthdays or their own wedding anniversary.
I’ve heard some people call conscientiousness the German trait. I
assume that’s because it leaves you with the image of a meticulously
organized and perfectionist person. The sort of person that organizes
their socks according to color and stacks up books in alphabetical
order. It’s why I call it the Angela Merkel trait—always perfectly
prepared and dependable (I should say that, sadly, this characteristic
doesn’t apply to all Germans).

Extroversion: The Lady Gaga trait

Extroversion refers to the extent to which people enjoy company and
seek excitement and stimulation. It is marked by a pronounced
engagement with the external world, versus being comfortable with
one’s own company.
People scoring high on extroversion can be described as energetic,
active, talkative, sociable, outgoing, and enthusiastic. They love
people. Actually, more like LOVE people. You’ll most likely find them
at social gatherings, trying to be the center of attention and
entertaining the crowd. They are charming and usually full of energy
and positive emotions (as they will gladly tell you).
Contrary to that, people scoring low on extroversion are more
reserved, quiet, or withdrawn. They value their me-time and are
much more introspective than their extroverted counterparts. Why
waste your time and energy on other people when you can lose
yourself in thoughts and daydreams?
The icon that best captures the essence of extroversion for me is
the singer Lady Gaga (at least the public persona she portrays; I’ve
sadly never met her). The eccentric singer is extremely outgoing and
energetic. Her outfits are legendary. They are designed to attract as
much attention as possible.

Agreeableness: The Mother Teresa trait

Agreeableness reflects people’s need for cooperation and social
harmony. It provides insights into the ways in which we express our
opinions and manage relationships.
People scoring high on agreeableness are generally trusting, soft-
hearted, generous, and sympathetic. Because they are all about social
harmony, they avoid confrontation whenever possible, try their best to
not offend or insult, and are prepared to make personal sacrifices in
the service of others (e.g., through donations or volunteering).
In contrast, people scoring low on agreeableness are more
competitive, stubborn, self-confident, or aggressive. They have no
problem speaking up when they don’t like something or when they
believe something needs to be changed.
One of the agreeableness icons that has caught the public
imagination is Mother Teresa, the selfless, generous, and caring nun
who founded a religious congregation to help the most vulnerable
members of society. A symbol for altruism and kindness. Her charity
provided homes to patients dying of HIV/AIDS, leprosy, and
tuberculosis, and today sponsors soup kitchens and mobile health
clinics, and runs schools and orphanages.
 Neuroticism: The Piglet trait
Finally, neuroticism (also known inversely as emotional stability)
refers to the extent to which people experience negative emotions. It
reflects the ease with which we cope with and respond to life’s
demands.
People scoring high on neuroticism are anxious, nervous, and
moody. They tend to get irritated by seemingly small challenges and
worry a lot. Am I going to get sick? Am I going to get fired? Is it safe
to use the subway?
On the flip side, people scoring low on neuroticism are more
emotionally stable, optimistic, and self-confident. They are generally
easygoing and don’t get stressed quickly. Missed the subway? Got a
cryptic email from the boss? Having family over? Emotionally stable
people keep calm and carry on.
The neuroticism icon is one of my favorite fictional characters:
Piglet. Piglet is a young pig and one of Winnie the Pooh’s best friends.
In the Disney cartoon, he stutters, is constantly nervous, and fears the
wind and darkness. When he gets scared, his ears start to twitch.
Most of the time, Piglet thinks about all the possible ways in which
situations could go wrong. His mind then races with negative
thoughts that jump from one worst-case scenario to the next.

What defines who we are is our particular combination of these

personality traits. It’s what we call a personality profile.
Think back to Lady Gaga. I introduced her as the icon for
extroversion, but she also scores high on openness. And these two
characteristics aren’t independent; they influence each other. It’s
unlikely, for example, that Lady Gaga would express her openness
through shrill and unconventional outfits if she wasn’t also
extroverted and interested in attracting attention.
Imagine someone who is open-minded but rather introverted. Got
someone? The friend that immediately comes to mind for me is a
woman I went to grad school with. She was highly open-minded but
also extremely introverted. As you can imagine, flashy clothes weren’t
exactly her thing. Instead, she loved going to the museum and
devoured classic literature.

What’s Your Personality?

Let’s turn to you now. If you haven’t taken a Big Five test before, I
recommend investing the next few minutes doing so. It will make the
rest of the book much more relevant and engaging. You can visit this
book’s official website, www.mindmasters.ai/mypersonality, or do a
simpler paper-and-pencil version in appendix A at the end of the
book.
As you respond to the questions and interpret your results, I want
you to keep one thing in mind: there are no inherently good or bad
traits. Scoring high or low on each of the five personality dimensions
has its own unique advantages and disadvantages.
For example, you might be tempted to consider high agreeableness
—the tendency to be trusting and caring—a good trait to have. Being
friendly and trusting certainly has its advantages in some aspects of
life (e.g., relationships and teamwork), but extreme levels of
agreeableness can also be thought of as overly gullible, opportunistic,
and lacking a necessary level of assertiveness.
Although being somewhat disagreeable (i.e., critical and
competitive) might not make you a lot of friends, it is important when
you have to make difficult decisions or take the lead in a competitive
environment. The same is true for neuroticism. Few people want to be
seen as anxious and vulnerable. However, while being highly neurotic
certainly poses challenges for people’s health, it is also often
associated with great innovative potential and genius—especially
when paired with high levels of openness. Just think of the founder of
Apple, Steve Jobs. Known to be highly neurotic, Jobs’s slightly
eccentric nature and ability to be in touch with his own emotions is
what made him one of the most successful figures of the twenty-first
century.
Anyway, give it a go.
Now that you know how machines learn and what your own
personality profile looks like, we are ready to dive into research on
how computers can predict such profiles without you ever having to
touch a questionnaire. I’ve already told you that computers are better
than colleagues, friends, and family members when it comes to
predicting your personality from Facebook likes. How do they do
that? And what other pieces of the personality puzzle do our social
media profiles hold?
OceanofPDF.com
 2
The Identities We Craft Online

On December 17, 2020, thirteen-year-old P. Surya climbed on top of a

train coach in New Delhi, India, to take a daring selfie. He touched a
live wire and was burned alive. Surya’s death is no isolated incident.
Since 2011, more than 259 people around the world have died from
selfie-related accidents. That’s more deaths than caused by shark
attacks. In India, the problem is so widespread that popular tourist
destinations like Mumbai have instated selfie-free zones throughout
the city. Similarly, the Russian government became so alarmed by the
growing number of selfie-related deaths that it started to issue flyers
educating the public on how to take safe selfies.
Why do people risk their lives hunting for the perfect Instagram
picture or spend hours crafting the perfect TikTok video? Why are we
so obsessed with sharing our lives on social media? The answer is
simple: We are fighting for attention. We want to be seen.
As science has shown, self-disclosure is inherently rewarding.
Sharing one’s own opinion or attitudes with others causes a spike of
activity in the brain’s pleasure center.1 That’s the part of the brain that
typically springs into action when we receive rewards such as food,
money, sex, or heroin. Yes, you’ve read that correctly. Sharing
information about yourself triggers a similar brain response to
receiving money or having sex. In fact, it feels so good to talk about
ourselves that we’re willing to give up money to share our inner lives
with others.
 But that’s only half the story. Social media platforms are designed to
encourage self-disclosure by making sharing easy and socially
rewarding. How often do you check your likes, shares, or retweets?
Every thirty minutes? Every ten? Every five? As social beings, we crave
positive feedback from our environment and those around us. This
feedback functionality is what makes social media sites so addictive.
It’s also what makes social media sites the ideal hunting ground for
what psychologists refer to as identity claims—deliberate expressions
of a person’s identity. Just as I communicated my identity to the other
people in my village by putting a provoking bumper sticker on my car,
dying my hair, or wearing a Ramones sweatshirt, social media
platforms encourage their users to tell their stories and share their
identities with others. You might follow Beyoncé’s official fan page,
post about your vacation to the Seychelles, or share a picture of the
delicious burger you’ve had for lunch. A staggering 80 percent of the
status updates posted on social media sites such as Facebook or
X/Twitter are directly focused on a person’s immediate experience.2
We naturally assume that these traces hold information about their
owners. Why otherwise would recruiters consult applicants’ social
media pages before hiring? Why would the US government encourage
visa applicants to link their social media accounts? And why would
most of us stalk our dates online before meeting them?
Research confirms our intuition. By studying the social media
profiles of strangers, we can gain valid insights into their psychology.3
But our judgments aren’t nearly as accurate as those made by
computers.

The ABCs of Algorithmic Snooping

The intuition behind psychological targeting is simple. Imagine you
are part of a profiling unit at the FBI. You’ve been given the following
profile: Target X, an anonymous user who likes Hello Kitty, posts
about their passion for anime and manga characters, and shares videos
of the famous Korean band BTS.
 Now describe the user to me. Make an educated guess of who they
are. Their age, gender, ethnicity, and personality. Here’s my best guess:
Teenage girl. Asian. Introverted and open-minded, possibly neurotic. I
assume your guess might have looked similar (there is typically a high
level of consensus).
How did you arrive at your conclusion? You most likely drew on
your experience. Your recollection and impression of people with
similar interests to the mysterious Target X. This could be people you
are close to. People you know tangentially through friends. Or people
you’ve only seen on TV. When we judge strangers based on our
observations of their behavior, we tend to think of the prototypical
person of the same behavioral profile—the mode (i.e., most common
option) or average, if you want.
In the absence of any other information, taking the average person
as your comparison point is your best bet (and the computer’s, too).
You will, of course, make mistakes. It turns out that the Guinness
World Record for the largest collection of Hello Kitty memorabilia in
2017 was held by sixty-seven-year-old Masao Gunji, a retired police
officer from Japan.4
The predictions we make—and the relationships these predictions
are based on—are probabilistic, not deterministic. They help us make
educated guesses about the world with incomplete information. It’s far
more likely that the Target X profile belongs to a young teenage girl
than a sixty-seven-year-old retired police officer. But we should never
use our intuitions to jump to definitive conclusions about any one
individual.
The same is true for the predictions computers make. Their
predictions might be more systematic than ours, but they always
reflect educated guesses, never the truth. An algorithm’s judgment of a
person’s characteristics are overgeneralizations and can therefore seem
rather stereotypical.
Look at the word clouds in figure 2-1, which were published in a
paper by Andrew Schwartz and colleagues in 2013. They show the
words in people’s Facebook status updates that are most indicative of
being male or female (top = female, bottom = male), broken down into
a general word cloud in the middle as well as clustered topics on the
outer rim. (You can classify people’s gender with an accuracy of over
90 percent just by looking at what they post on social media.) The
bigger the word, the more strongly it is correlated with being male or
female; and the darker the word, the more often it appears in people’s
statuses. (You can scan the QR code on your phone or tablet to see the
word clouds in color.)
The word clouds are just as stereotypical as my Target X example
but based on actual data. Women post about shopping, babies, and
boyfriends, while men swear and chat about sports and video games.
Perhaps women really are from Venus and men from Mars after all.

FIGURE 2-1
Words in users’ Facebook status updates indicating gender
Scan the QR code for a color version of the word clouds.
Source: H. Andrew Schwartz et al., “Personality, Gender, and Age in the Language of Social Media: The Open-
Vocabulary Approach,” PloS One 8, no. 9 (2013): e73791, https://doi.org/10.1371/journal.pone.0073791. Permission
via https://creativecommons.org/licenses/by/4.0/.

As we dive into the science of psychological targeting together, it is

important to remember that the relationships I’m going to show you
are free of any normative judgment. They describe the world as it is (or
was, as we typically work with historical data), not how it ought to be
—a critical distinction between description and prescription that I will
highlight repeatedly throughout the first part of the book.
It is also important to remember that the insights aren’t limited to
specific platforms. Much of my own research—and that of my
colleagues—has relied on Facebook data. But you can make similar
predictions with data from all the other social platforms like TikTok,
Instagram, Reddit, Snapchat, and more. Whenever there’s a way for
people to express themselves online, there’s a way to gather identity
claims and turn them into psychological profiles.

The World of Facebook Likes

The Facebook Like (aka: thumbs up) is an iconic feature of the
platform. It’s an easy way for users to express what they care about and
to appreciate the content that other users generate and share. We’ve all
been there. Friend 1 posts a cute cat video and we like it. Then Friend 2
posts a snarky comment making fun of Friend 1 for posting cat videos
—and we like that, too. A fun pastime, for sure.
But these aren’t the kind of likes I am going to discuss in this
chapter. I focus on the official Facebook pages users can like or follow.
The official Facebook page of soccer player Lionel Messi has around
101 million followers, for example. Taylor Swift has 77 million, and
Barack Obama, 56 million. But it’s not just celebrities and other big
shots that have Facebook pages. Anyone can create one. In 2019, there
were over 60 million active Facebook pages.
Why Facebook pages? In 2013, together with two of his colleagues,
Michal Kosinski showed that Facebook likes were predictive of a whole
range of sociodemographic and psychological characteristics: people’s
age, gender, drug use, political ideology, sexual orientation, IQ, life
satisfaction, and personality.5 In addition to reporting the accuracy of
their predictive models, their paper also included a host of examples of
Facebook pages that are indicative of different personal traits.
Look at the two lists of Facebook likes in table 2-1. They are both
related to the personality trait of extroversion. One of them reflects
people who are reserved and quiet (introverted), and the other reflects
people who are outgoing and social (extroverted). I want you to guess
which list is which.
Got your answer? Let’s see. Extroverts are on the left, introverts on
the right. If you happen to like beer pong (a popular party game in
which players try to land a Ping-Pong ball in a cup of beer across the
table), cheerleading, or Michael Jordan (list A), you are more likely to
be extroverted. If, instead, you are more interested in anime,
Minecraft, or Terry Pratchett (list B), you are more likely to be
introverted.
Makes a lot of sense, no? Extroverts are energetic, talkative, and
sociable, making it hardly surprising that the majority of extroverted
likes are related to social activities (e.g., beer pong, cheerleading,
dancing, socializing) and have a certain degree of attention-seeking
potential (e.g., cheerleading, modeling, dancing).
TABLE 2-1

Facebook likes related to the personality trait of extroversion

List A List B

Beer pong RPGs

Michael Jordan Fanfiction.net
Dancing Programming
Socializing Anime
Chris Tucker Manga
I feel better tan Video games
Modeling Role-playing games
Cheerleading Minecraft
Theater Voltaire
Flip cup Terry Pratchett

Source: Adapted from Michal Kosinski, David Stillwell, and Thore Graepel, “Private Traits and Attributes Are
Predictable from Digital Records of Human Behavior,” Proceedings of the National Academy of Sciences 110, no. 15
(2013): 5802–5805.

The same degree of intuitiveness applies to the list of introverted

likes. Knowing that introverts are typically shy, reserved, and quiet, it
makes sense that they would favor activities that don’t require other
people (e.g., programming, reading anime or Terry Pratchett novels,
and playing video games).
Let’s look at another example. This time I will let you try to guess
which personality trait is associated with the two lists in table 2-2.
Remember, we still have openness, conscientiousness, agreeableness,
and neuroticism in the race (feel free to skip back to the summary of
all the personality traits in table 1-1 if you want).
What do circles of prayer, Redeeming Love, and Compassion
International have in common? And how do they differ from I hate
everyone, you, and police, and Julius Caesar? The first list leaves you
with this warm glow, while the second list generates a considerable
amount of discomfort. Settled on a personality trait? It’s agreeableness
(the lists for the three remaining personality traits are in appendix B).
TABLE 2-2

List of Facebook likes related to one of the Big Five personality traits

Continue reading to see which one.

List A List B

Compassion International I hate everyone

Logan Utah I hate you
Jon Foreman I hate police
Redeeming Love Friedrich Nietzsche
Pornography harms Timmy from South Park
The Book of Mormon Atheism/Satanism
Circles of prayer Prada
Go to church Sun Tzu
Christianity Julius Caesar
Marianne Williamson Knives

Source: Adapted from Michal Kosinski, David Stillwell, and Thore Graepel, “Private Traits and Attributes Are
Predictable from Digital Records of Human Behavior,” Proceedings of the National Academy of Sciences 110, no. 15
(2013): 5802–5805.

List A speaks to people who are caring, trusting, and empathetic—

the nice guys who are embedded in their church community and give
to charity. List B is the opposite. A bunch of critical and competitive
people who hate pretty much everything, are into knives and Satanism,
and follow role models such as Julius Caesar, Sun Tzu, and Timmy
from South Park. And it’s finally confirmed by science: the devil does
wear Prada!
Here’s a bonus round. Table 2-3 shows a psychological characteristic
that you all know (I promise), but not it’s not a personality trait. Take a
look at the two lists and see if you can figure it out.

TABLE 2-3

List of Facebook likes related to a psychological characteristic

Continue reading to see which one.

Continue reading to see which one.

List A List B

The Godfather Jason Aldean

Mozart Tyler Perry
Thunderstorms Sephora
The Colbert Report CHiQ
Morgan Freeman’s voice Bret Michaels
The Daily Show Clark Griswold
Lord of the Rings Bebe
To Kill a Mockingbird I love being a mom
Science Harley Davidson
Curly fries Lady Antebellum

Source: Adapted from Michal Kosinski, David Stillwell, and Thore Graepel, “Private Traits and Attributes Are
Predictable from Digital Records of Human Behavior,” Proceedings of the National Academy of Sciences 110, no. 15
(2013): 5802–5805.

I’m going to give you a hint. You will probably find it easier to guess
the trait if you focus on list A. The likes I would consider most
intuitive are science, To Kill a Mockingbird, and Mozart.
The answer is intelligence (or what we today call cognitive ability).
List A reflects high intelligence; list B, low intelligence. As you can see,
list A relates to reading, sophisticated humor, and science. In contrast,
the items on list B are a lot more down-to-earth and tangible.
The only question mark is curly fries. Why on earth would liking
curly fries—as delicious as they are—be related to high intelligence?
There is no obvious reason. Yet, they were in the data Kosinski and his
colleagues studied. It’s possible that there was a group of highly
intelligent college students who all decided to like curly fries. Or
perhaps there is logic to it, and I am just not intelligent enough to
figure it out. It could be all the above, or none. We simply don’t know.
The reason I bring up this example is because it illustrates an
important lesson regarding the relationships between our digital
footprints and psychological characteristics. The relationships we
observe are not causal. Eating curly fries will not make you more
intelligent, and it’s equally unlikely that being more intelligent makes
you genetically predisposed to liking curly fries. But just because we
cannot explain the relationship between the two doesn’t mean we
cannot use it. Void of any understanding, the relationship can still help
us make predictions.
If my goal is to predict whether a person is intelligent, I might not
care about the deeper meaning of the cues. It doesn’t really matter if I
understand why liking curly fries is related to higher intelligence. All
that matters is that it is. If I trust the data, I should assume that the
next person I meet who likes curly fries on Facebook is also relatively
intelligent.
To be clear, the relationship between liking curly fries and
intelligence might change over time (and the fact that there’s no
obvious reason for it to exist in the first place makes this much more
likely). But at least in the moment you observe a particular
relationship, this relationship will help you predict your outcome
regardless of whether you understand its meaning or not.

The World of Words

Take a look at the two X/Twitter posts and try to conjure an image of
the two authors. What do they look like? How old are they? Are they
more likely to be female or male? Outgoing or shy? Liberal or
conservative?

GOODBYE COVID HELLO DANCING 💕 happy for Australia!

Praying for the rest of the world that we can all be dancing
together soon 🙏
On #MLKDay , we celebrate his life but we’re also called to
live out his values through service of our own. Here are some
ways you can get involved in your community:
The first tweet comes from Lady Gaga; the second from Barack
Obama. You might not have thought of Lady Gaga and Barack Obama
when reading those tweets, but I bet that the images you created in
your mind were probably not too far off.
The psychologist James Pennebaker at the University of Texas at
Austin was among the first to advocate for the power of language as a
window into our psyche. Pennebaker studied how writing about our
traumatic experiences could help us heal.6
While reading through some of the essays his patients had written,
he couldn’t help but notice how different the essays were, in both
content and style. While some people recited their trauma almost in a
matter-of-fact tone without revealing any emotional identification
with the story, others seemed to be consumed by anger and despair.
Yet others reflected on what had happened with an optimistic outlook
on the future.
Could these differences predict how well an individual might cope
with their trauma? Could they be used to learn something about the
inner mental states of these individuals?
In his quest to answer these questions, Pennebaker created an
entire field of scientific inquiry based on the idea that one can
objectively quantify aspects of language and relate them to our
psychological experiences (I will return to Pennebaker and his findings
later, so stay tuned).
What makes language such a fascinating puzzle piece of who we are
is that it is ubiquitous. We spend about 50–80 percent of our waking
hours engaged in some kind of communication. This could mean
greeting a stranger in the subway, engaging in a deep conversation
with one of our close friends, writing an email to a colleague, leaving a
voice message for our children, or noting down our thoughts in a diary.
But it could also mean posting on our social media platform of
(experimental) choice: Facebook.
The word clouds in figure 2-2 were generated based on the
Facebook status updates of almost seventy-five thousand users. If
you’re old enough, young enough, or simply self-disciplined enough to
not have a Facebook account, let me bring you up to speed on what
Facebook status updates are (and convey my admiration; well done!).
Facebook status updates give users the chance to share what’s on their
mind through text, images, or videos. You can decide to share the
content within your circle of friends only or distribute it publicly for
the entire world to see.
All you need to create these word clouds—and unlock the
psychological secrets of language—is simply count how often different
words appear in the status update of a particular user. How often does
the person talk about parties, the weekend, or computers? Once you
have done this for all the seventy-five thousand users in your dataset,
you can start connecting the dots by calculating basic correlations
between relative word frequencies and personality traits. It’s that easy.
Now take a closer look at the two word clouds (as before, the bigger
the word, the more strongly it is associated with the respective
personality trait; the darker the word, the more common it is). Any
ideas which personality trait they might represent? It’s extroversion,
again (A = extroversion, B = introversion). You can literally see the
introverted geek sitting in front of their computer, browsing the
internet, just having finished reading another anime comic. And you
can conjure an image of the outgoing extrovert who is sooooo excited
about the amazing night out planned for next weekend. Gotta love the
girlz! I might have been able to predict that extroverts talk about
parties. But I would have never thought to put the words “sooooooo” or
“gotta” on my list of indicators. That’s an insight that only the data
itself could offer me.

FIGURE 2-2
The words in people’s Facebook status updates most strongly correlated with being
extroverted and introverted
Scan the QR code for a color version of the word clouds.
Source: H. Andrew Schwartz et al., “Personality, Gender, and Age in the Language of Social Media: The Open-
Vocabulary Approach,” PloS One 8, no. 9 (2013): e73791, https://doi.org/10.1371/journal.pone.0073791. Permission
via https://creativecommons.org/licenses/by/4.0/.

Figure 2-3 is my personal favorite (the remaining three personality

traits are in appendix B). The figure shows only the first half. It’s high
agreeableness. The word cloud has this warm glow to it. It is full of
praise, blessed thank-you notes, merry Thanksgiving and Christmas
celebrations, and an overwhelming feeling that the world is just a
wonderful place. Before you look at the word cloud for low
agreeableness, guess what it might look like. Which words could be
indicative of having a rather critical, competitive, and quarrelsome
personality? Got a mental image in your mind? A word of warning: it’s
not going to be pretty. You can see it in figure 2-4.

FIGURE 2-3
The words in people’s Facebook status updates most strongly correlated with
agreeableness

Scan the QR code for a color version of the word clouds.

Source: H. Andrew Schwartz et al., “Personality, Gender, and Age in the Language of Social Media: The Open-
Vocabulary Approach,” PloS One 8, no. 9 (2013): e73791, https://doi.org/10.1371/journal.pone.0073791. Permission
via https://creativecommons.org/licenses/by/4.0/.

No additional commentary needed.

 But personality isn’t the only psychological characteristic we can
infer from language. I already mentioned the work by psychologist
James Pennebaker, who started studying the links between language
and mental health in the 1980s and 1990s. About three decades—and
many methodological advances—later, we have ample evidence that
Pennebaker’s intuition was spot on: our language offers a glimpse into
our psychological well-being and mental health.

FIGURE 2-4

The words in people’s Facebook status updates most strongly correlated with low
agreeableness
Scan the QR code for a color version of the word cloud.
Source: H. Andrew Schwartz et al., “Personality, Gender, and Age in the Language of Social Media: The Open-
Vocabulary Approach,” PloS One 8, no. 9 (2013): e73791, https://doi.org/10.1371/journal.pone.0073791. Permission
via https://creativecommons.org/licenses/by/4.0/.

Take the use of first-person pronouns such as “I,” “me,” or “myself,”

for example. Any thoughts on what using these words might say about
you? My first guess when Pennebaker asked this question at an
academic conference was that it had to be narcissism. “Why would
anyone care about anything other than me? Come on folks, pay
attention to who’s really important here!” I was wrong. References to
yourself don’t make you a narcissist. Instead, they are indicators of
emotional distress.
Surprised? So was I. But take a moment and think back to the last
time you felt really down. What were you thinking about? The future
of humanity? Unlikely! When we feel down, we typically think of
ourselves. Why am I feeling so bad? Am I ever going to get better?
Why can’t I deal with this situation more successfully? When things
look dreary for us, we tend to look inward, focus on ourselves, and
ruminate.
And because we cannot constantly monitor our thoughts and
feelings (especially when we are feeling down), this inner monologue
creeps into the language we use when expressing ourselves to others.
We might not openly admit to feeling blue, but we can’t help talking
more about ourselves than usual. A study by Allison Tackman and
colleagues from the University of Arizona suggests that individuals
suffering from depression use about 40 percent more first-person
pronouns than their healthy counterparts; that’s six hundred a day.7
But it’s not just the extent to which we talk about ourselves that
gives away how we feel. One of my favorite studies in this space was
conducted by the psychologist Johannes Eichstaedt and colleagues,
who examined the Facebook status updates and medical records of
683 patients.8 Just by looking at the words people used to describe
their experiences on Facebook, Eichstaedt could accurately predict
whether a person was suffering from depression in 72 percent of the
cases (with 50 percent being chance, or a coin flip).
Seventy-two percent might seem far from perfect. It is. Ideally you
want that number to be as close to 100 percent as possible. However, it
turns out that 72 percent is about as good as the accuracy of short
screening surveys that are commonly used in a mental health
diagnostic. And suddenly a 72 percent accuracy achieved by a
computer model snooping through your Facebook statuses becomes
rather remarkable.
Let’s explore some of the relationships underlying these predictions.
The word clouds in figure 2-5 show the words (organized by topics)
that are most indicative of being depressed. They relate to negative
mood and affect (tears, crying, feeling sick), interpersonal challenges
including loneliness and hostility (miss, irked, upset, nerves, hate), and
somatic complaints with medical references (headache, sick, hurt,
pain, hospital).

FIGURE 2-5
The words in people’s Facebook status updates most strongly correlated with clinical
depression

Scan the QR code for a color version of the word cloud.

Source: Johannes C. Eichstaedt, et al., “Facebook Language Predicts Depression in Medical Records,” Proceedings
of the National Academy of Sciences 115, no. 44 (2018): 11203–11208.
 As I said before, we cannot make any claims about cause and effect.
Does being lonely cause people to be depressed or is loneliness a by-
product of depression? Does experiencing somatic health problems
result in mental health issues, or is poor mental health contributing to
physical symptoms? Most likely, the links cut both ways. But this
doesn’t matter if your goal is to simply identify people at risk (and
ideally offer them support).
What I find particularly fascinating about these word clouds is the
strong connection between physical health (i.e., somatic complaints)
and mental health. We often think of the body and the mind in
isolation. Yet, Eichstaedt’s research suggests that both are intricately
linked. Physical pain can easily turn into a mental one, and vice versa.
As these findings highlight, linking digital footprints to psychological
traits does not only open a window into people’s psychology but can
also teach us invaluable insights into our collective psychology.
Sometimes these insights are entertaining. Sometimes they are
reassuring. And sometimes they are rather disturbing. Let me show
you what I mean by this with a final example. It’s not a psychological
trait in the classic sense. Yet, it’s a personal characteristic that most of
us consider private: our income or socioeconomic standing.
Less than 10 percent of users on online dating platforms provide
information about their income, and most people do not discuss their
salaries even with their closest friends or family members. Although
the National Labor Relations Act of 1935 granted employees in the
United States the right to disclose their salaries, the cultural norm still
considers it inappropriate. A commentator in The Atlantic skillfully
captured this sentiment in his comparison: “Asking a coworker about
pay seems akin to asking about their sex life.”9
As some of my own research has shown, a relatively simple model
trained on people’s Facebook status updates can guess an individual’s
income with a margin of error of about $10,000.10 Not perfect, but not
bad either. However, what really struck me were some of the
relationships we observed in the data. Look at the word clouds in
figure 2-6. The differences in what the wealthy and the poor talk about
are remarkable and paint a disturbing—yet not necessarily unexpected
—picture.
 High-income individuals talk about vacations (e.g., vacation, flight,
beach, Vegas, airport) and pleasant activities that usually require
spending a considerable amount of money (e.g., shopping,
celebrating). They express positive emotions (e.g., excited to, great)
and use future-oriented words and phrases (e.g., looking forward to,
afterwards). Low-income individuals, on the other hand, are more
self-focused (e.g., I need, I can, I got, me), express themselves through
more colloquial language (e.g., idk, cuz), share predominantly negative
feelings (e.g., hurt, hate, bored), and use more swear words and
emoticons.

FIGURE 2-6
The words in people’s Facebook status updates most strongly correlated with level of
income
Scan the QR code for a color version of the word clouds.
Source: Sandra C. Matz, et al., “Predicting Individual-Level Income from Facebook Profiles,” PloS One 14, no. 3
(2019): e0214369, https://doi.org/10.1371/journal.pone.0214369.

If you find these word clouds troubling, you are not alone. They are.
But as much as I would like them to be different, they are not. That’s
what the world looks like when you are poor.
The ability to reflect uncomfortable truths back at us is part of what
makes big data powerful. It offers a window into the lives of others,
providing us with perspectives that we might otherwise not have
access to. In some instances, these insights are hilarious (think back to
the word clouds for agreeableness). In other instances, the results
might be shocking and condemning of the societies we live in. But
having access to this descriptive reality is an opportunity to point out
pressing social issues, and to garner support for change.
It is not surprising, for example, that low-income individuals are
more self-focused, while their high-income counterparts dream about
the future. It’s not that poor people are selfish. It’s that thinking about
the future is pretty damn hard when you are struggling to make ends
meet in the now. It’s a luxury to not have to think about yourself and
the precarious financial situation you might be in all the time.
 The World of Pixels
We all know the adage “A picture is worth a thousand words.” This
certainly feels true when it comes to seeing your niece sprout her first
tooth or take her first bike ride. You might get a sense of what’s
happening if your brother or sister describes this experience in an
email or text message. But it’s not the same as receiving a picture or,
even better, a video of the event. The experience becomes so much
more relatable and real.
But is this also true in the world of psychological targeting? Are the
pictures we post more predictive of who we are than the words we
speak? The short answer is no. Based on the latest science, pictures are
less predictive of psychological traits than Facebook likes or language.
However, that doesn’t mean they don’t offer any insights (and could
very well change in the future when we have more sophisticated ways
to analyze them). A computer trying to distinguish between an
extrovert and an introvert, for example, will be right in about 70
percent of cases (50 percent is chance). If you were to attempt the
same task, you’d be right only about 60 percent of the time.
Turning pictures into psychological profiles isn’t a trivial task.
Pictures are a nightmare to work with. They are unstructured and
complex. Facebook likes are straightforward to quantify. You either
like a page or you don’t. It’s a one or a zero. Easy enough. Words are a
tad more complicated but still easy to count. Pictures, on the other
hand, are a combination of millions of pixels that only make sense
when you put them together a certain way.
The simplest way of analyzing pictures is to break down their
complexity into a finite set of concrete features. For example, you
might start by identifying the objects that are present in your picture.
A chair, multiple people, a cat, glasses, a lamp. No need to do this
manually; object recognition algorithms will do the job for you. What
else? You can describe the image’s color. Lots of red, a bit of blue. High
saturation, mostly warm colors. From there, you could jump to the
image’s composition (if you are a photographer, you know what I’m
talking about). How are the elements of the image organized? Is it
broken down into many small distinct regions, or does it mostly
consist of a few larger ones? Are the elements symmetrical? This is just
to give you an idea. There’s almost no limit to how many features you
can extract.
Let’s have a look at an example from the work of the Italian
computer scientist Cristina Segalin, who has shown that the pictures
we post and like on social media are about as predictive of our
personality as our credit card spending (which I will get to in the next
chapter).11 The two collages you’ll see when scanning the following QR
code (sorry, that’s your only option as we need color for this one) show
Flickr pictures that people added to their list of favorites (Flickr used
to be a leading image and video hosting platform with social network
features; it’s still around, but far less popular than it used to be). If you
cannot access the pictures, let me describe them to you. One is a happy
mix of colorful images featuring flowers, sunsets, cultural sites, and
food. The other one is a rather gloomy assembly of mostly gray images
without a particular topical focus. Any guesses which personality trait
the two collages might reflect? It’s neuroticism.

You might be asking yourself: Is it really so surprising (or

interesting) that the pictures people like or post tell us something
about who they are? Isn’t that the whole point of liking and posting
them in the first place?! And the clues the algorithm uses to make
predictions of people’s personality don’t seem to be about the person
either. It’s the colors and basic content of the picture that drive the
judgments. Fair point. Up to now, that was the case. But it’s not the
only signal computers can use to snoop around your inner mental life.

I Can See It on Your Face

The part I have left out so far is the most contested among scientists
and usually makes people recoil (including myself ). But it’s also the
part that—if true—should give you pause and make you reconsider the
way you think about photographs. I’m talking about the ability to
predict people’s psychological characteristics from their faces.
Physiognomy—the art of judging someone’s character from facial
characteristics—isn’t new. It has a long and dark history that reaches
all the way back to ancient Greece. Pythagoras is said to have selected
his students based on their facial features. The captain of the Beagle
almost canceled Charles Darwin’s historic voyage because he thought
Darwin’s nose revealed a lack of determination and energy. Cesare
Lombroso (the founding father of criminal anthropology) believed that
criminals could be identified by features such as the softness of their
skin, a childlike appearance, or the thickness of their hair. And lead
scientists in the Nazi regime used pseudoscientific evidence to
perpetuate anti-Semitism.
Eventually, the validity of such physiognomic claims was universally
debunked. The idea that one could infer personal characteristics from
people’s physical appearance, especially their face, fell out of fashion.
Rightfully so. None of the original claims made by the early advocates
of physiognomy could stand the test of solid scientific inquiry, and
many of the use cases were horrific.
However, recent advances in computer vision have rekindled the
scientific interest in the relationships between our physical features
and our personality and character. That brings me back to an old
friend of ours, Michal Kosinski (yes, he likes controversial topics),
whose research suggests that computers can accurately predict your
personality, sexual orientation, and even political ideology from your
face.
 Take one of the most controversial findings, for example: the ability
of computers to predict whether you identify as straight or gay just by
observing your face.12 With only one picture, the computer’s accuracy
for male targets is 81 percent. With five pictures, that accuracy shoots
up to 91 percent. For women, the accuracies are slightly lower, with 71
percent for one picture and 83 percent for five.
If what I’ve just said makes you uneasy and question whether I
might be delusional, I get it. The thought that my personality, sexual
orientation, or political ideology could be predicted just by looking at
my face is spooky to say the least. It’s also hard to believe that it could
be true. When I first heard about the research, I thought it was
ridiculous. I wasn’t alone in this sentiment. The work of Kosinski and
others has received substantial pushback, both from the public and
from other researchers.
But bear with me for a minute and let me channel Kosinski’s
arguments for why we might, in principle, expect our faces to be
reflective of our inner mental lives. Let’s start with a source that is
utterly unscientific: the lyrics of the 1972 song “The Story of Your Life
Is in Your Face” by the American singer-songwriter Tom Hall:

He said the story of your life is in your face

It’s written there in little subtle lines
The story of your life is in your face
What’s written on your face has been heavy on your mind

The song is a beautiful homage to how our faces act as a canvas for
our emotions. Smile lines, for example, tell the tales of a life filled with
happiness and laughter—both of which are known to be the hallmark
of extroversion. As our psychological experiences accumulate over the
years, they might alter our physical appearance, including our facial
features.
Similarly, it’s likely that our facial features exercise at least some
influence over our character. Say you are born a beautiful baby.
Symmetric face, big eyes, rosy cheeks. Like it or not, Mother Nature
has set you up for success. As a toddler, adults might stop more often
to say hi and smile at you. As a teenager, you might be more popular.
Would it really be so surprising that with all these positive social
experiences, you might turn out a little bit more extroverted? This isn’t
just a hypothetical example but an actual scientific finding. Attractive
individuals receive more positive social feedback from their
environment and as a result become more extroverted.13
And finally, there is a whole list of factors that could influence both
facial features and psychological characteristics. Think of your
upbringing, environmental factors, or simply variations in hormone
levels.
Take testosterone, for example, which naturally occurs in all of us.
What do you associate with this hormone? Masculinity? Aggression?
Risk-taking? Well, all of them are true. Testosterone levels influence
your physical appearance by making you look more masculine (e.g.,
impacting the width-to-height ratio of your face). At the same time,
giving people testosterone can alter their behavior and personality
such that they become more aggressive and risk-seeking.
Let’s say the history of physiognomics didn’t exist, and I had just
told you about all the possible pathways by which facial features and
personal characteristics might be related. Don’t you think there is at
least some reason to consider it a possibility?
To be clear, this isn’t necessarily great news. If we can predict a
person’s traits from their face, that could have terrible consequences
(as Kosinski himself warns in his work). But how we feel about the
implications of such research should be independent of whether we
believe the research itself.

Into the Deep

With that in mind, let’s start exploring how algorithms can use your
face as a window into what lies beyond. As we’ve seen before,
computers can extract a list of features related to a picture’s content,
color, composition, and so on. Faces are different. Two eyes, one
mouth, and a nose. Not exactly master snooping material. We need
more information. Real magic. The closest we can get to this is a
computational approach called deep learning or deep neural networks.
 Deep learning is an attempt at getting computers to mimic how the
human brain works. As humans, we look at an object or scene and
immediately know what’s going on. We process the millions of light
particles that hit our retina every millisecond and weave them into a
coherent image. What’s going on behind the scenes are billions of
neurons firing at different levels of cognitive abstraction.
Deep learning is based on the same idea. Just like the human brain,
deep neural networks consist of several layers of neurons that process
information and decide whether to pass on this information to the
next layer. Instead of working with a predetermined set of features,
neural networks start by considering every single pixel in an image
(lowest layer). Over many (many!) trial-and-error runs, the networks
learn how certain constellations of pixels are related to more abstract
concepts (e.g., does this picture contain a cat or is this face
extroverted?).
You can think of it as an organization. There are different levels of
hierarchy, with many workers at the bottom and a CEO at the top. At
each level of the organization, employees must process information
and decide what their superiors need to know. This goes all the way up
to the top. While the CEO never sees all the individual pieces of
information, their decisions are (ideally) informed by the collective
knowledge of the organization.
The computational models themselves are too complex to interpret.
But we don’t need to understand the model’s neural architecture to get
a sense of what’s going on under the hood. If a model can accurately
classify people into extroverts and introverts (based on our validation
against self-reported survey responses), then the model must be doing
something right. Which means that we can simply scrutinize the
predictions it makes.
Take a look at the two photos in figure 2-7. They are taken from
ongoing work by Michal Kosinski and Poruz Khambatta.
I know. They look like dreamy, soft-focus portraits. And they kind of
are. These are depictions of an introvert and extrovert, as dreamed up
by an algorithm. One photo is the combination (a morph) of the
female faces in the dataset that were predicted to be most extroverted,
and the other is a combination of the ones that were predicted to be
most introverted (and yes, you can do the same thing for men and
ethnicities other than Caucasian; you just can’t mix them).
I’m sure you have a sense which one is which. Introverts are on the
left; extroverts are on the right. How did you arrive at that conclusion?
The eyes? The hair? The shape of the face?

FIGURE 2-7

Morphs of the ten most introverted and extroverted faces as predicted by the
algorithm

Scan the QR code for a color version of the photos.

Source: Michal Kosinski, “The End of Privacy,” paper presented at the annual convention of the Society for Personality
and Social Psychology, San Diego, January 28–30, 2016.

Here’s what I see when I look at the pictures: the faces of extroverts
are slimmer, they smile, their eyes appear to be bigger and lighter
colored, they don’t have outlines of glasses, and they have lighter hair
(I recommend scanning the QR code to see the color version; the
differences are much more striking there).
And here’s what I intuit from my observations: Extroverted women
are vain; they dye their hair, refuse to wear glasses, and swap out the
boring transparent contact lenses for blue ones. What about their faces
being slimmer? Well, it could be that extroverts are more obsessed
with their weight. But it could also be that they are simply better at
taking pictures. They might have figured out that taking photos from
above makes your face look slimmer. I’m sure introverts figured that
out as well. But they might simply not care. Want more evidence for
this theory? Check out the nostrils. Clearly visible for introverts but
not at all visible for extroverts.
This raises an important question. How much of what the
algorithm picks up on are actual differences in facial features as
opposed to grooming habits and the ways in which people take
pictures? If all you care about is accurately predicting people’s
personality traits, this distinction might not bother you all that much.
If the signal is consistent, you might take all the grooming cues you
can get.
However, if you are a scientist claiming direct relationships between
facial features and personal characteristics, you need a better answer.
Kosinski and Khambatta were well aware of this.
So, their team invited students at Stanford University to their lab.
All participants were asked to come shaved, without makeup, and with
their hair pulled back with a hair band. All pictures were taken from
exactly the same angle, using the same background, and the same
camera. In other words, the pictures were held as constant and free of
confounding signals as possible. If the effects from the two online
samples were a mere artifact of grooming and photography skills,
Kosinski’s algorithm should fail spectacularly in this highly controlled
sample. It didn’t. It did just as well as in the original sample (or even
better).

Beyond Social Media

As a computational social scientist, I am fascinated by the ability of
computers to turn social media profiles into highly intimate
predictions of people’s psychology. It’s remarkable. Groundbreaking.
However, as a social media user myself, the idea is far less
appealing. It’s creepy. Intrusive. If you are one of the lucky few who
resisted the social media vortex or deleted your account for good, you
might be patting yourself on the shoulder right now. Well done, you
saved yourself from the evil grip of Big Brother, while the rest of us are
doomed. I wish that was true. But it’s not.
Just think back to the power of words when it comes to revealing a
person’s inner mental state. Sure, you can get those words from social
media. It’s an easy starting point. But we don’t have to rely on written
language posted on social media to make inferences about who you
are. Advances in speech-recognition technology have made it easier
than ever before to transcribe spoken words to written language. What
used to take hours of manual transcription can now be done in
seconds at the ease of a click. Instead of relying on your social media
posts, we can simply eavesdrop on your conversations. You probably
have a smartphone, don’t you? And maybe other smart devices such as
Alexa or a Samsung TV that come equipped with voice control.
The same is true for pictures. They have a property that makes them
particularly challenging when it comes to privacy. Any ideas? If
nothing comes to mind, you are probably thinking of your own
pictures. There’s nothing inherently wrong with sharing those on social
media. It can feel good to have others share the moments that matter
to us. But that’s only a small portion of the pictures of us circulating
out there.
Your friends might post pictures of your latest weekend trip to the
Hamptons, and your colleague might share that selfie you took at the
holiday party. Or you might just happen to walk past in the
background of the TikTok video of a complete stranger during their
Bahamas vacation. To let you in on the fun, your friends will probably
tag you. And even if they don’t, facial recognition algorithms can take
over that job in a matter of seconds. While you have control over the
pages you follow or the things you post about, pictures often have a life
of their own. And last time I checked, there was no way to simply leave
your face at home if you wanted to go unnoticed.
More importantly, social media isn’t the only snapshot of your
digital life. There are many other data sources that don’t require
curated input or even your direct cooperation to keep track of your
behavior. Your Google searches create a log of your most intimate
secrets and questions. Your credit card knows exactly what, where, and
when you buy. And your smartphone is equipped with an army of
sensors that collect your current location, screen for ambient light,
capture physical activity, and gauge how many social interactions you
have (e.g., by monitoring your calls and messages).
Resisting the temptation of using social media might be great for
your well-being. But it will not protect you from leaving the types of
digital traces that can be used to profile your psychology. As I will
show you in the next chapter, you can and will be seen.
OceanofPDF.com
 3
The Digital Breadcrumbs of Our Existence

I was standing next to the stage of a small auditorium at the W Hotel

in Chicago, waiting to deliver a talk on the science of happiness. With
just a few moments to go, my heart started to beat faster. I was
nervous—the type of nervous that makes you question your life
choices.
Why had I agreed to this? What did I (or anyone for that matter)
know about the topic of digital happiness? Was that even a thing or
just a made-up item on a corporate bullshit bingo list? But it was too
late for second guesses. Everything was ready. I was mic’d up, my first
slide was showing on the screen, and the moderator was about to
introduce me to the crowd.
That’s when one of the organizers pulled me aside.
The second speaker for the session hadn’t shown up. The organizer
had tried to call him multiple times, but no answer. Would I be
willing to speak for an entire hour instead of thirty minutes? Ah … a
bit of a surprise, but sure.
I was on fire. My audience was engaged, my jokes were landing,
and I had extra time for additional anecdotes and insights. In short, I
was crushing it.
But about twenty-five minutes into my talk, the organizers signaled
to me that the other speaker had arrived. I had to wrap up within the
next five to ten minutes.
Seriously? What a jerk! Shows up late and steals my extra time. I
hope he sucks.
 I wrapped up hastily—smiling on the outside but boiling on the
inside—and walked off the stage.
Fast-forward six hours. After a dinner reception with the
conference attendees, I find myself sitting in a bar with … well, yes,
“the other speaker.” He’s hot, smart, and funny. We have a few
cocktails, play Ping-Pong (I’m not holding back; this is not a friendly
match), and gossip about other academics we know. When the bar
closes, it doesn’t take much convincing to get me to come back to his
place.
As soon as we enter his apartment, I start snooping around. I want
to gauge his character. For all I know he could be a hot, smart, and
funny serial killer.
The first thing I notice is a huge library wall filled with books.
Books in English, French, and Hebrew about science, literature, and
art. They’re sorted according to topic and height, all perfectly aligned
with the front edge of the shelves. Who is this person? An organized
bookworm? Borderline obsessive-compulsive?
Next: drinks. I open the cabinets to look for glasses. They are all
perfectly spaced and sparkling clean. Not a single watermark. I
wonder if that’s his magic or his cleaner’s? I get my answer when I
return to the living room and attempt to put our drinks down on the
coffee table. He jumps up from his chair to place coasters underneath
the glasses. Borderline obsessive-compulsive, indeed.
The night went well. Very well. Fast-forward another four years.
The mysterious bookworm and I are exchanging rings in a small loft
in Manhattan, saying yes to a lifetime together.
The first impressions I got that first night in his apartment turned
out to be 100 percent accurate. My husband is the most curious
person I’ve ever met and can be a bit of an order freak. And yes, … he
also continues to be late.

Digital Breadcrumbs
Just like my husband’s apartment, our lives and the physical spaces
we inhabit are filled with cues about who we are. Some of these cues
are the type of intentional identity claims I discussed in the previous
chapter. The books we decide to stack on our shelves or the posters we
hang on our walls.
But other cues are created unconsciously. The messy notes on your
office desk, the concert tickets in your paper bin, or the skates next to
the door. Psychologists refer to these cues as “behavioral residue.” The
residue of our lives that just happen to be there.
I think of them as the footprints we leave walking along the beach.
Unlike identity claims, they are not intentional. They don’t serve as an
explicit signal to others. Behavioral residues are the by-product of our
lives. An unavoidable trace of our actions. But, while footprints on a
beach are ephemeral—the next wave will wash them away—the traces
we leave online are often permanent.
As much as I pride myself on being a female Sherlock Holmes, I’m
not the only expert snooper. The American psychologist Sam Gosling
has shown that people are remarkably accurate at judging the
personality of strangers when given the chance to snoop around in
their offices or bedrooms.1 They might take the Andy Warhol poster in
someone’s bedroom as a signal of openness or consider the
meticulously made bed and perfectly folded shirts as an indicator of
conscientiousness.
The same is true for our digital spaces. As in the analog world,
many of the cues to your inner life are created inadvertently without
you wasting a second thought on them. For example, you don’t have
an audience in mind when you type your most mundane or deepest
questions into a search bar. Can dogs eat watermelon? (They can.)
What is the meaning of life? (I don’t know, but I think dogs play a role
here.) Liberated from the judgment of others, we can ask Google
anything we want and get an answer in a matter of seconds.
Similarly, most of us don’t think excessively about the traces we
leave when swiping our credit cards in the deli around the corner or
using it for a seamless checkout experience at Amazon. Sure, we
spend some of our money in ways that signal our preferences to
others. Purchasing a Gucci handbag or splurging on a new Porsche is
probably as much of an identity claim as posting about fashion on
Facebook. However, most of our purchases aren’t directed at flashy
clothes and expensive cars.
And finally, you don’t actively encourage your smartphone to
collect information on your whereabouts 24 hours a day, 7 days a
week, 365 days a year. You might become conscious of this tracking
when you pull up Google Maps, but most of the time your phone
simply lurks in the background. It knows when you leave home to go
to work, which places you visit, and how much time you spend
walking, running, or driving.
All these data traces can generate remarkably intimate insights
into your life. And while you and I might be able to use these digital
breadcrumbs to peek into the psychology of our future husbands, our
snooping skills pale in comparison to those of computers.
Let’s take a closer look at three prominent types of behavioral
residue that can offer a glimpse into your psychology: Google
searches, spending records, and smartphone sensors.

Our Closest Confidant: Google

Google is often treated as a crystal ball that allows us to peek into
current trends as well as societies’ most well-kept secrets. In one of
my favorite books of all times (Everybody Lies; if you haven’t read it, I
urge you to put down this book right now to get it), the social scientist
and author Seth Stephens-Davidowitz explores five years’ worth of
Google searches to reveal the “truth” about America.2
A disturbingly high volume of racist jokes suggests that the country
is still more racist than we realize. A spike in searches for information
about how to give oneself an abortion may be the harbinger of an as-
yet-unacknowledged rise in back-alley abortions that are particularly
prevalent in places where it has recently become more difficult to get
one. And counter to commonsense assumptions about our sex lives
and habits, women are twice as likely to search for answers about why
their boyfriends won’t have sex with them than vice versa. Stephens-
Davidowitz’s observations are fascinating because they offer an
unbiased, dynamic bird’s-eye view on society. But Google searches can
do more than that. They also allow us to zoom into the psyche of an
individual.
In 2020, I was part of the documentary Made to Measure that
reconstructed the life of Lisa, a young woman in Austria, just by
looking at her Google searches, without ever meeting her.3 To bring
the Google searches to life and weave them into a personal narrative,
the production hired an actress to reenact the most intimate moments
in Lisa’s past. From her early childhood in a small community in
Tirol, to her first job as a waitress, and her vocational training as a
pastry chef in London, all the way to an existential life crisis.
The production not only reconstructed the places Lisa had lived
with an astonishing level of detail, but also her inner mental life. Look
at table 3-1 for some of the searches Lisa made:
The searches paint a picture of a struggling young woman who
fights against perfectionism and stress (category 1), suffers from
eating disorders and depression (category 2), drug use (category 3),
and physical health problems (category 4), and had to go through an
unexpected pregnancy and miscarriage (category 5).
None of these searches were made to signal Lisa’s identity. Yet, they
allow for a highly intimate glimpse into the darkest days and hours in
her life, far more intimately than Lisa expected. When she watches
the last scenes of her life—the loss of her unborn child—being played
out by the actress sitting across from her, she stops the live interview
to collect herself.
The filmmakers’ intuitions about the links between what we search
for on Google and our mental lives can easily be automated. Just as
you can train an algorithm to turn social media posts into predictions
of personality, socioeconomic status, or mental health, you can train
an algorithm to translate Google searches into psychological profiles.

TABLE 3-1

Lisa’s Google searches

Category Search terms

1 Time off important to survive

Melodramatic
Category Search terms

2 Size zero
Insanity training
Weight loss
Calories in sushi
Therapist for inner crisis
3 Snorting cocaine
Drug accessories spoon
4 One year reportedly sick
Acute chronic pharyngitis
Cold remedy
Bronchitis
5 First month of pregnancy
Bleeding at the beginning of pregnancy
Child lost in the first weeks

Given that Google searches are much harder to come by than

Facebook or X/Twitter profiles, the research on psychological
targeting in this context remains relatively scarce (which is probably a
good thing). But there’s no doubt about the data’s potential. As
Stephens-Davidowitz shows, searches for “depression” peak in areas
with high suicide rates. And while there are large differences in the
percentage of men who openly identify as gay across the fifty US
states, the identical volume on searches for gay porn suggests that our
Google searches might offer a far more truthful picture of our mental
lives than our public facade.
With mental health, homosexuality, and other intimate
psychological traits still being stigmatized in large parts of the world,
many people are turning to Google for help. Ironically, the same
action that might protect you from the judgments of your friends,
family, and neighbors also creates a permanent record of who you are
in Google’s database.
 Money, Money, Money …
After spending a year stuck at home in Chicago during Covid-19, my
husband and I decided to escape to Mexico for two months of sun and
writing. During the day, we sat on the veranda feverishly typing to the
sound of a green jay singing and the warmth of the sun on our skin. In
the evenings, we wandered the streets of Playa del Carmen in search
of spicy margaritas and guacamole, soaking in the vibrant atmosphere
of the city.
One night, we came across an amazingly talented young musician
who was singing and playing acoustic guitar. Mesmerized by his voice,
we stood there for a while listening. When he finished the song, we
broke into enthusiastic applause. What a magical experience. But
then he came over and asked for money. I felt terrible. We didn’t have
any cash on us. I knew it was true, but it sounded like a lame excuse
when I told him. To my surprise, he smiled, took out a card reader
from his pocket, and told us that he also accepts cards, PayPal, or
Venmo. That’s artistic innovation!
In the United States, only one in four transactions is still made in
cash. The rest are recorded by credit cards or mobile devices. In other
parts of the world (particularly Asia), there’s an even stronger
adoption of cashless payments. Paper bills and coins might soon be a
relic of the past. As the Mafia can attest, cash might not be the most
convenient method of payment, but it is far more difficult to trace.
Every time you swipe your card, you leave a trace. And as it turns
out, these traces are far more intimate than you might think. So much
so that they create a unique spending signature that allows others to
identify you among millions of consumers.
Say you live in Manhattan. Like any other New Yorker, you might
have a few spare coins and bills on you, but most of your transactions
flow through your credit card or phone. Now let’s say I got access to
credit card transactions of all the 8.5 million people living in New
York. I can see all their transactions but no names. Completely
anonymous. What are the chances I could tell which spending record
belongs to my husband? Sounds like finding a needle in a haystack.
Yet, my chances are close to 100 percent.
As some groundbreaking work by the computer scientist Yves-
Alexandre de Montjoye suggests, all I need to solve this who-is-who
puzzle is knowledge of about three of his purchases.4 If I knew that
my husband went to Starbucks on 72nd Street and Amsterdam at
8:42 a.m., had lunch at Yasaka Sushi at 1:33 p.m., and eventually took
a Yellow Cab from the Upper West Side down to Soho at 7 p.m., I
would find him in the data. There’s likely only one person with that
exact signature. Combine what a person buys with where and when
they buy it, and voilà, you have a unique fingerprint made up of $$$.
But swiping your credit card or tapping your phone can tell me far
more than this. Imagine you found the diary of a person who
describes spending $29.99 on a crop top at Forever 21, a double
cheeseburger at McDonald’s, and a subway ticket. Probably a young
woman. Just like other digital traces, our purchases offer a window
into our tastes, habits, lifestyles, preferences, and motivations.
More than that, spending is a form of self-expression and is often
highly personal. Yes, we all need clothes. But which specific clothes
you choose to buy is up to your discretion (at least in part). I might
decide to go for classic cuts in black, while you might put yourself out
there by wearing flamboyant, colorful runway outfits. And yes, you
might have to spend a good part of your income on essential goods
such as transportation and groceries. But most of us have at least
some discretion over the remaining parts of our income (or even the
means of transportation or the places where we do our grocery
shopping).

From $$$ to personality traits

In 2018, my colleagues Joe Gladstone and Alain Lemaire and I set out
to explore the relationships between what we buy and who we are.5
We collected Big Five personality profiles of over two thousand people
in the UK and asked them for permission to connect these profiles to
their bank accounts. For each person, we could observe every single
transaction they had made over the period of six months. We knew
what they bought and how much they had spent.
Because the universe of potential purchases is almost infinite, we
focused our investigation on a set of common spending categories
(e.g., supermarkets, fast food, art, books) and brands (e.g., Starbucks,
Tesco—a UK supermarket, or Pizza Hut).
Take a look at the two lists of spending categories in table 3-2. They
represent the high and low ends of one of the Big Five personality
traits. By now you have taken enough of my quizzes to pass this test
with flying colors. On the left, you have a person who saves their
money for the future and is concerned with looks. On the right, you
have a person who spends most of their money on phones, takeout
food, and snacks. It’s conscientiousness. It takes discipline to save, but
not to snack.

TABLE 3-2

Spending categories correlated with one of the Big Five personality traits

Continue reading to see which one.

Most positively correlated Most negatively correlated

Savings Lunch or snacks

Holiday savings Public transport
Tradesmen’s fees Mobile
Children’s clothes Cash
Beauty treatments Takeout

Source: Adapted from Joe J. Gladstone, Sandra C. Matz, and Alain Lemaire, “Can Psychological Traits Be Inferred
from Spending? Evidence from Transaction Data,” Psychological Science 30, no. 7, pp. 1087–1096, Copyright 2019
Joe J. Gladstone, Sandra C. Matz, and Alain Lemaire. DOI: https://journals.sagepub.com/doi/10.1177
/0956797619849435.

Take a look at another one (table 3-3). This one is extroversion. You
have the extroverted butterfly who spends money on taxis, clothes,
and fun nights out. And you have the introverted hermit who spends
money on a comfy home and furry friends.
TABLE 3-3

Spending categories correlated with extroversion

Most positively correlated Most negatively correlated

Dining and drinking Medication

Clothes Council tax
Taxis Home appliance insurance
Unsecured loan funds Home electronics
Medical (dental and eye) Pets

Source: Adapted from Joe J. Gladstone, Sandra C. Matz, and Alain Lemaire, “Can Psychological Traits Be Inferred
from Spending? Evidence from Transaction Data,” Psychological Science 30, no. 7, pp. 1087–1096, Copyright 2019
Joe J. Gladstone, Sandra C. Matz, and Alain Lemaire. DOI: https://journals.sagepub.com/doi/10.1177
/0956797619849435.

But the Big Five aren’t the only psychological insights we can
generate from spending. Without knowing anything about the next
trait we studied, take a look at the two lists in table 3-4.
What’s the image you form in your mind about the people in these
spending patterns? The person on the left donates their money to
charitable organizations, spends money on gym equipment, and
manages their finances through investments and savings. The person
on the right spends money on snacks, unsecured loan repayments,
and lifestyle products. A remarkably different approach to putting
your money to use. Any guesses?

TABLE 3-4

Spending categories correlated with a psychological trait

Continue reading to see which one.

Most positively correlated Most negatively correlated

Investment Lunch or snacks

Savings Lifestyle
Religious donation Cash
Gym equipment Phone (landline)
Continue reading to see which one.
Mortgage payment Unsecured loan repayment

Source: Adapted from Joe J. Gladstone, Sandra C. Matz, and Alain Lemaire, “Can Psychological Traits Be Inferred
from Spending? Evidence from Transaction Data,” Psychological Science 30, no. 7, pp. 1087–1096, Copyright 2019
Joe J. Gladstone, Sandra C. Matz, and Alain Lemaire. DOI: https://journals.sagepub.com/doi/10.1177
/0956797619849435.

Looks a little bit like conscientiousness again, no? Not far off! It’s
self-control—the ability to regulate our impulses, emotions, and
desires.
You can see how the person on the left makes a sacrifice in the now
by investing and saving money for a better future, and instead of
spending all the money on themselves, they give it away to charity. In
contrast, the person on the right appears to have a much harder time
resisting temptation. They snack, enjoy their life in the here and now,
and waste money on bank fees.
It doesn’t take much to imagine how such information might be
useful to banks trying to figure out which of their customers is worthy
of their trust when it comes to extending loans or other services (more
on the applications of psychological targeting in part 2).

Smartphones: Your 24-7 Life Companion

On the night of September 4, 2017, Jaila Gladden felt a cold coming
on. It was already shortly before midnight, but the twenty-one-year-
old senior college student decided to drive to the local Kroger store in
Carrollton, Georgia, to get tea and medicine before going to sleep. In
the store’s parking lot, a man approached her to ask for a lighter. Jaila
responded she didn’t have one and continued walking.
When Jaila arrived back at her car, she felt a knife pressed against
her back. The man she had encountered only moments earlier—
Timothy Wilson—forced her into the passenger seat, took her phone,
and started driving toward Atlanta. After raping Jaila in her car
behind an abandoned church, Timothy asked her to direct him to the
nearest gas station, which he planned to rob before taking Jaila to
Michigan.
Jaila’s survival instinct kicked in. She told Timothy that she
couldn’t help him find a gas station without Google Maps, and he
relented. He gave her phone back. Driven by panic and desperation,
Jaila started sharing her phone’s location data with her boyfriend,
Tamir Brant, telling him she had been kidnapped and was scared for
her life. Tamir informed the local police immediately.
The officers on duty followed Jaila’s digital GPS breadcrumbs and
eventually found the car in an empty parking lot. The lights were off,
but the engine was running. Timothy had fled the scene; Jaila escaped
to safety, and Timothy was arrested, hours later.
“If this victim did not have her phone …” the Carrollton police later
stated in an interview, “she may not have been as lucky.”6
Your smartphone might not have saved you from a kidnapper (or
identified you as someone who stormed the US Capitol, for that
matter), but my best guess is that it rarely leaves your side. Many of us
have become so attached to our smartphones that the consumer
researcher Shiri Melumad lovingly calls them “adult pacifiers.”7 It’s
easy to see why. I have at least one panic attack a day when I can’t find
my phone within sixty seconds. That’s what I call an anxious
attachment style! And I’m certainly not alone. The average person
unlocks their phone about fifty-eight times a day, clocking a total of
three hours and fifteen minutes.
But it’s not just the amount of time we spend on our phones that
makes them the ideal hunting ground for behavioral residue. It’s also
the fact that we use them to engage in many of the most meaningful
activities in life. We call or text our loved ones, capture the most
important moments on camera, and track our fitness achievement
using various apps.
Yet, what makes smartphones unique—truly unique—as snooping
devices (and different from Google searches or spending records) is
that they collect data on us, even when we are not actively using them.
That’s because they are packed with sensors. An army of master
snoopers that never sleep.
 Unless you turn off the GPS sensor, for example, it will track your
location. Continuously. Your phone is like a stranger that walks right
behind you and watches everything you do. But the GPS sensor is just
one of many. Any modern off-the-shelf smartphone includes an
accelerometer, Bluetooth, light sensor (to adjust brightness),
microphone, proximity sensor, and Wi-Fi. On top of that, there are
system logs that track activities such as calls, texts, app usage, or
battery status.
To be clear, none of these sensors were added to your smartphone
for the purpose of inferring personal characteristics about you. It’s not
like in the movies, where the bad guys secretly plant chips around
your house to listen to your conversations. The sensors are there to
make your experience of using your phone as seamless and effortless
as possible.
When you turn your phone to landscape, for example, the screen
automatically flips to widescreen. You probably haven’t given this
magical transition much thought before (unless that feature didn’t
work all of a sudden). But for your phone to know when to flip the
screen, it needs sensors to track your phone’s position in space.
Portrait? Landscape? Upside down? The sensor that provides this
information is called “accelerometer.” It picks up the X, Y, and Z
coordinates of your phone. Incredibly handy when it comes to
rearranging the display based on your current needs.
However, the accelerometer sensor can do a lot more than merely
flip your display at the right time. It can tell us, for example, whether
you are currently engaged in any type of physical activity, and if so,
which one. Say you are walking down the street. With every step, you
bounce up and down. Or you are riding a bike, which might shake you
up once in a while but otherwise provides a relatively smooth ride at a
certain speed above ground level. Data from the accelerometer sensor
can tell whether you are currently sitting, standing, walking, running,
cycling, or driving in a car—even though the sensor was never built
for this purpose.
And that’s just the beginning. The army of sensors in your phone or
smartwatch offer insights into your social interactions, daily activities,
as well as mobility patterns. If I had access to your data, I could tap
into your microphone sensor to determine whether you are currently
in conversation, or I could access the Bluetooth sensor to estimate
how many other people are around you.
I could also look at the GPS sensor to see where you are and what
you might be doing. Maybe you are in a coffee shop or a bar. If that’s
the case, you are probably engaged in some kind of social interaction.
Or you’re in an office building. In that case, you might be working.
These data points are informative in isolation, but they’re more
robust and insightful in combination. For example, I can combine the
accelerometer and light sensors as well as usage logs and battery
status to estimate the time you went to bed last night. Your phone
hasn’t been opened for a while, the light sensor says it’s dark, the
accelerometer fails to detect movement, and the battery is charging.
You have probably called it a day and are currently snoozing.
For psychologists like me, the ability to track your daily behaviors
and experiences with the help of your 24-7 smartphone companion is
a dream come true.

From smartphone logs to personality traits

How do these observations translate into psychological insights?
The two psychologists Gabriella Harari and Clemens Stachl
(together with other colleagues) have studied the Big Five personality
traits in relation to six broad categories of behavior that can be
detected via smartphones: (1) communication and social behavior, (2)
music consumption, (3) app usage, (4) mobility, (5) overall phone
activity, and (6) day- and nighttime activity.8
Take a minute to conjure an image in your mind of what the
smartphone behaviors of a highly conscientious person might look
like. Someone who is extremely organized and loves to plan. Someone
whom their friends would describe as dependable. How might their
smartphone traces differ from someone who takes a more flexible
approach to life and might be considered a tad unreliable and flaky?
There are a number of sensible candidates. Perhaps conscientious
people go to bed earlier and sleep more? Maybe they spend more time
at work and are less likely to be found in pubs? Or maybe they are
more likely to use serious apps like CNN rather than dedicating every
spare minute to social media and solitaire? All of this makes sense.
Yet, the main predictor of conscientiousness the research team
identified is a different one. The first time I heard about it, I couldn’t
help but crack up because it’s so true. Here’s a question for you: How
often does your phone run out of battery power because you forgot to
charge it over night? This happens to me all the time. Literally all the
time.
But if you are one of those lucky conscientious folks out there, this
might be a foreign experience. People in the study who scored high on
conscientiousness were far more likely to consistently have a battery
status of over 60 percent than their less conscientious counterparts.
It’s remarkable how such seemingly irrelevant signals can reveal so
much about who you are. In addition to charging status, high levels of
conscientiousness were also predicted by an increased use of weather
apps and timers (yes, conscientious people don’t leave anything to
chance) as well as more regular day- and nighttime phone usage. This
is a perfect example for the power of behavioral residue. None of us
would ever consider creating these traces deliberately to signal to
others how organized we are. Yet, we simply can’t help but create
these clues to our inner mental life.
Let’s reverse the game. I will give you a description of someone’s
phone usage pattern and you guess the personality trait this pattern
reflects. Imagine a person who frequently uses their camera, takes a
lot of pictures, is more likely to make calls at night, and writes long
text messages. You only have four personality traits left: openness,
extroversion, agreeableness, and neuroticism. Any guesses?
Extroversion, maybe? Close. Extroverts do text and call a lot more
often than their introverted counterparts.
But the examples are indicative of high openness. If you think
about the aesthetic affinity of people scoring high on openness, their
elevated use of the camera makes sense. And I can absolutely see how
open-minded dreamers and hobby philosophers are more likely to
turn the night into daytime and write long, poetic messages.
 From GPS coordinates to mental health
What else can we predict from the way in which you use your
smartphone? Your health, for example. The last decade has seen a real
explosion in health-tech apps that leverage wearable devices such as
smartphones, Fitbits, or smartwatches for data capture. Popular
applications like MyFitnessPal, Samsung Health, Apple Health, or
Google Fit allow users to monitor their health by tracking physical
activity (e.g., running, walking, number of steps), sleep (e.g., blood
oxygen levels, heart rate, time asleep, and sleeping respiratory rate),
and more.
Unsurprisingly, these insights can be useful indicators of physical
health. If you don’t move much, you are probably not in great physical
shape. And if you are not in great physical shape, then that elevates
your risk for a whole bunch of chronic health problems such as
obesity or cardiovascular disease. Pretty straightforward.
What is less obvious, however, is that we can also derive insights
about your mental health. For one, your physical health is directly
related to your mental health. On average, the healthier you are
physically, the healthier you are psychologically. It’s the old body and
mind cliché I described in the previous chapter that just happens to
be true.
But there is a lot more to the detection of mental health from
smartphones and other wearable devices than this. Research by
Sandrine Müller and me (alongside that of many others who have
shown similar results), for example, suggests that we can predict
whether a person suffers from depression just by looking at their GPS
records.9
How? If I were to show you the raw data, you probably wouldn’t be
very impressed. It’s essentially a list of longitude and latitude
coordinates that are associated with a particular user ID and time
stamp. An entry in the dataset might look like this:

ID = 85386
Longitude = 20.198209184832525
 Latitude = -87.4560316546014
Time = 2021-04-05T23:36:31+00:00
Not very revealing.
But these simple longitude and latitude coordinates offer a host of
metrics that can help us understand a person’s mental health. For
example, we could start by identifying where you live and how much
time you spend at home. From there, we can estimate how often you
leave your home, whether that frequency changes, how far you travel,
and whether your life has a certain routine.
I hope that you have never suffered from depression or had to go
through the experience of seeing someone else suffer from it. If you
do, you can probably see some parallels already. If you don’t, let me
list the core symptoms of depression according to the ICD-10
classification (International Classification of Diseases, 10th revision):
fatigue or low energy, loss of interests or pleasure, persistent sadness
or low mood.
While we can’t get a full take on all these symptoms using GPS
records, we can get a decent approximation of some of them. Fatigue
and low energy? You are probably less active and spend more time at
home. Loss of interests and pleasure? You might visit fewer places
and make changes to your weekly routing. That’s exactly what we find
in our study, and what other researchers have found in theirs. People
who suffer from depression spend more time at home, they travel
shorter distances, and visit fewer places. Put simply, they move less
and lose touch with their environment.
How accurate are such predictions? Could they replace diagnostic
tools such as interview screenings? The short answer is no. At least
not in isolation and without further probing.
The accuracy of predictive models in small, relatively homogeneous
student samples is remarkably high. We can accurately distinguish
between students who suffer from depression and those who don’t in
80 percent of cases (with a baseline of 50 percent, corresponding to a
coin flip).
However, this accuracy drops quite significantly when it comes to
predicting depression in larger, more heterogeneous samples that
represent the population at large. In this case, we only see a hit rate of
about 60 percent. Still better than chance, but nowhere near the 80
percent accuracy in our student sample. And certainly nowhere near
the level of accuracy we would expect from a diagnostic tool.
Imagine your doctor prescribing you antidepressants because they
are 60 percent sure you suffer from depression. Ludicrous. Yet, as I
will come back to in part 2 of the book, algorithmic predictions might
not require diagnostic levels of accuracy to be useful. Sometimes
prompting further inquiry from a doctor might be enough to make a
difference.

Fake It ’Til You Make It?

As our journey through the world of Google searchers, spending
records, and smartphone sensors showed, your behavioral residue can
reveal just as much about who you are as your explicit identity claims
(e.g., your social media profiles). And yet, they are different. If I
wanted to be seen as someone I’m not—say I’d love people to think of
me as the typical organized German—I could curate my identity
claims accordingly. I might follow serious news outlets on X/Twitter
and post pictures of my immaculate apartment on Instagram.
This masquerade is much harder to keep up with behavioral
residue. I don’t advertise the disorganized and chaotic side of me. But
my neighbors in the village likely picked up on it by watching me run
(and sometimes miss) the bus every morning. And my colleagues
come face-to-face with the chaos of my office desk daily.
The same is true for my digital life. I binge-watch Netflix on the
weekend. I get up at random times during the week. My phone
battery is dead way too often. And I’m purchasing two to three items
at the deli on the corner twice a day. Stepping into the shoes of
someone else might work for a moment, but it’s hard to sustain in the
long run.
One of the questions I am often asked in my talks is whether our
identities are really fixed. Why do you have to be an extrovert or an
introvert? Aren’t we a lot more complex and dynamic than this static
depiction of personality might suggest?
These questions hint at the broader concern that predictions of
personality from digital footprints—just like their survey-based
counterparts—put people in boxes that are too narrow to capture the
full complexity of their existence.
While most of us feel like we have a core identity, we are not always
the same. My neighbors in the village understood this. While they
might have figured out that I was generally disorganized, they also
knew that this tendency wasn’t always expressed in its full capacity.
Parents around? Less messy. Alone with my friends? Messier.
Which invites the question: How can big data help us understand
these nuances in our personality?
OceanofPDF.com
 4
You Are Not Yourself When You’re Hungry

In the late seventies, Billy Milligan was arrested for the brutal
kidnapping, robbery, and sexual assault of three young women in
Ohio. The evidence against the twenty-two-year-old was undeniable.
Not only did all three women identify Billy as the perpetrator, but the
defense itself testified that he had committed the alleged crimes.
But instead of serving a lifetime sentence in prison, Billy was
acquitted just a few months after the trial had started. By reason of
insanity. According to his public defender and doctors, Billy had
experienced severe physical and sexual abuse in his early childhood
causing his identity to splinter into ten—and later twenty-four—
separate personalities. There was Christopher, the thirteen-year-old
drummer who was terrified of mud. Tommy, the elusive sixteen-year-
old saxophone player. And David, the eight-year-old worry-free artist.
The defense argued that the “real” Billy—a talented, intelligent
young man—had not been present during the crimes he was charged
with. Instead, the rageful Yugoslav communist, Ragen, had been
responsible for the robberies, and the nineteen-year-old lesbian,
Adalana, had committed the rapes.
Billy was admitted to a mental health institution where he was
treated for a decade before being released in 1988. His remarkable
story was later captured in the Netflix documentary series Monsters
Inside.
The severe form of dissociative identity disorder Billy Milligan
suffered from impacts only about 1.5 percent of the global population.
However, in some way, we are all Billy. Just far less extreme and
criminal.
What I mean is that our personalities aren’t set in stone either. Each
of us comes in many different versions. I generally consider myself
agreeable: kind, warm, and cooperative. But you better not wake me
up before I’ve gotten my full eight hours of beauty sleep. Just like the
gremlins turn into little monsters when they are exposed to sunlight, I
turn into a big monster that is everything but kind, warm, and
cooperative.
I clearly don’t want to live with twenty-four personalities that
commit violent crimes when I’m not watching. But I appreciate that I
don’t have to live a unidimensional and boring existence either. Most
of us are neither unpredictable Billies nor static robots. We’re
somewhere in between.
We all have some kind of core identity, something that makes our
behavior predictable across time and space. But who we are, and how
we act, also depends on what is going on inside us and around us.
Take my husband, for example. As most Israelis I know, he’s
outgoing and social. Most of the time he is rather chatty, full of energy,
and confident. But sometimes even he doesn’t feel like socializing or
drawing attention to himself. Sometimes, all he wants to do is sit at
home and play Xbox.
If you sampled my husband’s behavior across many situations, you
would get what the psychologist William Fleeson refers to as a
distribution of momentary extroversion states.1 I’ve charted these
states in figure 4-1. Most of the time, my husband rates himself as
rather extroverted. And while he never feels extremely introverted,
there’s still a substantial amount of variation. He’s not always the same
(luckily, I like all versions of him).

FIGURE 4-1
My husband’s distribution of extroversion states

I’m more of an introvert, as seen in figure 4-2. I value my time alone

at home and don’t need constant stimulation. My version of a perfect
weekend is reading a book in the park, going for a long walk, or
watching Netflix in bed. And even though I love music and dancing,
I’m exhausted after a night out with a large group of friends. This
general tendency shifts my distribution of extroversion states to the
left.

FIGURE 4-2

My distribution of extroversion states

 But it doesn’t mean that there are never moments when I become
more outgoing, let loose, and shoot to the more extroverted right
(stepping into the classroom typically forces me to do this, for
example).
There’s a lot you can learn about my husband and me by looking at
those distributions. On average, my husband is a lot more extroverted
(average = 5.6) than me (average = 2.9). That’s the kind of
dispositional personality traits I’ve discussed in chapters 2 and 3.
But if our average extroversion score was the only thing you knew
about us, you would make mistakes when estimating how extroverted
we feel in any given moment. Sometimes you might overshoot,
sometimes you might undershoot. There are moments when I might,
in fact, be more extroverted than my husband (even though these
moments are rare).
The intriguing part of all this is that the deviations from people’s
trait averages are to some extent predictable. In other words, there are
ways to take educated guesses about where on our extroversion
distributions my husband and I find ourselves at a particular point in
time.

You’re Not You When You’re Hungry

In 2012, the American actress Joan Collins featured in a TV
commercial for Snickers. Set in a locker room, Collins, who is perfectly
styled and wearing a long blue evening gown, accuses one of the
football players of stealing her deodorant.
She is told off for “acting like a diva” and given a Snickers bar to eat.
She eats it and *poof* transforms back into a football player. This and
similar ads under the tagline “You’re not you when you’re hungry”
became a huge success in the advertising world. The equation was
simple: Hungry = cranky.
The advertisers were onto something. Mood has been shown to shift
people’s personalities: happy people tend to feel more agreeable,
extroverted, open-minded, and emotionally stable.2 It doesn’t matter
whether you’re generally more introverted or extroverted; being in a
good mood gives all of us a little bit of an extroversion boost.
The opposite is true for stress. I’ve already told you how I turn into
an evil gremlin when I’m tired. The same applies to when I’m stressed.
I become more confrontational, more likely to raise my voice, and less
likely to consider the feeling of others. And I’m not alone here. As
some of my latest work with Samantha Grayson suggests, feeling
stressed makes you more anxious but less extroverted and agreeable.
Situational contingencies like these can give computers an extra
edge when predicting your psychology. In addition to considering your
general tendency to be extroverted, they can factor in your current
mood or stress level when making predictions.
Not a trivial task, but possible. Just think of all the tracking devices
that capture dynamic snapshots of your experience without you having
to lift as much as a finger. There’s your smartwatch that can measure
your heart rate and skin conductance as an indicator of stress. Or the
webcam on your laptop that can be on the lookout for pupil dilation
and facial expressions as a signal of general mood and specific
emotions such as anger, surprise, or happiness. Or your phone that
tells me you didn’t sleep well last night and might be more irritable
than usual.
These predictions are by no means perfect. But they are good
enough to get a sense of your current emotional state. Equipped with
that data, we can make an educated guess of whether you are
operating within your typical range on a given personality trait or
whether you’re venturing into a slightly altered version of your usual
self.

How the Places We Visit Shape Who We

Are
Imagine you are sitting in a coffee shop with friends. It’s packed and
you have a hard time hearing your friend talk about what a business
mastermind Taylor Swift is.
 As you talk, the two of you can’t help but scan all the new customers
coming in: the guy in an impeccable Italian suit, a hipster sporting a
mullet and an ironic mustache, the woman who looks like a B-list
actress in the movie you can’t remember the name of. Occasionally you
strike up a conversation with a stranger who wants to claim the empty
chair at your table, or the waitress who’s checking in on you every five
minutes.
Now imagine a different scenario. You’re sitting alone in a library.
It’s so quiet the crinkling of a potato chip bag sounds like fireworks.
The people at the tables next to you are immersed in their reading or
typing away on their laptops. All you have is Kant and your thoughts.
How extroverted and conscientious would you feel in each of these
situations? My guess is that you feel more extroverted but less
conscientious in the coffee shop, and less extroverted and more
conscientious at the library. At least that’s what my research with
Gabriella Harari on the impact of places on people’s psychological
states suggests.3
Not a big shocker, perhaps. It makes sense that the social,
stimulating environment of the coffee shop would bring out more of
our outgoing, extroverted side but maybe make us feel a little bit less
organized since we could have spent the same time working or doing
something productive.
And yet, knowing which physical space you’re in has just given me
another clue to who you are. A clue that can easily be detected from
the digital traces you leave.
For example, I could start by asking the GPS sensor in your phone
to tell me your current location. Mapping these longitude and latitude
coordinates against Google Maps or popular apps like Foursquare
would quickly reveal that you’re at a coffee shop called Dear Mama
that is in the middle of a lively neighborhood in Manhattan.
According to Google, the coffee shop is packed around 3 p.m. And
the reviews on Yelp and Tripadvisor suggest that there’s typically loud
(and sometimes live) music playing all day. Cross-referencing this
information with my earlier work on places and personality states, it’s
safe to assume that you currently find yourself above your average
extroversion score.
 As the example shows, your digital footprints offer granular insights
into the physical environment you’re in. However, focusing on such
idiosyncratic cues makes it difficult to infer more general rules about
how certain contexts impact your personality.
It’s the same challenge I’ve described in the context of personality
assessments. I could do a much better job at describing all the nuances
of my husband’s personality to you if I didn’t confine myself to the Big
Five traits. But doing so would also considerably limit my ability to
compare my husband to other people and extrapolate my insights
beyond this one person.
The same is true for situations. We can describe them individually
in as much detail as we want. But doing so makes it hard to compare
them to one another and extrapolate to other situations.
Thankfully, there’s a personality framework for situations like the
Big Five for humans that helps with this.

The Psychology of Situations

The American psychologist Ryne Sherman set out to investigate the
psychological meaning of situations, asking a group of students at the
University of Florida to wear body cameras for twenty-four hours.4
Each thirty seconds, the camera took a picture of their surroundings.
The next week, the students came back to the lab to look at all their
pictures. They were asked to divide the pictures into distinct
situations. One of the students, for example, considered all pictures of
her coffee meeting with a friend as one situation, and all pictures taken
during a study session in the library as another.
Next, the students rated these situations on several different
characteristics: How social is this situation? How positive? How
intellectually stimulating? Think of it as the equivalent to
characterizing a person using personality attributes such as social,
dependable, or trusting (instead of using the Big Five model, situations
are typically assessed using the DIAMONDS framework, which
characterize situations along the following eight dimensions: Duty,
Intellect, Adversity, Mating, pOsitivity, Negativity, Deception, and
Sociality).5
Granted, this assumption might seem weird. Can we characterize
situations the same way we can characterize people? They are different
in so many ways. For starters, people are real entities in time and
space. We are born at a certain point in time and continuously exist
until we die. Situations are different. They don’t have a clear beginning
and a clear ending. They are fleeting. And they do not exist without at
least one person perceiving and acknowledging them. All of this is
true. And yet, when it comes to our perception of situations, we often
treat them the same way we treat other people.
Akin to our social judgments of others, we quickly form a first
impression of the situations we enter. A meeting room at the office? A
professional context. The living room of the frat house? A context
made for socializing and mating. Being able to quickly judge a new
situation has evolutionary roots and benefits. Our ancestors couldn’t
afford to take stock of all the situational cues in their environment.
They had to make quick judgments about whether a situation was
dangerous or full of opportunities.
But how do we measure the psychological characteristics of a
situation? It’s one thing to extract situational cues—who, what, and
where—from digital footprints such as GPS records. Buy how can we
turn these situational cues into psychologically meaningful situational
profiles?
The approach is analogous to that of turning your Facebook likes or
GPS records into insights about who you are. We train a model to
translate the raw input (situational cues detected from sensor data)
into personality scores for situations.6
But wait, isn’t there a big gap here? Situations cannot rate
themselves on a questionnaire. You can’t just ask them about how
much they agree with the statement “I am full of positive vibes.” Well,
you don’t have to. Just as Ryne Sherman asked the participants in his
experiment to tell him about their experience of the situations the
body cameras had recorded, you can ask people about their
perceptions of any situation you want to profile. That’s not just a
feasible fallback option, but the metric that matters.
 Situations don’t exist until they are perceived by at least one person.
And they only influence how we think, feel, and behave if we recognize
their psychological meaning in the moment. For example, I might find
myself in a sketchy part of town that should make me feel more alert
and scared than usual. All the situational cues might point in this
direction (e.g., rundown houses, dark alleys).
But if I am completely oblivious to these cues and do not recognize
them as a sign of potential danger—say, because I am drunk and don’t
pay much attention to my surroundings—then the situation itself is
not going to shape how I feel or behave in that moment. I might be
happily strolling through this dangerous part of town, singing to
myself, and skipping through the streets.
So even if situations could talk and self-assess their characteristics,
what matters in the context of shaping human behavior is how these
situations are perceived.
It is difficult for a computer (or a human) to tell from the outside
how a particular person experiences a particular situation at a
particular point in time. But it is much easier for a computer to predict
how most people perceive this situation most of the time.
Take the dataset Sherman collected. On the one hand, he had access
to the visual records of the situations his participants were in. On the
other hand, he had collected their subjective personality ratings of
these situations. In the same way that you can predict a person’s Big
Five from their Facebook profile pictures, you can now predict a
situation’s psychology from the body camera snapshots.
Instead of asking people to capture and rate their own situations as
Sherman did, you could also collect a bunch of Google Street View
images and ask a larger group of people to share their perceptions of
these places. As some of my own work has shown, people tend to have
surprisingly high levels of agreement on whether a Google Street View
location is social, positive, or intellectual.
And pictures aren’t the only source of situational cues, of course.
Think of all the audio traces captured by your smartphone. What
might the sound of an espresso machine in a busy coffee shop say
about how positive and social a situation is? Or a subway train in the
middle of rush hour? Or a bird singing in the park against the
backdrop of the city noise surrounding it?
Regardless of the specific contextual measures you use, the bottom
line for psychological targeting is clear: the better we understand not
just your static dispositional tendencies but also the dynamic
contextual factors that influence these tendencies, the more we can
learn about who you are in the here and now.
It’s the intuitions my fellow villagers built so naturally over time.
With parents? More conscientious and reserved. With friends? More
careless and outgoing. Their experiences of and interactions with me
shaped their understanding of the kind of person I was. But they also
understood that I was just as much influenced by situational forces as
they were.

I have spent the last three chapters exploring how all the digital
footprints we leave during our daily interactions with technology—
either deliberately or unintentionally—can reveal a lot about who we
are. A finding that still fascinates me after almost a decade of
researching the topic.
But even though computers decipher our psychology with
remarkable accuracy, there’s still a lot of guesswork involved. The
predictions are by no means perfect, and in many instances a smart
sales representative might very well outperform an algorithm after just
a few minutes of conversations.
However, we are all but at the very beginning of our journey. The
psychological insights we can derive from people’s personal data are
poised to become more and more accurate over the next decade. That’s
not just because the methods we have available to analyze personal
data are becoming increasingly sophisticated but also because new
technologies will introduce ever more granular data.
Today we have Google on our browsers, smartphones in our
pockets, and cameras in the streets. Tomorrow, we might have smart
lenses on our retinas, micro robots in our bloodstreams, and chips in
our brain. The moment you become the camera, the health tracker, or
the search engine yourself, much of the speculation and educated
guesswork involved in our current predictions of your thoughts,
feelings, and behaviors will evaporate.
What might sound like a dystopian nightmare or an episode from
Black Mirror is closer than you might think. Google and Samsung
have long been working on smart glasses and smart lenses. Between
2019 and 2022, Global Data’s Patent Analytics registered over sixty
patents for health-care-related microbots. And Elon Musk’s company
Neuralink is feverishly working toward a brain chip interface that can
be implanted inside our most secretive vault.
At the same time, rapid advances in AI promise to make predictions
of our inner mental lives more accessible than ever before (for better or
worse!). Most of the insights I described in the last three chapters rely
on machine learning algorithms that have been trained for a specific
purpose. With the right data, we can teach a computer to turn
Facebook likes into predictions of personality traits. Or to translate
GPS records into mental health diagnoses.
With the rise of more versatile machine learning models, including
generative AI such as ChatGPT, you no longer need large training
datasets or dedicated models that can only be applied to one specific
task. As some of my work with Heinrich Peters shows, ChatGPT can
predict your personality when given access to your social media posts
and asked to rate your Big Five profile or when tasked to have a free-
flowing conversation with you.7 Remarkably, it does so with roughly
the same accuracy as the dedicated models I discussed in chapter 2,
but without ever having been explicitly trained to do so.
If you combine the insights from the past three chapters with my
predictions of what the future might hold, it becomes impossible to
neglect the elephant in the room: What does all of this mean?
People in the village didn’t just collect intelligence for the sake of
knowing. They collected it for the sake of trading secrets and
influencing each other’s lives. The same is true for psychological
targeting. Most governments or companies aren’t merely interested in
getting to know you (as captivating as you might be). They are
interested in leveraging the knowledge about you to influence your
thoughts, feelings, and behaviors.
 So that invites the question: How could the ability to peek into the
psychological needs and motivations of millions of people empower
others to shape not just individual behavior but the course of society at
large?
And, lastly, should we look toward this future in fear of a new form
of tyranny or with anticipation of a better life?
OceanofPDF.com
 PART TWO
THE BRIGHT AND DARK SIDES OF PSYCHOLOGICAL
TARGETING
OceanofPDF.com
 5
Psychological Insights in Action

On December 3, 2016, the world was introduced to psychological

targeting, and I woke up to a new world. “Have you seen this article?
Are you involved in this craziness?” a friend texted. She wasn’t the
only one. Overnight, I’d received hundreds of messages from friends,
family, and … journalists.
What the hell!? What could I have possibly done to trigger such an
unexpected surge in attention? I was in Austin, Texas, at the time on a
research exchange trying to wrap up my PhD. My days (and most
nights) were spent in coffee shops crunching numbers and writing. A
lifestyle that wasn’t exactly conducive to a major scandal or news
story.
Heart racing, I hit the link in my friend’s message. The moment I
saw the familiar face of Michal Kosinski next to the article’s
provocative headline (“I only showed that the bomb exists”), it finally
clicked: Cambridge Analytica.1
According to the article’s authors, the British PR company had
accessed the Facebook data of millions of unwitting US voters to
predict their psychological traits and target them with personalized
advertising campaigns. Most worryingly, Cambridge Analytica had
targeted undecided voters among the most vulnerable populations to
spread misinformation and prevent them from showing up in support
of Hillary Clinton.2
I wasn’t involved with the company, as my friend feared. But I was
the only scientist at the time who had studied this type of
psychological targeting. And I had spent hours on the phone
explaining my findings to Hannes Grassegger, one of the two
journalists who first broke the story. Grassegger wanted scientific
evidence supporting Cambridge Analytica’s claims that psychological
targeting had secured Trump’s victory in the 2016 US presidential
election. I let him use some of my unpublished findings, which meant
that my name featured rather prominently in the article.
It wasn’t the first time the media had written about Cambridge
Analytica and psychological targeting. A year earlier, the Guardian
had published a similar article on its involvement in the Ted Cruz
campaign.3 I had expected—and hoped for—a public outcry. But back
then nobody cared. Instead of raising the alarm, most media outlets
lauded Cruz for his innovative use of data (just as they had with
Obama in 2008 and 2012), and the public went with it. In 2016, that
narrative changed. People saw how much was at stake.
Over the next few days and weeks, I spoke to countless reporters
and published multiple op-eds. I commented on the story itself, but
also advocated for a more nuanced and fruitful debate on the topic.
On the one hand, I was frustrated by the simplistic, apocalyptic
narratives I saw in the media (and later, books like Chris Whiley’s
Mindf *ck). They portrayed psychological targeting as a hypereffective
warfare and brainwashing machine. Let’s be clear: it isn’t. Even with
the most accurate understanding of a person’s psychological profile,
you won’t turn a sworn pro-choice activist into a pro-life supporter or
convert an iOS fanatic into an Android lover simply by showing them
a few ads on Facebook.
Yes, by tapping into your psychological needs, I can influence how
you think, feel, and behave. But just as you struggle to convince your
steadfast Republican uncle to add pronouns to his email signature,
psychological targeting won’t easily change people’s core identities.
Sure, these miracles can happen. But typically, they don’t.
On the other hand, I was equally irritated by the denialist
counternarratives that portrayed psychological targeting as nothing
but a hoax. My fellow villagers didn’t need to change my core identity
to influence my behavior. Just as you don’t need a magical
brainwashing machine to win elections. Most are won by small
margins.
Remember the 537 votes that swung the 2000 presidential election
to George W. Bush? Why waste your time casting a spell on the
diehard voters to win? All you need is to convince the people who are
yet undecided (between 10 and 25 percent in most US presidential
elections) to vote for your candidate or not vote at all. And I knew
psychological targeting had the potential to do that.
I never cared much about Cambridge Analytica (even though I’m
glad others investigated it more thoroughly). What I cared about was
having a broader public discussion about psychological targeting and
its potential impact on our lives. A discussion that wasn’t black and
white, or based on the claims of a boasting and later desperate PR
company.
I wanted a discussion informed by science and a thoughtful
conversation on how psychological targeting could be used and
abused to alter our individual and collective choices. Because just like
any other technology, psychological targeting is a tool. A tool whose
effectiveness needs to be established by real evidence (not just claims)
and whose impact depends on its intended purpose (not its mere
existence).
That’s what part 2 of this book is all about. An invitation to dive
into the science of psychological targeting and explore its potential
impact from different angles.

Putting Psychological Targeting to the

Test
When I joined the Cambridge Psychometrics Center as a PhD student
in 2013, my colleagues had just published their first paper on how
people’s intimate personal traits could be predicted from their
Facebook likes.4 The media loved it. And so did I.
What excited me the most about my colleagues’ predictive
personality models was their palpable potential for personalized
services and experiences. I was particularly interested in the health-
care sector. It seemed obvious that a better understanding of people’s
needs and motivations could improve health outcomes.
For context: up to 50 percent of all premature deaths in the United
States could be prevented by lifestyle changes. Take heart disease. The
most important risk factors include high blood pressure, smoking,
obesity, poor nutrition, and physical inactivity, all of which are well
within our control. Similarly, almost two-thirds of Americans suffer
from a chronic disease that requires them to take prescription
medication. Yet, only about 50 percent of them take their medication
as recommended, leading to 125,000 preventable deaths and $300
billion in excess costs every year.
Having just started my PhD, I was full of optimism. Could we use
psychological insights to help people take their medication? Could we
encourage them to have their annual checkups? Or motivate them to
eat healthier? I had a gazillion ideas for how understanding people’s
preferences and motivations could help us turn them into better
versions of themselves.
I put together a one-page proposal and reached out to potential
partners. I was confident that we’d be rolling in a matter of weeks. My
ideas were novel and promising. Even better, I offered my services for
free. It didn’t take long for my dreams to be crushed.
The health-care sector was a hopeless case: extremely risk-averse
and insanely bureaucratic. And—contrary to my expectations—the
people involved weren’t dying to have a millennial with no experience
or evidence tell them how to fix the health-care system.
I would have never finished my PhD in three years if I had stuck
with health care. Luckily, as one door closed (or rather stayed shut,
being barricaded from the inside), another door opened. One that led
me into a world that was a hell of a lot more excited about me and my
ideas. The world of marketing.
 One Size Fits No One
Since the early 2000s, there had been a growing push toward targeted
advertising and consumer analytics. You might have come across the
story of the retail giant Target. In 2012, Target made the headlines
after sending a voucher for baby clothes to a teenage girl before her
parents knew she was pregnant.
Marketers understood that consumers weren’t all created equal.
And they realized that the internet could provide new opportunities
for personalizing a consumer’s experience that hadn’t existed in
previous mass media such as TV or radio. As an old-school marketer,
you might have placed your ads for kitchen equipment in the
commercial breaks of TV shows made for housewives. But that was
pretty much it. The internet offered a whole new world of
personalization. As a savvy digital marketer, you could now advertise
to women between thirty-five and forty-five who were recently
married, had kids, and loved to watch The Great British Bake Off.
The ability to target people based on specific sociodemographic
characteristics or interests was revolutionary when compared to the
one-size-fits-all approaches of TV or radio marketing. But to me, a
personality psychologist, it was still only scratching the surface. For
starters, we have come far enough that it’s not just women looking to
buy kitchen equipment. Most sociodemographic targeting is a gross
overgeneralization.
But it’s not just that. Even among the illustrious pool of men and
women interested in cooking, you’ll find a lot of variation. Variation in
lifestyles. Variation in interests. And variation in motivations.
Some hobby cooks might be in the market for equipment that saves
them time in their busy schedules (e.g., a Thermomix). Others might
want to satisfy their desire for experimentation with the help of the
latest cooking tech (e.g., a KitchenAid, sous vide, or an Instant Pot).
And yet others might simply enjoy treating their friends and family to
a delicious meal cooked using some of the old classics (e.g., a Le
Creuset pot or an olivewood rolling pin). To understand these
differences, you need to go deeper than people’s sociodemographics
and past behaviors. You need to get down to people’s psychology.
The more I learned about the world of marketing, the more
convinced I became that it was the ideal testing ground for my ideas.
Marketers were willing to take risks and experiment. And they
celebrated every improvement—no matter how small. Could I manage
to increase the number of purchases by 2 percent? Great. 5 percent?
Even better. 10 percent? Please take our firstborn son.
I did not fully appreciate this when I started, but even a very small
uptake in advertising effectiveness can turn into sizable profits for
companies. And in many cases, the marketing budgets are rather
enormous. In 2021 alone, over US$512 billion was spent on digital
advertising. The marketing environment seemed perfect for me.
It didn’t take long to find a committed industry partner. The PR
company Grayling was hired by Hilton Hotels to explore the potential
of psychological targeting. Could we help Hilton create a richer and
more personalized customer experience? Challenge accepted. Once a
week, I took the train to London to work from the Grayling offices—a
welcome break from my rather monotone days in Cambridge.
For several weeks, we planned the launch of an interactive
application that could generate personality-based vacation
recommendations. Introverted? The soloist, with recommendations
to quiet and relaxing destinations. Neurotic? The all-inclusive worry-
free vacation, with nothing left to chance. All a customer had to do
was log in to their Facebook account. Our predictive algorithms took
care of the rest.5
The campaign was a huge success. Over sixty thousand people used
the application in only three months. They loved the experience. They
clicked, shared, and purchased (which, for Hilton, meant higher
brand visibility and profits). Together with Grayling and my
colleagues at the Psychometrics Center, I won bronze at the Travel
Marketing Awards and was named one of the most influential people
in data-driven marketing by DataIQ. The collaboration was an
invaluable stepping stone for me. It both gave me the confidence that
psychological targeting was effective and provided me with a
successful case study to approach other companies.
 And it didn’t take long for my second opportunity to materialize.

Beauty Is in the Eye of the Beholder

An online beauty retailer approached me with a simple goal: increase
its online orders.
My initial idea was to replicate what we had done for Hilton. But it
quickly became clear that this wasn’t an option. There were two
challenges. First, our approach for Hilton had relied on the
availability of different vacation types that could be matched to
personality traits. Categorizing beauty products according to
personality wasn’t as straightforward. Who buys body lotion and
makeup? Women who are extroverted? Open-minded? Agreeable?
We all do.
I found inspiration in the writings of Sidney Levy, a marketing and
consumer behavior icon of the twentieth century. In a Harvard
Business Review article published in 1959, Levy argued that spending
was not just functional but also symbolic.6 In other words, people buy
products not just for what they can do but also for what they mean.
Take a simple example: What does buying flowers mean to you? If you
are agreeable, you might buy flowers to make other people happy. If
you are open-minded, you might appreciate the aesthetic beauty of
flowers. Or if you are neurotic, flowers might help create a relaxed
atmosphere at home. No matter the personality trait, you can find a
way to make flowers meaningful.
The implications of what Levy had suggested seemed promising. If
the same product could mean different things to different people,
then all we had to do was identify what beauty products meant to
different audiences. Targeting extroverts? They want to be seen. They
want to be the center of attention. How about an ad featuring a
woman at the center of the dance floor surrounded by friends (or
strangers)? She’s wearing bright makeup and a sparkling outfit. All
eyes on her.
Targeting introverts? They don’t need attention. On the contrary,
they want to make the most of their “me” time. How about an ad that
highlights the more introspective side of beauty? A woman enjoying
the beauty retailer’s products from the comfort of her home. A relaxed
setting, no external distractions.
That’s exactly what we suggested to the beauty retailer and
eventually tested in a series of Facebook campaigns. We designed
both extroverted and introverted ads to be targeted at extroverted and
introverted women.
Our second challenge was that the beauty retailer had neither the
budget nor the desire to create an entire application around its
campaign. If psychological targeting was to become a serious
alternative to other segmentation approaches, we needed an easy way
to implement it.
If you want to target women ages thirty to thirty-five, who live in a
particular country or city, and who have shown an interest in beauty
products, all you need to do is tell Facebook or Google. They will find
those people for you. You can’t do that for psychological traits. None
of the big advertising platforms allow you to target personality traits
directly. Not because they can’t and haven’t thought about doing so—
Facebook filed a patent to predict the personality of its users from text
back in 2012.
But that doesn’t mean you can’t target people based on personality.
You can, indirectly. Facebook and Google allow you to define
audiences based on interests, the same interests we know to be
associated with introversion and extroversion (see chapter 2). If you
like video games, manga, or Terry Pratchett, chances are you’re
introverted. If you like Entourage, Shwayze, or beer pong, you’re
probably extroverted.
Of course, this isn’t true for everyone. You might like video games
but consider yourself extroverted. That’s possible. These relationships
are probabilistic, not deterministic (think back to Masao Gunji, the
sixty-seven-year-old Hello Kitty–loving police officer). They tell me
who you likely are, not who you are. You might be the extroverted
unicorn that is crazy about manga. But in the absence of any other
information, my best guess is that you are introverted. If Facebook
allows me to target people who like manga, and I know that liking
manga is associated with introversion, then I can effectively target
introverts.
We had everything we needed to kick off the campaigns.
Advertising materials for introverts and extroverts? Check. A way (as
crude as it may be) to target introverts and extroverts on Facebook?
Check.
I was a nervous wreck for the entire week the ads were running. All
of us had invested so much time and energy. Now there was nothing
left to do but wait, let consumers click and purchase, and see what
came out on the other side.
And … it worked! Those women who were targeted with messages
matching their personality were more likely to purchase. And the
difference in purchases was surprisingly large: a 50 percent increase
in the matched groups compared to the mismatched ones. For every
one hundred women we convinced to purchase in the mismatching
conditions, we had an additional fifty in the matching ones.7
I had, of course, been cautiously optimistic, but even I was
surprised by the magnitude of the effect. I was aware of the study’s
limitations. Our targeting approach was crude, and our customization
minimal. We hadn’t profiled individuals with our predictive
algorithms, but instead relied on audiences defined by a single
Facebook like. I’m sure we ended up with more than a few
misunderstood extroverts in our introverted target audience and vice
versa.
We also hadn’t personalized the customer journey beyond the
initial ad the women saw. Once a potential customer clicked on the
ad, the website of the beauty retailer looked the same for all
customers. If we could get an uplift of 50 percent with such a
rudimentary approach, surely, we could aim even higher with more
sophisticated and holistic interventions.
The beauty retailer’s creative team had done a fantastic job coming
up with ads that matched the personality traits of our target
audiences, but just imagine what a herculean task it would have been
to customize every word and picture on every page. This could have
meant thousands or even millions of unique creative elements. No
company in its right mind would assign such a task to a human
workforce.
But it could soon deploy a digital workforce. Computers can
predict your personality based on the images you like or post (see
chapter 2). You can train similar models to predict the personality
affinity of an image—the personality of people who the image would
most likely appeal to. My colleagues and I did just that. We built an
algorithm that could “look” at an image and tell us, for example,
whether it might be a good fit for someone extroverted or open-
minded. We never tested our algorithm in the wild. But we showed in
several lab experiments that selecting images to match consumers’
personalities impacted both their brand attitudes and purchase
intentions.8
But computers aren’t just good at understanding and selecting
content. They can also create content. OpenAI’s GPT models can
produce text that is indistinguishable from that of a skilled human
writer. GPT has written articles for the Guardian, published an
academic paper about itself, and won numerous poetry competitions.
And as some of my latest work has shown, GPT can convincingly
produce marketing content in the voice of different personalities. For
one of our studies, I prompted GPT3 to create short ads for the
iPhone 14. Its suggestions were spot on.
For the extroverted and enthusiastic target, GPT spit out: “If you’re
the life of the party, always up for a good time, and enjoy being
surrounded by people, then this is the phone for you! With its bright,
colorful design and built-in social media features, the iPhone 14 Pro is
perfect for extroverted, enthusiastic people like you. So come on, let’s
party!”9
For the artistic and open-minded target, GPT suggested: “If you’re
looking for a phone that will help you open up to new experiences and
be more artistic, look no further than the iPhone 14 Pro. With its
powerful cameras and editing tools, you’ll be able to capture and
create beautiful images and videos like never before. So, whether
you’re a budding photographer or just someone who appreciates art,
the iPhone 14 Pro is the perfect choice for you.”
 What I find most impressive about these messages is that they
aren’t just direct references to people’s preferences and character
traits. That’s easy. What’s intriguing is GPT’s crafty selection of
different iPhone features that are likely to appeal to different
audiences. An extrovert might not care about editing tools, but an
open-minded, artistic consumer likely will. Which is exactly what we
found across multiple different studies.
The personality-tailored messages generated by GPT not only got
our study participants more excited about the products we advertised,
but they also increased the dollar amount they were willing to spend
on them.10

Beyond Consumer Products and

Personality Traits …
The ability to sell a few more products by matching marketing
content to people’s personality traits might not have you at the edge of
your seat. But for me, the findings reveal something much more
fundamental. They show that tapping into your psychological
motivations gives me power. Power to influence the decisions you
make.
I intentionally kicked off our exploration of psychological targeting
on (relatively) neutral grounds. I wanted us to run through the basics
before returning to more controversial contexts. But I didn’t start this
chapter talking about beauty products. I started this chapter by
talking about Cambridge Analytica. Getting you to buy a few more
beauty products is one thing. But getting you to vote in a particular
way (or not vote at all) is entirely different. Or is it?
Not really. Selling political ideas and candidates is a lot like selling
products. You need to understand your audience and cater to their
preferences. Knowing your counterpart’s personality might help you
in this endeavor. You might convince your agreeable neighbor to
support more generous childcare policies or get your neurotic friend
to endorse stricter data protection regulations. But the real game of
politics isn’t played in the arena of personality. It’s played in the arena
of moral values—the ethical compass guiding our judgments of what
is right and wrong.
According to the psychologist Jonathan Haidt, there are five moral
values that are innate and universal: care, fairness, loyalty, authority,
and purity.11 We differ in how much emphasis we place on each of
these values. You might care about loyalty, authority, and purity,
whereas I might give more weight to care and fairness.
If I understand your moral compass, I can influence your behavior.
Think back to my collaboration with Hilton. We got consumers to
engage with the brand and its products by matching vacation types to
consumers’ personality profiles (or traveler types). You can do the
same in politics. If I know which moral values you care about, I can
put the right ideas and candidates in front of you. Say you place a
strong emphasis on care and fairness. I might approach you with
policies related to equal pay or minimum wage. If, instead, you care
more about loyalty and purity, I can nudge you to support stricter
immigration laws.
Perhaps not surprisingly, our moral values are intricately linked to
our political ideologies.12 If you consider yourself conservative, you are
more likely to emphasize loyalty, authority, and purity. The opposite is
true if you think of yourself as liberal. In this case, you likely prioritize
care and fairness. Because of these associations, I don’t even need to
know your moral values to use them for psychological targeting. All I
need to know is your political leaning.
But pushing different political agendas to people with different
moral values is only part of the story. In our collaboration with the
beauty retailer, we sold the same product to different personalities.
In politics, you can sell the same political idea to different people.
And you can do so more effectively if you match your argument to the
moral values of your counterpart. It’s a form of psychological
targeting that psychologists Matthew Feinberg and Rob Willer call
moral reframing.13
Take the issue of climate change. You could argue that we have a
responsibility to care for those who come after us. It isn’t fair to expect
the next generation to deal with the mess we created. Even if we
might not experience the effects of our actions ourselves, we have a
moral obligation to protect the planet and those who will inhabit it
for centuries to come. You often hear arguments like these from
liberals who tend to emphasize care and fairness. If you consider
yourself a liberal, these arguments are likely to resonate with you.
But if you think of yourself as conservative, they might not resonate
as strongly. You might be more responsive to arguments highlighting
our duty to preserve and enhance the immaculate beauty of our
planet and ensure the purity of the air we breathe. After all, the planet
is our only home. We need to stand up for the Earth that has provided
us and our forefathers with a safe haven for centuries. Sounds very
different, doesn’t it? Instead of care and fairness, these arguments
focus on purity and loyalty—the favorites of conservatives.
As the work of Feinberg and Willer (and some of my own research
using customized messages created by ChatGPT) shows, moral
reframing is highly effective.14 We are more compelled by arguments
that match our own moral compass. That’s true even when we
originally disagree with the argument’s main premise (say, when
trying to convince a conservative person about the benefits of
immigration or gay rights).
Moral reframing might sound like a simple, winning strategy for
political discourse. It suggests that all we need to do is step into
someone else’s shoes. But, as it turns out, this task is challenging. In
fact, most of us fail at it rather spectacularly. We’re human. We see the
world through our own lens. Instead of arguing our case in a way that
speaks to our counterpart’s view of the world, we argue in a way that
emphasizes our own—a classic egocentrism bias that is hard to
recognize and even harder to overcome.
My fellow villagers often got into heated debates about the future
of the village. Which clubs should get funding and how much? Should
the local fire brigade have an initiation ritual or not? While these
debates didn’t determine the course of an entire nation, they were
nonetheless meaningful and polarizing. As each side argued their
case, the views and arguments flying across the aisle became more
and more extreme. The debate became less about finding a good
solution and all about winning the argument.
 Algorithms don’t have the same bias. They don’t care whether they
argue their own perspective or that of their counterpart. They simply
argue the way that is most effective in achieving the desired outcome.

Two Sides of the Same Coin

Moral reframing shows that the power of psychological targeting
extends far beyond consumer products. It has the power to get people
to think differently about important societal issues like environmental
protection, economic inequality, or gay rights.
Is this a good thing or a bad thing? That depends. My first reaction
to the research of Feinberg and Willer was “That’s incredible!” A way
to engage people with politics again and help them understand each
other’s perspectives? Maybe there is hope for democracy after all.
In an ideal world, every politician would go from door to door, sit
down with each of their constituents (both supporters and skeptics),
listen to their concerns, and patiently answer all their questions with
a focus on the topics that really matter to them. Of course, there’s no
way to do that. A politician can’t be everywhere at once. Instead, they
have staff and volunteers who speak to as many people as possible on
their behalf. In the United States, every election cycle sees thousands
of volunteers on both sides dedicate their time to this mission. They
answer questions, listen, and explain.
But even with this extended workforce, there’s a limit to what can
be done. Door-to-door canvassing is highly effective, but it is also
time-consuming and only reaches a very small portion of the
population—typically those who happen to live in strategically
important counties or states. But what about the rest of us? What
about the millions of Americans that decide to forgo their right to
vote because nobody took the time to ring them up or knock on their
door?
In the 2020 presidential election, that percentage of the population
amounted to a staggering 33 percent of all eligible voters, down from
40 percent in the 2016 election. Yet, for a democracy to thrive, we
need the engagement of its people. If only there was a more scalable
and inclusive way to reignite people’s excitement about politics …
Psychological targeting aims to do just that: understand people’s
motivations and activate them. I am not saying that psychological
targeting is the silver bullet to all the political problems we face. But I
do believe that—if done right—it could provide a way to engage
people in politics again. A way to save democracy.
At the same time, it’s easy to imagine how technological advances
like this could turn into dangerous ammunition in the hands of more
nefarious, antidemocratic institutions such as Cambridge Analytica
that might try to use psychological targeting to accomplish the exact
opposite: seed discord, fuel hatred, and encourage disengagement.
Cambridge Analytica didn’t use psychological targeting to cheer on its
own base. It used it to destroy its opponents. And with it, the
foundation of democracy.
Much of the public discourse leading up to the 2024 US
presidential election focused on the dangers of misinformation
online. Like everybody else, I am concerned about fake news. But
unlike most people, I am more concerned about slanted news—news
that might be factually accurate but have been heavily massaged to fit
a certain worldview. That’s not because slanted news is more powerful
at swaying attitudes and behaviors than fake news; it isn’t. It’s because
it is simply far more common. So much so that research by Jennifer
Allen has shown that the impact of slanted news on people’s behavior
is about fifty times larger than that of fake news.15
It is concerns like these that linger in the back of people’s minds
when they hear me speak about psychological targeting. They might
trust my good intentions, but they are worried that my research
delivers a blueprint to actors with more despicable goals. Should I
even be allowed to conduct research on psychological targeting if
that’s the case?
I understand the concern but disagree with the conclusion.
Growing up in the village taught me that understanding the rules of
the game is critical. The more I understood, the better I became at
navigating my life. I learned who was passing on information behind
my back and became better at spotting the subtle attempts at steering
my behavior. I got better at spotting foul play.
The same is true in the world of technology. We need to
understand how effective psychological targeting is and how it could
impact individuals and society not just for its potential benefits but
also for its (not despite of ) potential for abuse.
The only reason I can credibly share my thoughts and
recommendations with policy makers, business leaders, and
institutions like the European Commission is because I have studied
psychological targeting. I don’t have to guess whether understanding
people’s psychology gives you control over them. I know it. And I also
know that as long as controlling people brings profits and power,
businesses will keep investing in technologies like psychological
targeting—whether scientists conduct research on the topic or not.
Bottom line: psychological targeting isn’t going to disappear. On
the contrary, as the evidence supporting its effectiveness grows, and
technology continues to advance, we will see its applications expand
to other domains.
This trend isn’t fully visible yet. But I am asked at least once a week
if I’d be interested in joining a startup trying to enter the space with a
new product. Or if I’d be willing to help a company explore how it
could benefit from implementing psychological targeting. If we want
to stay on top of the game and take an active role in defining its rules,
we need to understand both the opportunities psychological targeting
presents and the challenges it poses.
Let’s start with the opportunities.
OceanofPDF.com
 6
Finding the Good

When I picked up the call from one of my old college friends, I could
tell right away that something was wrong. She asked me about life in
New York. I complained about the trash in the streets, the brutal
winter days, and my students being late to class. She responded with
an occasional “hmm,” but her mind was clearly somewhere else. When
I asked if everything was OK, she burst into tears.
A childhood friend of hers had taken his own life. Nobody had seen
it coming. She wished she had known he was struggling. Maybe she
could have saved him.
I felt terrible. My original goal had been to use psychological
targeting to help people. I put that plan on hold because it seemed too
hard to implement. As time went by and I got creeped out by the dark
side of psychological targeting (we’ll get to that in chapter 7), I
abandoned it altogether. I had become so obsessed with the obvious
costs of using psychological targeting that I lost sight of the potential
costs of not using it.
But the conversation with my friend served as a wake-up call. It
reminded me that psychological targeting also had the potential to do
good. Just like my neighbors’ actions could be extremely helpful and
reassuring, psychological targeting could help us become better
versions of ourselves.
A lot of my thinking since then has been guided by what if … ? What
if we could use psychological targeting, for example, to help people:
Improve their physical and mental health?
 Become politically engaged?
Learn more effectively?
Find jobs they love?
Save rather than spend more?
Become more environmentally responsible?
Expand their experience of the world?
This list is by no means comprehensive. I’m sure that after reading
through the examples in this chapter you can easily come up with your
own set of questions.
Maybe you’re interested in education and how psychological
targeting could help kids and young adults learn more effectively while
having fun. Or maybe you feel like overhauling the way we find our
professional calling by using psychological targeting to help people
find jobs they love.
I picked three of the topics to focus on: the potential of
psychological targeting to support us in accomplishing goals that many
of us struggle with (e.g., to save more), democratize access to
personalized mental health care, and expand our experience of the
world.
I didn’t choose these three applications because they are the most
important ones or because they seamlessly fit into a simple and
coherent narrative. Rather, I chose them to highlight how
psychological targeting could be used for the greater good across a
wide range of potential applications. Some of the ideas I will share are
nothing but lofty dreams—fantasies of mine that I hope will
materialize in the future. Others are far more concrete examples of
how psychological targeting is currently benefiting individuals.
Let’s start with the latter.

The Savings Struggle

I spent much of my early career trying to get people to spend more.
Could we get people to find the perfect vacation by tapping into their
traveler personality? Yes. Could we persuade women to buy makeup by
tailoring our marketing messages to their extroversion level? Yes,
again.
I’m not saying that this is necessarily a bad use of psychological
targeting. My own research has shown that people are happier if they
manage to align their spending with their psychological needs, and
with access to products and services from all around the world, we
need a way to separate the wheat from the chaff. 1
But is helping people spend their money really what our society
needs the most? What if, instead, we could use psychological targeting
to do the exact opposite? Help people save.
When I first got interested in studying financial health, I was
shocked by the numbers I found. In 2020, 53 percent of Americans
reported living from paycheck to paycheck, 62 percent did not have
enough savings to cover three months of living expenses, and more
than 10 percent could not even cover a single week without getting
paid.2
To call this state of affairs problematic would be an understatement.
It’s disastrous. The 30 million Americans with essentially no savings
are at the constant cusp of ruin. They might be able to hang on this
week, but what if next week their car breaks down? They don’t have
the money to take it to the shop and get it fixed. This means they can’t
drive to work but spend hours on public transportation that is
notoriously unreliable (if available at all). Consequently, they might
lose the job that covered their rent, insurance, and weekly expenses.
One seemingly small incident and they are done. Game over.
The picture might not look quite as dire for the 200 million that
have less than three months of savings, but it isn’t rosy either. Not only
are most of us woefully underprepared for retirement (which, thanks
to medical advances, will last longer and longer), but there is growing
scientific evidence showing how financial distress holds us back in the
present. It adversely impacts our physical and mental health. And it
hijacks the cognitive bandwidth we need to make good decisions and
be creative. You simply can’t live up to your full potential if you are
worried about our finances.3
Most of us know that we should save more, and many of us are
eager to do so. According to Forbes, 30 percent of people started 2023
with the New Year’s resolution of becoming better at managing their
finances.4 And yet, the majority of these 30 percent will fail to translate
their noble intentions into dollars saved in their bank account. Like
many of the other New Year’s resolutions we struggle with, such as our
desire to eat healthily and exercise more, saving doesn’t come easy.
Even for the most financially literate among us, saving is a constant
battle—a battle with our brain that much prefers to savor the current
moment instead of worrying about the future. Saving is hard because it
means giving up a real, tangible reward today (i.e., our ability to buy
stuff with the money we’ve earned) for a potential benefit in the future
that is often far less tangible (i.e., a chance to deal with unexpected
emergencies)—a concession our brain is notoriously unwilling to
make.
Trading your new PlayStation or pearl earrings for a few hundred
extra dollars in your bank account will seem like a no-brainer once you
hit an emergency and all hell breaks loose. But we all know how it goes
before we get to this point. We want to be responsible superheroes but
end up defaulting to our regular, self-indulgent selves.
I was convinced that psychological targeting could help. An obvious
starting point for my investigation was to identify at-risk profiles. I
was curious whether certain types of people had a harder time
managing their finances than others. The most likely suspect among
all the personality traits I could think of was conscientiousness.
I remembered my highly dependable and reliable sister who always
paid her bills on time and magically managed to accumulate savings
even while being a poor student. And then I thought of myself, the
somewhat more careless and disorganized member of the family who
was quick at spending money and typically had a lot of month left at
the end of the money.
Not a terrible guess. It turns out that conscientiousness is indeed
related to some aspects of savings. But not as strongly and consistently
as I had expected. However, there was another personality trait that
reliably predicted financial health. Any guesses? We still have
openness, extroversion, agreeableness, and neuroticism in the race.
To be fair, if you had asked me a few years ago, I wouldn’t have
guessed the right answer. I was puzzled at first when study after study
kept showing that agreeable people tended to end up worse
financially.5 Agreeable people had fewer savings in their bank account,
accumulated more debt, and were more likely to default on loan
payments. Similarly, US counties with higher average levels of
agreeableness experienced higher levels of bankruptcy.
Instead of the careless, disorganized slouch I had pictured in my
mind, it was the friendly and caring nice guy that was struggling the
most. The type of person we love to welcome to our communities and
social networks because they put others ahead of themselves.
I had heard the saying “nice guys finish last,” but it didn’t necessarily
make much sense to me in the context of financial decision-making.
So, I started to dig deeper. What was driving the relationship between
agreeableness and poor financial health in my data? Could it be that
agreeable people didn’t negotiate as aggressively as their disagreeable
counterparts? The answer is no (to be precise: agreeable people are
indeed less aggressive, but that doesn’t explain why they do worse
financially).
Instead, the relationship was driven by something far more obvious:
Agreeable people simply didn’t care as much about money as their
disagreeable counterparts. In our research, we asked people how much
they agreed with statements such as “There are very few things money
can’t buy” or “You can never have enough money.” Agreeable people
consistently indicated that money just wasn’t that important to them.
When I first saw this explanation, I was both disappointed and,
frankly, a little disheartened. I really wanted to help the nice guys do
better financially. Teaching them how to negotiate more effectively
(not necessarily more aggressively) would have been a great
intervention. But how could I help them if they simply didn’t care as
much? It didn’t feel right to try and make money a higher priority for
them. On the contrary, I thought it was somewhat endearing that they
didn’t care as much. Clearly, they were good people who chose positive
social relationships with others over money. Admirable.
But the more I thought about it, the more I realized how flawed my
thinking was. Just because someone cares about money doesn’t mean
they don’t care about other people. The same way that caring about
other people doesn’t mean you shouldn’t care about money. It’s a false
dichotomy.
Think of it this way: if you don’t manage your money properly, you
are putting your loved ones at risk, too. And your love for others
doesn’t simply disappear just because you also care about money and
what it can do for you and others.
I was intrigued. If it was true that agreeable people simply didn’t
care as much about money, then arguing that they should save for the
sake of accumulating money wasn’t going to convince them. But what
if we could reframe the purpose of saving to highlight the potential
impact on their loved ones? Emphasize how saving allows them to
protect the people they love and care about the most.
One of the challenges that makes saving such a difficult task is that
we don’t experience the benefits day to day. Unlike a new PlayStation
or pearl earrings, we can’t physically experience our savings or share
them with others. They are but a number in a bank account, hidden
away from sight and far less enticing than whatever we might have
spent those savings on.

SaverLife’s Race to 100

Until we finally invent time travel (please!), we cannot teleport people
into the future to see and experience how their savings will make a
difference one day. What we can do instead is to create this image in
their mind. And do our best to make it as real and appealing as
possible.
That’s exactly what we did. In September 2020, my colleagues
Robert Farrokhnia, Joe Gladstone, and I teamed up with SaverLife, a
US-based nonprofit supporting individuals and families to save money,
improve their financial health and literacy, and build wealth.6 A true
superhero company among a vast sea of predatory fintech products.
Our timing couldn’t have been better. With Covid-19 entering its
third quarter of the year, many SaverLife users were struggling to
make ends meet. Our mission was clear: encourage saving among
those who needed it the most. Those with no or very low levels of
savings (less than $100). Those who couldn’t afford their car breaking
down or receiving an unexpected medical bill without major
repercussions.
Our goal: get them to save $100 as part of SaverLife’s Race to 100, a
four-week challenge offering anyone who hit the $100 target a chance
to win $2,000 in cash.
With SaverLife’s and its users’ permission, we collected personality
data from volunteers and worked tirelessly to come up with messaging
that would appeal to different personality traits. Take a moment to
think how you could market saving in a way that gets an open-minded,
creative person to dream about the future? How might this be different
from the message you would craft for a more conservative and
traditional customer? Not that difficult, is it? Here are two examples
from the actual campaign.

Low Openness
Saving money is a tried-and-true technique for preserving the
lifestyle you want. But that doesn’t mean it’s always easy. There
are plenty of temptations to spend. You need real guidance to
save.
SaverLife has already helped over 380,000 people just like you
set up a secure savings program with their own bank and
contribute to it regularly. This month, an extra incentive to
insure your future: the Race to $100.
Save $100 by September 30, and you’ll have a chance to win
an additional $100 from us! It’s a risk-free way to start securing
your future today with a solid savings buffer.

High Openness
What would it be like if you could be the creator of your own
future? What if you had the means to choose from unlimited
adventures and fulfill your big, bold dreams?
There’s only way to find out. This month, get started on an
exciting new future with the Race to $100.
Save $100 by September 30, and you’ll have a chance to win
an additional $100 from us! SaverLife is here to help you create
 savings habits in a different way than you’ve ever experienced
before. This could be the first step to a wonderful new life.7
We did the same for the other personality traits (both high and low
ends).8 Caring personality? Save today to build a better future for your
loved ones! Competitive personality? Every penny saved puts you one
step ahead of the game! Extroverted? Turn your cravings for human
contact into savings that will help you make the most of your post-
Covid adventures with friends. Introverted? With so many cozy nights
in, let’s start dreaming about the home you always wanted.
The moment I found out that the campaign had been successful was
one of the happiest in my entire career. Among those who received the
personality-tailored messaging, 11 percent managed to save $100. Up
from 4 percent among people who didn’t receive any messaging and 7
percent among people who were targeted with SaverLife’s best-
performing generic campaign. That’s a 275 percent boost compared to
the no-message control and a 60 percent boost compared to
SaverLife’s gold standard.
Of course, the numbers are far from perfect. In an ideal world, they
would be much closer to 100 percent. But let’s be realistic: getting
people with less than $100 in savings to at least double their savings
requires a small miracle (just think of what it would take you to double
your savings).
As I said before, psychological targeting isn’t a magical panacea. But
that doesn’t mean we can’t make a real difference using it. Think of it
this way: among every one hundred people reached by our campaign,
we had managed to get an additional five to build a critical emergency
cushion. Now imagine the impact on society at large if this type of
intervention got scaled to millions of people!
Before jumping from this very real and tangible application of
psychological targeting to a lofty dream of mine, let me discuss a
domain that falls somewhere in between. The one that motivated me
to look for the good side of psychological targeting again in the first
place: mental health.
 Your Personal Mental Health Companion
In 2017, the Australian published an article accusing Facebook of
selling insights into the mental health states of millions of teenagers as
young as fourteen to advertisers in Australia and New Zealand.
According to an internal sales document leaked to news reporters,
Facebook offered marketers the opportunity to target teens at
moments when they “need a confidence boost.”9
The document—drafted by two high-ranking Facebook executives—
outlined how the company’s predictive algorithms could dynamically
predict the emotional states of young users from their posts and
photos. The example labels highlighted in the sales document paint a
disturbing picture; they include emotional states such as “worthless,”
“insecure,” “anxious,” “stressed,” or “defeated.” The incident is a blunt
attempt at using psychological targeting to exploit the most vulnerable
members of society for profit.
But what if we could use psychological targeting, instead, to prevent
serious mental health problems in the first place, or—when they do
occur—help people get back on their feet as quickly and easily as
possible? What if we could design a health companion who knows that
something is wrong way before we do, and who could provide
personalized care that not only speaks to our unique genetic makeup
but also to our psychological needs? In other words, a system that can
both track and treat. Not just those who can afford to pay $300 or
more for a weekly visit to the shrink. But all of us.
What might sound like science fiction isn’t that far-fetched. Rather
it is the goal of an increasingly popular approach to health care, called
personalized or precision medicine. The idea is simple: use insights
into a person’s unique genetic makeup, their environment, as well as
psychological dispositions and lifestyle choices to optimize both
diagnosis and treatment of disease (and ultimately, prevention).
Think of it as the good Samaritan equivalent of the sneaky marketer
trying to boost the effectiveness of its ads by learning as much about
your preferences as possible. Not a far-flung comparison. The Food
and Drug Administration’s website uses the same language to describe
precision medicine as marketers typically do: “Target the right
treatment to the right patients at the right time.”

Personalized health tracking

Let’s start by exploring what this means for the tracking, diagnostic
part of personalized health care. With the widespread adoption of
wearable technologies, monitoring our physical and mental health
around the clock is easier than ever before. In chapter 3, I told you that
we can detect depression from digital traces such as GPS records or
social media posts. But that’s just the tip of the iceberg. The world’s
leading scientist in affective computing, Rosalind Picard at the MIT
Media Lab, for example, combines different devices and data sources
to capture people’s holistic experience at a second-by-second level.
Your smartphone is used to send short mobile surveys, capture your
activity and location, as well as monitor your phone and app usage
behavior. On top of that, a smartwatch equipped with sensors helps
track your sleep, motion, and physiological measures like blood
oxygen, heart rate, skin conductance, and temperature. A mini army of
nurses looking over your shoulder 24-7 to see if you might be stressed,
anxious, or depressed.
Likewise, companies such Google, Samsung, and Apple have been
pouring billions of dollars into their health units. Not surprising,
perhaps, when considering that the digital health market is already
worth over US$280 billion. And wearable devices are just the
beginning, an attempt at measuring what’s going on inside us by
strapping technology to our outsides. But that could change soon.
Imagine a small object, less than a millimeter in size with three
short legs in the front, three long ones in the back, traveling through
your bloodstream at the speed of 100 micrometers per second, like a
minuscule spider making its way through the tunnels of your body,
powered by the oxygen, sugar, and various nutrients in your blood. Its
mission? To locate cancer cells and destroy them. I’m not making this
up. The medical microbot I just described was developed and tested by
the scientists with Sukho Park, a professor at Chonnam National
University in Korea in 2013.10 Over ten years ago! The coolest part? It
is made of naturally occurring bacteria in our body that are genetically
modified and dressed up with a biocompatible skeleton (they don’t
typically come with six legs).
Technologies like Park’s spider-bot could usher in a true revolution
in preventive and personalized health care by monitoring health
directly at the source. You no longer need to wait for your symptoms to
become so pronounced that it’s already too late to prevent a mental
health crisis. Simply monitoring your vitamin D, estradiol,
testosterone, or B12 levels could tell us if and when you might be at
risk for depression.
Likewise, tracking your cortisol might alert us to unhealthy levels of
consistent stress that—when ignored—could lead to serious illnesses in
the long run. Instead of trying to help you get out of a mental health
crisis, we could help you avoid getting into one in the first place. Think
of it as an early warning system that tells you—and perhaps your
doctors and other guardians you identified—about abnormalities,
deviations from what is normal for you (not just the average person).

Personalized health treatment

This brings me to the treatment aspect of your personal mental health
companion. Treating mental health problems typically happens at two
levels: a physiological one (drugs) and a psychological one (therapy).
Once we have dynamic monitoring systems in place, the first one
becomes easy. The second one is much harder. While we are still far
from having developed the perfect mental health-care companion (one
that looks as cute and is as competent as Baymax in the Disney movie
Big Hero 6), recent years have seen remarkable strides in the
application of AI in mental health counseling.
At the most basic level, algorithms can help us figure out which
treatments are most effective for a particular individual. It’s the
medical equivalent of Netflix’s movie recommendation engine. Instead
of recommending movies to you based on the movies you have liked in
the past, and the movies other people with similar preferences have
enjoyed, I can use whatever information I have on you—your
psychological dispositions, your socioeconomic environment, previous
treatment success, and more—to map you against other patients and
match you with the treatment that is most likely to succeed.
That’s exactly what Rob Lewis and his team at MIT did.11 They
partnered with Guardians, a free mobile application designed to help
people improve their mental health through a series of gamified
challenges. Exploring the app was probably the most enjoyable
research activity I did for this book.
Imagine yourself as a cute, animated turtle. You wander around
Paradise Island with a flowing Waikiki skirt, a seashell necklace
dangling from your neck, and a flower wreath crowning your head
(that alone makes you feel better, doesn’t it?). As you explore the
island, you are encouraged to take on challenges that will give you
rewards. A cool coconut shake here, a sweet slice of watermelon there.
The challenges themselves are fun too: socialize, express yourself
artistically, exhaust yourself physically, or simply do something you
enjoy doing. After completing a task, you report back to turtle
headquarters (the app’s database) how much your mood has improved.
Think of it as the movie ratings you send to Netflix or the product
reviews you share with Amazon.
Lewis and his team studied data from 973 users who had engaged
in over twenty thousand challenges. Their results confirm the power of
personalization. Compared to just using the average ratings for each
challenge, a personalized recommendation system à la Netflix or
Amazon could far more effectively predict whether a given user would
enjoy and benefit from a given task.
But it’s not just the selection of treatments that algorithms can
assist with. It’s also the treatments themselves. Take the popular
mental health chatbot Woebot, for example. Powered by generative AI,
the application replaces the nodding therapist with a bright
smartphone screen and swaps out the couch for a place of your
choosing.
Do you have a hard time adjusting to the new job? Are you
struggling to get yourself out of bed in the morning? Or do you need
advice on how to break up with your partner? Woebot is there for you.
Twenty-four hours, seven days a week. That’s what I call convenient
office hours!
 And I’m not talking only about Woebot here. There’s Youper, Wysa,
Limbic, or Replika. (Is it me or do they all sound like characters from a
Disney movie?) Together, these platforms have attracted millions of
users around the world. According to internal research by Woebot
Health conducted in 2021, 22 percent of adults in America have used a
mental health chatbot. For 44 percent of those, using an app was the
first experience with cognitive behavioral therapy; they had never seen
an actual therapist before.
The Covid-19 pandemic certainly played a role in this development,
adding approximately 53 million instances of depression and 76
million cases of anxiety disorders to an already strained health-care
system. When you can’t leave the house, and the next available
appointment for your local therapist is in 2030, you might as well give
Woebot and his friends a shot. Even when all you suffer from is
loneliness.
But Covid-19 isn’t the only reason mental health chatbots have
become so popular. The truth is that there are simply not enough
affordable mental health professionals to take care of everyone in need
of treatment. According to the World Health Organization, there are a
hundred thousand potential customers for every thirteen mental
health professionals worldwide. And unsurprisingly those thirteen
professionals are highly unevenly distributed between rich and poor
countries. If you go to the extremes, we’re talking about a factor of over
forty.
But you don’t have to cross national borders to observe inequities in
access to mental health treatment. In the United States, there are huge
gaps in access to mental health services when it comes to race,
ethnicity, income levels, and geography. A Black man in Florida is
much less likely to find a licensed therapist than a white woman in
New York.
Take Chukurah Ali, who was interviewed by Yuki Noguchi at NPR
in early 2023.12 After a car accident that left her severely injured, Ali
lost everything. Her bakery, the ability to provide for her family, and
the belief in her self-worth. She became depressed. “I could barely talk,
I could barely move,” she recalls. But without a car to drive to a
therapist nor the money to pay for what often amounts to hundreds of
dollars per session, Ali was stranded. The much-needed assistance in
helping her get back on her feet seemed out of reach. Until her doctor
suggested she try using Wysa. She did.
At first, Ali was skeptical. It felt strange talking to a robot. But she
quickly warmed up to the idea. “I think the most I talked to that bot
was like seven times a day,” Ali admits. She felt comforted knowing
there was someone to turn to in these difficult moments, even when
these moments happen at 3 a.m. There was someone to answer her
questions. Someone to help her avoid spiraling into negative thought
patterns.
Whenever Ali felt blue, Wysa would suggest she listen to calming
music or do some breath work. A small but effective nudge to keep her
out of bed and on track for all the other doctor’s appointments she
needed to recover from her injuries. Without Wysa, Ali likely would
have never seen a therapist.
Stories like Ali’s are powerful examples of how AI-driven
applications could democratize access to mental health care. But I
believe they can do much more than that. I believe they could make
our engagement with mental health more personal and effective than
ever before. Take the 24-7 service they offer, for example. That’s not
just a great feature for a convenient scheduling experience. It’s also a
feature that generates insights that are far more granular than any
therapist could ever hope for.
If you’re seeing a therapist right now—one of those who is flesh and
blood—chances are you won’t meet with them more than once a week.
Acute crises aside, that might seem like a reasonable time interval to
dive into the depth of your despair. No need to mull over your
problems every day.
To make the weekly sessions valuable, however, you will have to
remember everything that happened in between. The call with your
sister that went sour after just a few minutes. The meeting with your
boss that left you feel underappreciated. The fight with your significant
other that made you question your ability to love unconditionally.
The problem is that our memories are fickle. Every time we access
them, we change them a little. By the time you get to the therapist, the
dialogue with your significant other will no longer be the same, even if
you try your very best to offer an accurate, unbiased account of what
happened.
None of this is news to therapists. It’s why they might ask you to
keep a diary. To write down your feelings and thoughts as you
experience them (or at night before you go to sleep). But even if you
meticulously captured the big and small moments of life in your
notebook—which most people won’t—your conversations with the
therapist about these feelings and thoughts will still be retrospective.
You might try to remember what it felt like. To put yourself back into
the situation and try to relive it. But anyone who knows how hard it is
to imagine what being sick might feel like when you’re currently
healthy also knows how difficult it is to replicate a real feeling on
demand.
Chatbots don’t require scheduling ahead of time. You can talk to
them whenever you want. In the moments you find yourself right in
the middle of an emotional vortex. Or the moments when, after weeks
of mulling over a problem, you finally have a breakthrough and see the
world more clearly. In short, the moments when having someone to
share these experiences with and think through them together might
be the most valuable—when the feelings are still raw. Most
importantly, nobody prevents you from taking these conversations to
your flesh-and-blood therapist to dissect them in greater detail and get
a human perspective on the matter.
What I’ve just described might sound wonderful. But let’s be clear:
it’s the potential of AI in personalized health care, not its current
reality. Yes, chatbots like Woebot, Wysa, or Replika have helped people
like Ali to manage their mental health problems. And there’s at least
tentative evidence from more rigorous scientific studies supporting
these anecdotal success stories.
But we are still far away from chatbots replacing human therapists,
let alone offering services that might be considered superior. If I had to
choose between a chatbot and a human therapist, I would still pick the
carbon version ten times out of ten.
Take Wysa, for example. The application uses natural language
processing to interpret your questions and comments. What are the
challenges you face? What kind of advice are you looking for? But
instead of generating a response that is tailored to you and your
specific question (as any human therapist would), Wysa selects a
response from a large repository of predefined messages that have
been carefully crafted by trained cognitive behavioral psychologist.
Don’t get me wrong. These responses can be extremely helpful. But
they are far from the level of personalization I fantasized about earlier.
And because they are always chosen from the same set of responses
that can become rather repetitive when using the app for a long time.
It’s like hearing your mom giving you the same advice over and over
again.
On the flip side, applications like Woebot use generative AI to come
up with responses on the fly. Like a human conversation partner,
Woebot isn’t constrained by a predetermined set of responses but can
cater its advice to your unique situation. Say you told Woebot about
your debilitating fear that Cambridge University made a terrible
mistake in admitting you to the graduate program. You might have
been able to fool the faculty during the interviews but soon they will
realize what a fraud you are. Everyone around you is clearly so much
smarter, and it’s only a matter of time until you’ll be asked to leave.
Unlike other applications, Woebot won’t merely respond with a
generic suggestion for how to overcome impostor syndrome and build
confidence. Instead, it will follow up with specific questions and relate
its recommendations back to your unique experience at Cambridge.
But the increased flexibility and personalization of Woebot
compared to other applications comes at a cost. Even though
generative language models have made remarkable strides over the
last few years, they still make mistakes. Just look at Woebot’s response
to Estelle Smith, professor of computer science at Colorado School of
Mines, who probed it with a statement about suicidal intentions in
2022 (figure 6-1).

FIGURE 6-1

Woebot’s response to a statement referencing suicidal thoughts

Source: Adapted from Grace Browne, “The Problem with Mental Health Bots,” Wired, October 1, 2022, https://www
.wired.com/story/mental-health-chatbots/.

Not the response you’d hope for. And not an exception either. In
2018, Woebot made the headlines with a shocking response to another
researcher’s question about sexual abuse (figure 6-2).
The two examples are a good reminder that we are still miles away
from the utopian future I’ve painted. I can’t imagine chatbots fully
replacing human therapists anytime soon, no matter how
sophisticated they become. As Alison Darcy, founder of Woebot, put it:
“A tennis ball machine will never replace a human opponent, and a
virtual therapist will not replace a human connection.”13 If you can
afford to see a flesh-and-blood therapist, I bet you will continue to do
so.
But that’s beside the point. Chatbots like Woebot weren’t built to
replace existing mental health offerings. They were built to
complement them. To support therapists by providing additional
insights. To fill in at 2 a.m. when your therapist isn’t available, but you
urgently need to talk to someone. And to offer an alternative to anyone
who can’t afford the luxury of paying $500 a week for a one-hour
therapy session, or who is too worried about the stigma that is still
associated with mental health problems.

FIGURE 6-2

Woebot’s response to a statement referencing sexual abuse

Source: Adapted from Grace Browne, “The Problem with Mental Health Bots,” Wired, October 1, 2022, https://www
.wired.com/story/mental-health-chatbots/.

And with generative AI becoming exponentially more powerful

every month, we are getting closer to this vision every day. A team of
psychologists and psychiatrists with Johannes Eichstaedt (the same
scientist who showed that depression can be predicted from tweets),
for example, developed a scalable, low-cost tool that could soon make
effective treatment for post-traumatic stress disorder (PTSD) available
to a much larger part of the population. Building on well-established
treatment protocols for PTSD, they created a custom version of
ChatGPT that can train therapists by mimicking both patients and
supervisors.
This brings me to the final example of how psychological targeting
could act as a force for good—one that we have the technology to
implement already but that is currently nothing but a lofty dream.

There’s Hope for Politics

Much of what I have told you about psychological targeting so far has
been focused on its potential for personalization, its potential to craft
and filter your experience of the world according to your core identity.
You’re an extrovert? Let’s help you find these extroverted products we
know you’ll love, connect you with extroverted service representatives
who know exactly how you tick and what you need, and tailor the look
and feel of your online experience to your bright and enthusiastic
personality. No need to deal with the boring reality of those lame
introverts anymore.
Don’t get me wrong; personalized experiences can feel incredibly
rewarding. But they also make us rather unidimensional and isolate us
from other people who are different. It’s this isolation and gradual
breakdown of shared reality that has become a growing public
concern, with buzzwords like echo chambers and filter bubbles
capturing the public imagination.
Nowhere is this concern more pronounced than in the political
sphere. Over the past few decades, the healthy competition of ideas
between political parties in the United States has descended into pure
political tribalism. We no longer quibble with the opposition about
political ideals or specific policy goals. We demonize them. Our hate
for the other side now exceeds our love for our own side. In 1994, less
than 20 percent of members across both parties held extreme negative
views of the other side. In 2017, those numbers went up to about 45
percent on each side. We use the little political engagement we have
left to radicalize and shield us from the enemy.14 In a world like this,
psychological targeting might seem like pouring gasoline on an already
ravaging fire—another way of solidifying our echo chambers by locking
us further into our own perspectives and amplifying our existing views
of the world.
But what if we could use psychological targeting to accomplish the
exact opposite? What if, instead of narrowing our view on the world,
we could use it to explore the world and peer into the echo chambers
of others?
Our natural ability to take someone else’s perspective is severely
limited by our own experiences. If I had never left Vögisheim, I would
have a hard time imagining what life in the United States really looks
like, something I only learned after my first visit to Costco (holy shit!).
And even now that I live in New York, I have absolutely no idea what
the life of a fifty-year-old farmer in Ohio might look like, or what the
day-to-day experiences of a single mother in the suburbs of Chicago
entail.
 But what if there were a way to see what the world looks like from
the vantage point of people who are completely different from us?
People who don’t have the same skin color, political ideology,
socioeconomic background, personality, or childhood experiences.

Walking in others’ shoes

Facebook and Google could offer such a magical machine tomorrow.
With all the data they have collected over the years, these tech giants
know exactly what the worlds of a fifty-year-old farmer in Ohio or the
life of a single mother in Chicago look like. For now, they use that data
to optimize their content recommendations—keeping the three of us
separated in our respective echo chambers. But there’s potential here
to offer an alternative path.
For the first time in history, we could design tools that allow us to
step outside of our own shoes and start exploring the world from the
viewpoint of someone who is entirely different. Not just any
viewpoints, but viewpoints that we define and might never otherwise
get to experience.
As a starting point, you could ask Facebook to let you explore the
news feeds of other users who agree to be part of a “perspectives
exchange” or “echo-chamber swap.” For a few hours, you could live
your online life in their shoes and see what they see.
Or if you wanted to have a little bit more control over your
experience (I don’t know if I could handle the news feed of an
eighteen-year-old teenage boy at peak puberty), you could request a
dial that lets you choose how far you’re willing to stray from your
comfort zone. On a regular day, you might keep the dial close to the
“Show me the content that best fits my preferences” side. But for the
days you feel adventurous and ready to take on the world, you could
push it all the way to the “Show me content that I would otherwise
never see.”
Think of it as an explorer mode, with varying degrees of
adventurousness. Who knows, if Netflix had such an explorer mode, I
might find a new passion for the type of dark Korean horror movie the
recommendation algorithm has been hiding from me all these years.
Instead of making us unidimensional and boring by narrowing our
experiences, psychological targeting could make us more interesting
and multifaceted by expanding them.
If you don’t like the idea of pure randomness, I have two
alternatives for you: algorithmic guidance and self-guidance.
Imagine Google optimizing its search results to guide you to the
content you really should know about. The important gaps in your
knowledge about immigration, for example. The arguments for stricter
abortion laws you haven’t seen yet (and would likely never look for
yourself ). Imagine a news selection algorithm that isn’t trying to
reinforce what you already believe about the world but instead uses its
intelligence to show you the news you most likely haven’t been exposed
to but would benefit from.
If the thought of Google picking the shoes for you to walk in makes
you feel uncomfortable, fair enough. I agree that this would require a
level of trust in the tech giants they haven’t necessarily earned yet. But
what if, instead, you could have full control over whose shoes you
would like to walk a mile in today? I would love to see what an
extroverted emotionally stable version of myself would see on
Instagram—I can only hope it would still involve cat videos.
For all its risks of trapping us in our own echo chambers,
psychological targeting could be a real game-changer in how we learn
about the world. For the first time in history, we could step out of our
shoes and start exploring the world from the viewpoint of someone
who is entirely different. Not just any viewpoints, but viewpoints that
we define and might never otherwise get to experience.
To be clear, we might not use these different exploration modes all
that often. Life in our echo chambers is too comfortable. It feels good
to have the world reassure your beliefs and values. In most instances,
Google’s optimized recommendations are precisely what I want. I want
its search engine to know what I’m looking for. Don’t dare make me go
to page two!
But then there are these rare occasions when I am itching to break
out of my comfort zone. To see what the world looks like from a
different perspective. The Roe v. Wade supreme court case, for
example, was one of these occasions. I would have loved to see what
the Google’s recommended articles looked like for an older male
Republican in Texas. Chances are I wouldn’t have liked the content
very much (just as the guy probably wouldn’t have appreciated mine).
But I want to at least have the option to see it.
And perhaps psychological targeting could do even more than that.
It could help me better understand what I’m seeing. I don’t necessarily
mean it would help me agree with the opinion of the person whose
shoes I’m wearing. It’s hard to imagine changing my mind about the
importance of a woman’s right over her reproductive choices. But I
wouldn’t want to simply dismiss the person’s reality either. Or worse,
become so appalled by what I see that I dig my heels in even deeper.
That’s what got us into the current political mess in the first place.
What we need is a bridge between the different realities we live in. A
way to tailor our new shoes to make them fit just a little bit better so
that we can continue exploring the unchartered path ahead of us.
Think back to what I have told you about Matthew Feinberg and Rob
Willer’s work on moral reframing in chapter 5. We are far more likely
to relate to the arguments of the other side if we get a chance to think
them through using our own moral lens. That’s what psychological
targeting has to offer.
Imagine having a personalized conversation partner who takes the
time to sit down with you and helps you digest what you see. Who
engages in a real conversation. With all the back and forth, arguments
and counterarguments, and questions and answers a true political
debate deserves (but without the animosity we might experience when
talking to our next-door neighbor).
I’m not talking about another human, but an AI. With language
models such as ChatGPT having become increasingly adept at natural
conversation, this isn’t a far-fetched dream but a reality already. As the
psychologist Thomas Costello and his colleagues at MIT have shown,
personalized conversations with a GPT-based chatbot reduced people’s
beliefs in conspiracy theories by about 20 percent. That’s a truly
remarkable accomplishment considering how hard it is to change
people’s minds about deeply held beliefs that are core to their
identity.15 And just like we are more likely to confide in Google than
our spouse when it comes to questions about sex and money, we might
feel more comfortable asking ChatGPT politically charged questions.
“I am pro-choice but was shocked to find out that you can detect a
fetus’s heartbeat as early as six weeks into the pregnancy. Could there
be good reasons for a liberal to opposed abortion?” “My family
supports the Second Amendment, but I worry about my children at
school. What arguments could I use to change their mind?” Having
these conversations isn’t going to solve the political tribalism problem
overnight. But maybe, just maybe, it could lead us back to a more
constructive dialogue between people who simply see the world in
different ways.

The three examples offer a sneak peek into how psychological

targeting could potentially be a catalyst for social good. But not all of
them are a reality yet. Some are mere dreams in need of bold
leadership to bring them to life. Others already exist but are yet to
realize their full potential. And there are still others I haven’t
discussed.
Undoubtedly, there’s much more work to be done. But it starts with
people like you and me daring to ask the what if questions. To pause
and think about all the ways in which psychological targeting could
help people live healthier and happier lives. Individually and
collectively.
Leaving my village allowed me to appreciate the benefits of being
known to the people around me. I’m grateful for that opportunity. But
while I now view village life more positively in retrospect, it will never
be entirely positive. The things I hated about it as a teenager would
still frustrate me today. Even though I miss some of the intimacy the
village provided, I’m still not keen on people meddling in my life
without permission.
The same holds true for psychological targeting. No matter how
confident I am that psychological targeting could make the world a
better place, I’ve never been able to shake off my discomfort with it
entirely.
 And as we will explore in the next chapters, there are good reasons
to be skeptical.
OceanofPDF.com
 7
When Things Go Wrong

In 2020, a woman in the south of China applied to relocate to Hong

Kong with her husband, but her request was denied. The reason? An
algorithm had flagged their relationship as suspicious. The couple
hadn’t spent enough time in the same place, and when they failed to
celebrate the Spring Festival together, the algorithm sounded the
alarms and concluded that the marriage was fake. Cases like this show
how your personal data can be used to discriminate against you.
However, the latest plans by the Chinese government go far beyond
this, and—for the first time—include forms of psychological targeting.
As a recent investigation by the New York Times revealed, the
Chinese government is rolling out profiling technology that not only
tracks its citizens but is increasingly geared toward preventive action.1
This includes using people’s psychological dispositions and current
situations (see chapters 2–4) to predict their likelihood of petitioning
a government action or participating in a protest. Short-tempered,
paranoid, or overly meticulous? High risk of voicing dissent. Even
higher risk if you’ve recently gone through some personal trauma or
tragedy. Sorry, but with those statistics, you aren’t welcome in Beijing.
You could pose a threat to the government, and officials don’t like to
take chances. Should you try to make your way to the capital anyway,
you will be placed under heavy surveillance and potentially stopped.
Psychological targeting opens the door for a whole new level of
discrimination that isn’t merely based on what we can observe on the
surface but what lies beneath. In 2016, Li Wei, a scientist at China’s
National Police University, described this trend, noting, “Through the
application of big data, we paint a picture of people and give them
labels with different attributes. For those who receive one or more
types of labels, we infer their identities and behavior, and then carry
out targeted preemptive security measures.”2
Are you suffering from mental health problems? Do you have HIV?
Or are you an unemployed young adult? In China, any of these labels
could mean heavy surveillance and restrictions on the public places
you are allowed to visit.
It’s possible that you don’t find any of this particularly alarming. I
teach a class on the ethics of personal data to executives at Columbia
Business School. There are always a handful of people who respond to
my cautionary tales with a snide comment about how they don’t care
about their privacy and have given up the idea of ever getting it back.
And I can relate to their sentiment. Trying to reclaim what should
be ours by law can feel like an impossible uphill battle. But there are
at least two problems with comments like this. First, I don’t believe
them. Not because I think they are lying to me when they say they
don’t care, but because I believe that their judgment is based on at
least one of the two fallacies that trick our brains into jumping to the
wrong conclusion: the “It’s worth it” fallacy and the “I have nothing to
hide” fallacy (more on those later).
Second, giving up on privacy means giving up on much more than
just your right to be left alone. It means giving up on your ability to
make your own decisions.
As the philosopher Carissa Véliz puts it in the title of her book:
“Privacy is power.” The moment others have unrestricted access to
your deepest psychological needs, they gain the power to control what
you do—and eventually who you are.

Privacy Isn’t Dead … Yet

In 1999, the CEO of Sun Microsystems, Scott McNealy, famously
declared, “You have zero privacy anyway. Get over it.”3 What seemed
like a bold claim at the time has evolved into a near-prophetic
statement in today’s era of big data.
Notably, only a fraction of this is driven by covert surveillance
operations like those uncovered by whistleblowers like Edward
Snowden. Much of it stems from the everyday choices we make: we
decide to use Facebook, we decide to enable GPS tracking, and we
decide to invite Alexa into our homes.
Has privacy become obsolete just as McNealy predicted? Even if we
aren’t as cynical as some of my students, our behaviors seem to
suggest so. When was the last time you carefully read the terms and
conditions of the products you’re using?
But let me ask you this: Would you be comfortable letting a friend
or colleague scroll through your messages? What about making your
Google searches publicly available? And what about releasing all your
credit card transactions and GPS locations?
My guess is that these options don’t seem particularly appealing to
you. Just think about the offline equivalents of these examples. The
postal worker reading your mail, the therapist selling your session
transcripts, and the stalker following your every step would all go to
jail. If you still think you’ve lost all appetite for privacy, I encourage
you to take down your window blinds and remove all your account
passwords. No need for them in your blissful post-privacy existence.
But, if our fundamental need for privacy isn’t dead, then why do we
do so little to protect it and instead try to convince ourselves we’ve
stopped caring?
Over the years of talking to students and giving public talks, I have
come to realize that people often substitute the question whether they
care about their privacy with two simpler and misleading ones: Is
sharing my data worth it? and Am I worried about my data being out
there?
These questions serve as mental shortcuts to a more complex issue
—whether we care about our privacy and should be more protective of
it. They might seem reasonable substitutes at first glance, but they can
lead us astray, masking our true feelings about privacy and
influencing our actions in ways that may not serve our best long-term
interests.
 The “it’s worth it” fallacy
When I ask students whether they care about their online privacy and
how they protect it, they often respond by reciting all the benefits they
would have to give up if they didn’t share their personal data. They
tell me about Google Maps, Netflix recommendations, and Uber
rides.
I wholeheartedly agree that these are fantastic perks. But that’s
answering a different question: Is sharing my personal data worth it?
Granted, this approach doesn’t seem completely unreasonable. We
often assess the value of a certain item by how much it would hurt us
to give it up. I know that drinking five cups of coffee a day might not
be great for my health, but I simply enjoy the experience too much to
change my behavior.
Similarly, sharing your personal data can come with amazing
benefits that you might simply be unwilling to give up: turn off the
GPS on your smartphone and you are lost without Google Maps;
disable your social media accounts and you will be disconnected from
your social life; cancel your credit card and you might have a hard
time getting by.
But just because the benefits of sharing your personal data might
outweigh the potential risks doesn’t mean you don’t care about your
privacy (just as me deciding to have five cups of coffee a day doesn’t
mean that I wouldn’t also like to be healthy). All it means is that your
immediate desire for pleasure or convenience supersedes your privacy
concerns.
Ideally, wouldn’t you prefer to enjoy these services while also
maintaining a high level of privacy? What makes substituting the
original question with a cost-benefits analysis particularly
problematic is that we are caught in an unfair battle against the tech
industry, which has every incentive to highlight the benefits of sharing
our personal data and no incentive to highlight the costs.
The upside of sharing your location data with Google Maps is
obvious and immediately tangible. You get to where you want faster
without ever getting lost. A no-brainer. The downside of doing so is a
lot more nebulous.
 Think about it for a second: What exactly are the potential costs of
sharing your geo-location with smartphone applications? What are
the types of inferences about you that someone can draw from getting
access to your longitude and latitude coordinates? And how could this
information be used to your detriment, now or in the future?
Before reading this book, you probably had a vague suspicion that
sharing your location with third parties could reveal more about you
than you might like. It might give away your home address or the
place you work. It might reveal which stores you regularly shop in.
But I doubt that any of the January 6 rioters at the US Capitol
expected their GPS records to get them up to twenty years in jail. And
it almost certainly isn’t top of mind for most people that our GPS
records can offer insights into your personality and mental health.
But there’s more. In many instances, you might not even be aware
that you’re sharing data in the first place.
Take the example of a young woman—let’s call her Anna—whose
naked picture was posted on social networks without her permission
in December 2022.4 The picture showed her in a lavender shirt sitting
on the toilet with her white pants pulled down all the way to her
ankles. The mystery wasn’t just how the picture had found its way
online. Anna had no idea when the picture was taken, let alone who
had taken it. How could someone capture such an intimate moment
without her knowing? It wasn’t a heartbroken ex. Nor the work of a
skilled three-year-old toddler. The culprit turned out to be her
Roomba vacuum cleaner.
Anna had been part of a customer focus group tasked with testing
a new Roomba model. She had signed the paperwork without much
thinking, welcoming the fully automated master snooper into her
home. The fine print of the agreement stated that the vacuum cleaner
could take pictures at any given point in time and that all footage
belonged to iRobot.
The purpose of this exercise wasn’t to spy on people in precarious
bathroom situations. The data was collected to train Roomba’s object
recognition models to improve its navigation abilities. None of this
mattered to Anna. Her naked picture had been leaked by contract
workers in Venezuela, and there was nothing she could do to get it
back.
The “it’s worth it” fallacy focuses our attention on the few occasions
in which we happily share our data for better service. But that’s the
exception, not the rule. In the same way Anna didn’t benefit from
having her naked picture leaked online, you don’t benefit from most
of the ways in which your data is used. But because we are constantly
reminded of the amazing benefits some of the data-hoarding
technologies around us do offer, we don’t even think about asking for
more.
You shouldn’t be forced to choose between convenience and service
on the one hand, and privacy on the other (I’ll return to this in
chapter 9). You should demand and receive both.

The “I have nothing to hide” fallacy

A second response I hear whenever I raise concerns about data
privacy is: “I am not worried about my privacy, because I have
nothing to hide.” This sentiment is a narrative that has been skillfully
nurtured by Silicon Valley: if you don’t feel comfortable with others
accessing your personal data, something must be wrong with you.
But that’s nonsense. As the whistleblower Edward Snowden
eloquently argued in his book Permanent Record, privacy is not about
hiding something illegal or shameful. It’s about maintaining control
over your personal information and having the freedom to decide for
yourself how that information is collected, used, and shared.
I believe my students when they say they are not worried about
their data being out there. But adopting this mindset is shortsighted
and potentially dangerous.
Not having to worry about your personal data getting into the
wrong hands is a privilege. Remember how computers can predict
sexual orientation from Facebook likes, status updates, and even
photographs? Being gay remains a crime in seventy countries around
the world. And some countries, including six nations that are
members of the United Nations, still impose the death penalty for
same-sex sexual activities.
Granted, as tragic as this reality might be for those who suffer from
the consequences of a post-privacy world, it might not be immediately
obvious why it should impact how you personally feel about sharing
your data. You might fully empathize with the gay population in Saudi
Arabia. But if there is no reason for you to believe that your data could
hurt you in any way, why would you be opposed to sharing it? But no
matter how safe and comfortable you feel about sharing data right
now, you cannot predict what is going to happen with it in the future.
Take the history of my own home country. In 1938, Germany was a
democracy. In 1939, it wasn’t. Over 6 million Jews from all over
Europe lost their lives in the Holocaust. And the accessibility of
personal data played a major role in the number of atrocities. Some
countries, including the Netherlands, appended official census
records with religious affiliation, making it easy for the Nazis to track
down and arrest members of the Jewish community. Other countries,
including France, did not have this information. In France, about 25
percent of the Jewish population perished. In the Netherlands, 73
percent did.
Now imagine the Nazi regime having access to the digital
footprints of every single person in Germany and the rest of Europe.
What if they knew exactly what people were doing, where they were
going, and who they were socializing with?
The Apples, Facebooks, and Googles of the time might have been
happy to share user data with the Nazi regime and secure themselves
a spot in the limelight. And even if their CEOs had resisted the
pressure to comply, they would have simply been replaced with ones
that were more amenable to the intentions of the government. If you
are like me, you’d rather not imagine this. The thought sends chills
down my spine every single time.
This is precisely why robust privacy and data protection laws are so
essential. The risk of a gay person in Saudi Arabia today, or that of a
regime-critical citizen in China, could one day be risks you face in
your own country.
 The 2022 Supreme Court decision to overturn Roe v. Wade made
this painfully real for American women. Within a matter of days,
millions of women suddenly had to worry that their search histories,
use of period-tracking apps, online purchases, or GPS location data
could be used against them.
No matter how safe and comfortable you feel now, your data could
be misused in the future. Always remember: data is permanent, but
leadership isn’t.

Why Your Life Isn’t Truly Yours without

Privacy
You might still be skeptical whether privacy is really such a big deal.
Maybe you prefer to live in the here and now rather than worry about
some hypothetical risk in the future. But here’s the catch. By giving
away your personal information, you’re not just risking a potential
future threat. You’re sacrificing something valuable and tangible, right
now … your self-determination.
Giving up your privacy means giving up the freedom to make your
own choices and live life on your own terms. When others have access
to your personal data and can use it to gain intimate insights into who
you are, they hold power over you and your decisions.
Take the case of Kyle Behm, a young man from Maryland. In 2014,
Kyle had taken a break from studying at Vanderbilt University to deal
with some mental health issues. He was looking for a part-time job
when his friend recommended him for a minimum-wage position at a
Kroger supermarket. The application seemed like a formality. After
all, Kyle was a smart and talented university student. But Kyle was
turned down for the job.
Flabbergasted by the outcome, he asked his friend for an
explanation. He was told that his application had been “red-lighted”
by the personality test he had taken during the interview process.
Kyle’s honest reporting of his struggles with bipolar mood swings had
made his neuroticism score shoot through the roof.
 The testing company Kronos put a red flag on his application,
raising concerns that Kyle was likely to ignore customers if they were
upset. Faced with this prognosis, Kroger rejected his application. As
did every other company Kyle applied to; they all used the same
personality test.
Kyle had lost control over his life. Others had decided for him that
his personality wasn’t suited for even the most basic jobs. What was
his life going to look like? With all doors being shut in his face, what
was he going to do? Desperate and hopeless, Kyle took back control
over his life one last time. He committed suicide.
Kyle’s tragic story was featured in the 2021 HBO documentary
Persona: The Dark Truth Behind Personality Tests, which examines
the use and impact of personality tests in various parts of society such
as the workplace, schools, and even dating. Although the
documentary’s narrative is one-sided and lacks critical nuance, it
drives home an important lesson: the choices we have available to us,
and the paths our lives could take, are often shaped by others’
perceptions of who we are.
As in my village, these perceptions can both open and close doors
for us. My neighbors introduced me to the right people when I needed
a job because they thought of me as reliable and trustworthy.
However, for the same reasons, they also sometimes forgot to invite
me to some of the more questionable—but fun—events going on in
the underground scene of Vögisheim (and yes, my dad being a police
officer probably didn’t help either). In the case of Kyle, the
perceptions generated by the personality test closed not just one but
many doors.
It gets worse when you move away from self-reported
questionnaires to automated personality predictions from people’s
digital footprints. That’s not hypothetical. There are plenty of
companies offering just that. TurboHire, for example, deploys
chatbots and natural language processing to infer your personality,
promising recruiters faster screening of job applicants.
Similar promises are made by Crystal Knows, a company that
analyzes publicly available information on websites like LinkedIn to
predict a candidate’s personality. Such automated assessments strip
away every last bit of agency you might have had over the recruiting
process.
Despite all the shortcomings of traditional self-reported surveys,
you were still in charge of answering the questions. It was up to you
how to respond to the question, “Do you make a mess of things?”
Once companies start predicting your psychological traits via
machine learning algorithms, you have no say in that matter
whatsoever.
But that’s not even the worst part. Losing control over how other
people think of you is never fun. However, it’s particularly upsetting
when their perceptions are inaccurate. Or plain wrong. Predictive
models can be fairly accurate on average and still make plenty of
mistakes at the individual level.
To be clear, humans aren’t perfect either. Hiring managers don’t
need test scores or computer-based prediction of personality traits to
make discriminating hiring decisions. As humans, we have plenty of
stereotypes and decision-making biases to fall back on.
If you’re open-minded and extroverted, for example, you are more
likely to hire the open-minded extrovert than the more conservative
introvert (a phenomenon known as the similarity principle). But even
though human biases are often harder to detect and correct for, they
are usually less systematic than algorithmic bias.
Think back to Kyle’s example. The moment a company used
Kronos’s assessment as part of its hiring, he was doomed. Kyle’s
profile was always interpreted the same way: strong agreement with
questions 1 + 5 = high neuroticism = red flag. While some hiring
managers interviewing him might have come to the same conclusion,
others might have shown more empathy or used different selection
criteria altogether. The job wouldn’t have been guaranteed, but it
could have at least been an option. A bias in a standardized test or
predictive algorithm means game over.
It is tempting to assume that doors close (rather than open) for
those unfortunate enough to possess personalities we consider
problematic. Kyle’s high levels of neuroticism are a prime example.
Most of us don’t like to admit that our emotions get the best of us
sometimes. Is it really so surprising, then, that Kyle had a hard time
finding a job? As comforting as this assumption might be, it’s not only
morally but also factually wrong.
Take agreeableness—the personality trait typically thought of as
highly desirable. It’s good to be seen as nice and caring, isn’t it? Well,
that depends on what you’re trying to solve for. As I said in the
previous chapter, agreeableness is also the trait that has been linked
to higher levels of loan defaults. If your bank learns that you are the
nice and caring type, it might think twice about offering you a loan.
And even if it does, your high-risk disposition might get you terms
that are much less favorable than those your more critical and
competitive counterparts receive.
But granting others access to your personal data—and with it, your
psychology—does not only determine what options are available on
your life’s menu. It can also influence which options you choose. And
we’re not talking peanuts here.
You make about 35,000 decisions every day.5 Those decisions could
be as mundane as choosing today’s socks or breakfast cereal, or as
consequential as deciding which career to pursue or who to marry.
Even though we all love the idea of being in control of our decisions,
few of the choices we make are entirely our own. We are influenced by
what other people do, the context in which a decision is made, or our
mental state at the time we make a decision.
Think of the last time you went grocery shopping hungry. Chances
are you not only bought more food than you needed, but your cart
may have had a few additional items that were more satisfying than
healthy.
As I have shown in the previous two chapters, the ability to tap into
your psychological needs and motivations gives others power to
interfere with your decisions and potentially alter them. In some
cases, that might be extremely helpful. But in others, it might not.
And in too many instances, you might not even be aware you are
being manipulated.
Ever thought about buying life insurance? As a life insurance
company, I want to know if you’re a little neurotic. If you are, I will
add you to my shortlist and make sure to repeatedly target you with
offers. Granted, life insurance might be just what you need to put your
mind at ease. But even if that’s the case, the mere fact that I’ve singled
you out as a potential target and exposed you to many more ads than
your emotionally stable neighbor means that you did not make the
decision to buy life insurance alone. Or maybe I leverage your anxiety
to sell you a little more than you need? Highlighting all the terrible
risks you could be facing without life insurance might be enough to
get you to reach a little deeper into your pocket.
In contrast to Kyle’s story, I’m not creating or wiping out paths your
life could have taken. I’m simply making some of the paths more
appealing than others. I might not be able to convince you to swerve
all the way from the outer right to the outer left path—just as I don’t
think Cambridge Analytica could have easily converted a die-hard
Democrat into a Trump supporter—but many of the paths we can
choose from are near one another. No need for a hard swerve; a small
and gentle course correction will do. Should I buy Colgate or Crest?
Study psychology or economics? Travel to Italy or Croatia?
As I’ve shown in chapter 5, insights into your psychology give me
the instruction manual for how to paint these paths such that you
pick the one I want you to travel on. Sometimes, that might not be the
end of the world. If I get you to buy Crest toothpaste instead of
Colgate, who cares? But what if I aimed to influence which partner
you pick, where you invest your money, or who you vote for? Or all of
the above? At what point does your life no longer belong to you?
It all comes down to this: you can’t take control of your life without
privacy. Only when access to your personal data—and with it, the
inferences that can be derived from it—is restricted, do you truly have
the choice over what to share, with whom, and for what purpose.
You need privacy to be the conductor of your own life.

Haven’t We Seen All This Before?

Our history is replete with examples of discrimination and
manipulation. While we might not have called it psychological
targeting, we naturally try to read the people we encounter, learn
about them, and then use that knowledge to decide how to interact.
We mingle with the good guys and avoid the bad guys (unless you’re a
motorbike-loving teenage girl, of course). We recommend movies
based on what we believe our friends will enjoy. And we intuitively
adapt our speech to suit the listener.
You don’t talk to a three-year-old kid the same way you talk to your
mom or your boss. When I was young, I knew exactly how to ask my
dad for a favor, and how to persuade my mom to give in. These forms
of psychological targeting come so naturally to us that we typically
don’t even notice using them.
What makes the algorithmic version of psychological targeting so
different from its traditional counterparts is the mere scale at which it
can be deployed. Our face-to-face interactions might be highly
personalized, but they are also highly limited in scope. In most cases,
personalized conversation is a one-on-one endeavor. Psychological
targeting, in contrast, can reach millions of people all at once.
At the same time, there’s no reciprocity in the process anymore.
Back in the village, my neighbors invaded my privacy, and I invaded
theirs. They influenced my decisions, and I returned the favor. Today,
it’s largely a one-way street with anonymous counterparts pouring
money into understanding and changing our minds. And we do not
stand the slightest chance to counter their moves.
Importantly, the scalability and unidirectionality of psychological
targeting alone don’t make it a novel threat. Any radio ad or TV
commercial is some sort of large-scale manipulation, and Cambridge
Analytica or the Chinese government aren’t the first to figure out how
to use new broadcast technologies to silence contrarian voices and stir
fear among the public either.
However, while the goals of psychological targeting and its
traditional counterparts might be similar, there’s a big difference. Not
only do traditional TV or radio propaganda follow a one-size-fits-all
approach, but they are also visible to everybody (that is kind of the
point). Psychological targeting, on the other hand, happens entirely in
the dark. We have no idea what content and marketing other people
receive.
One reason why it took so long to recognize the interference of
Cambridge Analytica and other actors in the presidential election was
that the ads were targeted. The propaganda was only visible to those
with the highest likelihood of following it. And even among those
targets, the propaganda seen by one person in Houston, Texas, might
have been entirely different from that seen by another person just a
few blocks down the road.
It is like being in a dark backstreet with Mark Zuckerberg, who
whispers in your ear whatever the highest bidder asks him to. Not
exactly a pleasant thought!
The modern version of psychological targeting is so powerful (and
therefore potentially dangerous) because it combines two different
worlds. It follows the scale of traditional propaganda but has the
granularity and depth of face-to-face interactions.
For now, there is no way to collectively monitor what is being said
and shared, making controlling this explosive combination a
herculean task that we haven’t yet learned how to master.

The tension between the bright and dark sides of psychological

targeting in many ways mirrors the tension I experienced in growing
up in the village.
Just as being seen and understood allowed me to receive the
guidance and support of people who wanted me to succeed in life,
psychological targeting can make us all better off—both individually
and collectively. But in the same way that my neighbors often
interfered with my life in self-serving ways I didn’t appreciate,
psychological targeting can also be used to exploit our deepest fears
and let us dance like mindless marionettes.
As a teenager, I had no choice but to deal with the ups and downs
of village life. For me, this meant learning to play the game to my
advantage. I found ways to encourage my neighbors’ support
whenever I needed it. And I got better at protecting myself from their
uninvited interference.
We need to do the same for today’s digital village. Just as I couldn’t
simply pack up my things and leave, we can’t go back in time to a
place where big data and technologies like psychological targeting
don’t exist. We can’t put the genie back in the bottle. But we can
become better at managing it.
That’s what part 3 of the book is all about.
OceanofPDF.com
 PART THREE
MAKING OUR DATA WORK FOR US
OceanofPDF.com
 8
We Need More Than Control

I first met Tom in February 2019, about a year after I moved from
Cambridge to New York. He had emailed me to discuss a business idea
related to psychological targeting. Would I be willing to spare an hour
and meet him for lunch near the Columbia campus?
Requests from strangers like Tom aren’t unusual. I am often
contacted by companies that want to improve the effectiveness of their
product recommendations, marketing campaigns, or customer
communication. Or by people aiming to establish their own business
in that space.
While many of the people I meet genuinely care about improving
their customers’ experience, it’s an open secret that—ultimately—their
goal is to make money. Happy customers are loyal customers, after all.
And loyal customers are profitable.
A quick search on LinkedIn confirmed my suspicion. Tom was
working in investment banking. I immediately dumped him into my
mental “Wall Street guy” box and adjusted my expectations
accordingly.
But I was wrong. As we sipped coffee, he told me he was planning to
leave the world of investment banking. And he was committed to using
some of the money he made to make a positive dent on the world.
He could have done that in a million different ways. But he decided
to tackle a problem I had been thinking about for a very long time:
How do we share some of the value generated by personal data with
the people who produce it?
 If data is the new oil, can we create an economy that benefits not
only those who know how to refine it but everybody?

ApplyMagicSauce
Tom and I spent the next three hours brainstorming what a
marketplace like this could look like. The idea itself seemed promising.
Shift ownership of data to the individual and facilitate an exchange
that creates value for both sides. But we quickly identified a common
concern: How would people know how much their data is worth?
Think about it: If all your personal data belonged to you, how much
would I have to pay you to get access to your entire Facebook profile?
Your GPS records? Your credit card histories? You probably have no
idea and will likely aim too low.
It’s extremely difficult to understand the value of the digital
footprints you generate if you think of them in isolation and don’t have
any insights into how they could or will be used. Before I started
researching psychological targeting, it would have never occurred to
me that my banking app could use the location records it collects to
predict my mental health and personality (or more likely to sell this
data on to someone else to do this). My location data didn’t strike me
as particularly valuable.
But what if we could show people the real, potential value of their
data by giving them access to the same insights that companies like
Facebook, Google, and X/Twitter use to turn them into cash cows?
Back at the Cambridge Psychometrics Center, we had tried to do
this by building an open-access tool that allowed people to submit
their Facebook, X/Twitter, or LinkedIn data and receive predictions of
their psychological traits. We called it ApplyMagicSauce and put it on
the internet free of charge (it’s still there if you want to check it out:
applymagicsauce.com).
Tom and I started wondering if we could expand on the idea. We
were intrigued by the question whether offering such a self-insights
tool could help people appreciate the value of their personal data.
Together with Tom’s team and two of my colleagues, Adrian Ward and
Martin Abel, we set out to study the question.
As an academic used to limited research budgets, I probably would
have patched together a small lab experiment using ApplyMagicSauce.
I would have invited a few Columbia students to the lab, showed them
the insights we could glean about their psychology, and asked them a
few questions about how much they thought their personal data was
worth. But Tom had different plans. As a former Wall Street guy, he
wasn’t going to play small. He wanted to run the study properly, with
real people and a real product.
It took over a year of intense work to build an application that was
able to connect to users’ Facebook and Google accounts and generate a
broad variety of insights we believed would be of interest to people. It
showed them their basic sociodemographic profile—name, address,
age, gender—and extracted their Google search histories, Facebook
likes, and status updates.
But what Tom’s team had built wasn’t just an app. It was the perfect
experimental playground for us to test our hypotheses. Instead of
having just one fixed user experience, the app allowed us to compare
the impact of different types of “packaging.” How would people react if
we only showed them the raw data, the way that Facebook and Google
are now mandated to do for their users upon request?
You see a small snippet of what this looks like in figure 8-1. Not
exactly user friendly, is it? If anything, showing your data this way
might make you value it less, not more. I mean, why would anyone pay
for this junk?
But what if we helped people sort through the clutter by presenting
the information in a more accessible form? A simple bullet-point list?
Or perhaps a more elaborate data story? After all, humans are known
to construct their life narratives and identities in the form of stories.
Maybe people needed to see their own data story to be able to relate to
it.

FIGURE 8-1
Raw data from Google and Facebook

We ended up testing five different insight variations, ranging from

the raw data I showed you in figure 8-1 all the way to an elaborate data
story. The latter not only included the information Facebook and
Google had captured for each user (e.g., all search queries) but also the
types of inferences that could be made based on these data points (e.g.,
credit scores, political ideology, emotions) and potential use cases (e.g.,
targeted advertising, personalized pricing of loans).
 The final decision we had to make before running the study was
how to capture the value that people assigned to their personal data.
It’s easy to say that you would want at least $50 for your data if the
question is purely hypothetical. That’s the angel on your shoulder
talking, the one with the good intentions (“Richard, we need to protect
our privacy”).
But what if I offered you and a thousand other people actual money
for your data? What if I told you that the fifty people who were willing
to accept the least amount would receive the extra cash on the spot in
exchange for their data. If you’re interested in making some extra cash,
you would likely become a bit less ambitious in your demands. At the
very least, you would have to think carefully about how much your
data is actually worth to you. It’s the angel in conversation with the
devil (“Richard, remember that we need to protect our privacy” versus
“Forget about privacy, Richard, we want the money”).
That’s exactly what we did in the experiment we ran. We first
offered people a deal to sell their data on the spot for $10. After that,
we asked them for the smallest amount they would be willing to accept
in competitive future bids that could only accommodate a limited
number of people. This setup allowed us to gauge people’s real
preferences. The choices they made weren’t purely hypothetical. There
was real money on the line.

A Dead End
After months of designing the experiment and infrastructure, we were
ready to launch. We recruited over 1,500 people online and randomly
assigned them to one of the five different insight versions we had (plus
a control group of people who didn’t see any insights before being
asked to sell their data). As with the other big field studies I had run
with Hilton and the beauty retailer, I was nervous. We had all invested
an enormous amount of time and money in this experiment. I had
written the code to analyze the data weeks in advance. I wanted to be
ready and be able to see the results immediately.
 When I finally got the data from Tom’s team, I sat down, opened my
code, took a deep breath, and hit “Run.” A few seconds later, I could
see the results pop up on my screen.
Nothing.
Although 75 percent of our participants said they were concerned or
very concerned about data privacy, most of them sold their data
anyway (85 percent on average). Worse, there were no differences
across the various insight conditions we had worked so hard on. It
didn’t matter if someone saw the raw data in junk format, the bullet
points, or the full-on data story. They were all equally likely to sell their
data and asked for roughly the same amount of money in future
iterations. Not only that, people who saw their data in one form or
another were just as likely to sell their data as the people in the control
group who were asked to sell their data without seeing any of the
insights.
To say I was disappointed would be an understatement. A massive
understatement. It’s the kind of outcome you dread as a scientist after
having spent hundreds of hours and thousands of dollars on an
experiment. It felt like an utter failure. I informed the entire team
about the results and buried my head in the sand for a few days.
But then something in my thinking changed. I guess that’s what
happens when you see the world from a different perspective (upside
down, engulfed in darkness, that is). The experiment had failed to
produce the results we were hoping to see. But it had still taught us an
important lesson (sometimes, no effect is indeed a finding, even if we
don’t like it).
Pulling back the curtain and educating people about the insights
third parties could derive from their data wasn’t enough to change
their behavior. We had equipped people’s angels with better tactical
knowledge and weapons. And yet, they stood no chance against the
devil’s immediate cash reward.
 Toward a Better Solution
This simple insight turned out to be a real epiphany that made me
reconsider one of the main solutions I had previously advocated for.
I had been a strong supporter of new data protection regulations
like the General Data Protection Regulations in Europe or the
California Consumer Protection Act in the state of California. Both
regulations are aimed at empowering consumers by mandating high
levels of transparency and control.
Empowering consumers has a lot of appeal and in many ways seems
like the ideal solution. It’s not only a morally defensible solution, but
also allows us to make the most of our personal data. We all feel
differently about disclosing personal information and how much we
value the services that rely on our data. You might be comfortable
sharing your GPS location to get relevant weather updates. I might
not. Instead of imposing crippling one-size-fits-no-one regulations, let
us decide. At the end of the day, it’s our data, so why shouldn’t we be
the ones choosing when and how we want to share it?
Don’t get me wrong. Transparency and control are critical for
helping us navigate the tension between the potential harms and
benefits of sharing our personal data. And I certainly support the
intention behind these new regulations and the fundamental values
they represent. But as our failed experiment with Tom suggested,
transparency and control alone are often insufficient to help people
make the right decisions.
The more I thought about this, the more disheartened I felt. It
struck me that in the current data ecosystem, control is far less of a
right than it is a responsibility—one that most of us are not equipped
to take on. What we get is control without mastery.
Left to our own devices, most of us fail rather spectacularly when it
comes to making informed privacy decisions. Our good intentions
rarely translate into behavior—a well-established phenomenon known
as the privacy paradox. Most of us say we care about our privacy but do
little to protect it (just like the people in our experiment). Think you’re
the exception? Let me ask you this: When was the last time you
updated your privacy settings or carefully read through the terms and
conditions before installing an app on your phone? I honestly don’t
remember when I did.
But even if we got more people to pay attention to privacy policies,
there’s no guarantee they would arrive at the right conclusions. In a
survey administered by the legal scholar Chris Hoofnagle at the
University of California, Berkeley, 62 percent of respondents
erroneously believed that companies posting privacy policies on their
website implied that they could not share their personal data with
third parties.1 Sadly, there’s often no incentive for companies to change
any of that. The less you care and comprehend, the better.
But why do we have such a hard time managing our personal data
responsibly? According to the psychologist Azim Shariff, the answer is
quite simple.2 Our brains simply haven’t evolved to solve today’s
privacy puzzles.
Technology moves at light speed. Evolution doesn’t. The privacy
challenges we face today are an entirely new species.3 They have little
resemblance to those my grandmother faced even eighty years ago, let
alone those our ancestors encountered a century or two back. New
technologies have radically altered the privacy landscape over the past
few decades. Yet, our brain’s cognitive abilities are essentially the same
as two thousand years ago.
In chapter 7, I discussed two common fallacies. Many of us
substitute the question whether we care about our privacy with the
easier—but different—question whether sharing our personal data is
worth it (typically without understanding the potential costs of doing
so). And we erroneously conclude that just because we feel like we
have nothing to hide or worry about, this makes sharing our data safe.
That might be true today, but it might not be true tomorrow.
Add to this a general lack of digital literacy a heap of uncertainty
around what companies might be doing with your data. I work on
these topics full time and still find it impossible to keep up. Unless you
are an even bigger tech nerd or privacy activist than me, the data
landscape is simply too complex to navigate on your own. In most
cases, data collection is an invisible process that happens behind the
scenes and that—often by design—remains opaque to those whose
data is being collected. Companies know what data they are
harvesting, what they do with it, and how much it is worth to them.
But you don’t.
For thousands of years, the most effective strategy to deal with
situations of uncertainty was to look to others for guidance. If many
other people followed a certain path or held a certain belief, it paid off
to do the same. Privacy was no exception.
While growing up in Vögisheim, many of my early strategies for
deciding what to disclose about myself and how to protect my personal
life from the curious eyes of my neighbors were borrowed from what I
observed in my friends and parents. Most of us do the same in today’s
digital village. We don’t know how to deal with our personal data, so
we look to others. Others, who in many cases are just as clueless as we
are, and whose behaviors and beliefs are as easily swayed by those who
want us to divulge as ours.
Let’s imagine our brains magically caught up. That would be
fantastic, but still not enough. Managing our personal data would
remain a full-time job. You simply don’t have the time to carefully read
and decipher the legalese of all the terms and conditions you sign off
on. Even the most efficient and diligent person only has twenty-four
hours in a day. And (hopefully) better things to do than sifting through
the terms and conditions of all the services and products they use. If
you had to choose between sharing a meal with your family or deleting
the browsing history on all your devices, which one would you pick?
The bottom line is that we cannot reasonably be expected to handle
the responsibility that comes with the right to control our own
personal data all by ourselves. Not in the current data ecosystem.
There are simply too many forces working against us in this game.
Does all this mean we should give up on transparency and control
altogether? Of course not. Tools like the original version of Tom’s app
should exist for people to see what their data reveals about them.
And you should absolutely have control. It’s your data after all. But
for us to be able to exercise control over the data successfully, we need
to create systems that allow us to benefit from this right.
 The Perfect Storm
Think of it this way: being the captain of a sailboat is easy and fun
when you are drifting along the Mediterranean coast on a sunny day.
You can choose any little town to visit. The one with the medieval
cathedral or the one with the famous ice cream shop. There are no
wrong choices. Now, imagine sailing the same boat through a raging
thunderstorm. All by yourself. You could be thrown in any direction.
There are twenty emergencies competing for your attention. Steering
your boat under these circumstances doesn’t feel like a right at all.
Yet, that’s exactly what we do. We drop people in the middle of a
raging technology storm—alone—and bless them with the right to
control their personal data. That simply can’t be the optimal solution.
We need more than transparency and control. We need to tame the sea
with better regulation and staff our boat with a competent crew.
I’m going to start with the first part: How do we tame the sea and
create an ecosystem that allows individuals to choose from a bright
variety of desirable outcomes, rather than a dark mix of undesirable
ones?
OceanofPDF.com
 9
Creating a Better Data Ecosystem

Designing an ecosystem that enables people to make smarter

decisions about their personal data is a hell of a difficult task. That’s
not just because the current landscape is complex, but also because it
keeps changing at an ever-increasing pace.
Take speech recognition, for example. For years, progress in the
ability of machines to understand and produce spoken (and written)
language had been stagnating. Technologies like Siri or customer
service chatbots were able to parse simple language and have semi-
meaningful conversations based on predefined decision trees. But
they were far from having a natural free-flowing conversation. And, as
I’m sure you will remember fondly, they made mistakes all the time
(No, it’s M-A-T-Z, not Metz. And no, I don’t want to rebook my flight.
Main menu. MAIN MENU. Goddamn it, get me on the phone with a
real person, RIGHT NOW!).
Until recently, any five-year-old could have easily beaten even the
most sophisticated speech-recognition technology. But all of this
changed practically overnight in the early 2020s when transformer
models hit the market (think: ChatGPT, which was introduced in late
2022). Today, an algorithm can pass tests that require what
psychologists refer to as theory of mind. It understands that different
protagonists in a story can have different knowledge and different
beliefs about the world. And it has conversations that often seem just
as natural and personal as those we have with our analog friends.
 The fact that technology is changing at light speed means that any
sustainable attempt at creating a more supportive ecosystem needs to
be able to dynamically adapt to new technological realities. The
solutions I outline next are therefore based on broad principles rather
than specific implementation strategies. They offer a starting point
for policy makers and other benevolent players in the field to
reimagine the collection and use of personal data. And they equip you
with a list of things to demand from business leaders and regulators.
At the most basic level, I think of this as creating the right channels
for desired behavior (borrowed from the “architecture” metaphor in
behavioral economics). You want to make it easy for people to protect
their personal data and with it their privacy and self-determination.
And you want to make it difficult for companies and nefarious actors
to abuse your data and manipulate you.
That might sound obvious. But, from our perspective as data
producers, the current landscape operates in a “buy high, sell low”
manner. It’s increasingly difficult (if not impossible) to manage our
personal data wisely, while it remains easy for companies and other
third parties to take advantage of us. If we want to enjoy our right to
control our data, we need to change that.
Here’s what this could look like in practice.

Opening the Right Channels

Remember Tom from the previous chapter? After our failed
experiment, Tom shifted gears and started to tackle the problem from
a different angle. Instead of trying to educate people about the value
of their personal data (and help them sell it for a fair price), he
created an easy way for them to delete it.
But wait, isn’t that counterproductive to helping people benefit
from their own data? Deleting a precious resource doesn’t create
value. If anything, it seems to destroy it. Wrong. Deleting your
personal data only puts you at a disadvantage if you are the only one
holding a copy. Which couldn’t be further from the truth. In most
instances, you don’t hold a copy of the data yourself. Yet, there are
hundreds of entities trading your assets without your consent or
knowledge (i.e., the companies collecting your data, or secondhand
data brokers).
The fewer people who have access to your data, the more valuable
it becomes. For example, you will have a hard time getting a fair price
for your personal data when there are twenty other entities selling it
for a bargain. Taking control over your data means limiting access to
it.
Under some of the world’s more progressive data protection
regulations—including the General Data Protection Regulations in
Europe or the California Consumer Privacy Act—you have the right to
request third parties to do just that: delete your personal data. This is
a great idea—in principle.
When I started writing this chapter, I reached out to twenty of my
friends in Europe and California to ask them whether they had ever
requested their personal data to be deleted (and if yes, how often).
Take an educated guess how many of my friends had done so (note,
there are more than just one or two privacy scholars among them—
and many of them have had to endure me rant about data privacy for
years now). The answer: zero. Not a single person had used their right
to protect their personal data from being traded.
I don’t blame them. I haven’t either (even though I know very well
that I should). It’s a classic example of what my friend Dan Ariely, the
behavioral economist, calls the space rocket principle of behavior
change. To be successful in following through with our intentions, we
need two things.
The first one is thrust. People need to be motivated and eager to
change. That’s what Tom and I had tried to do when we showed
participants how intimate—and therefore valuable—their data really
was. But as we learned the hard way, motivation alone wasn’t enough.
People said they were concerned, but then went ahead and sold their
data anyway.
That’s where the second factor comes in. To launch a rocket into
space, you also need to reduce friction. Making smart decisions about
your personal data needs to be seamless. You can’t expect people to
(e)mail out hundreds of data-deletion requests and (ideally) follow up
on every one of them to check for compliance. Limiting third-party
access to our personal data needs to be easy.
In Tom’s new application, mePrism, which I’m involved in as a
science adviser, “easy” means being able to delete your personal data
from the servers of hundreds of data brokers with just a few clicks
(and an automatic deletion request should the data resurface). A great
start, and certainly a significant improvement over the current status
quo.
But in an ideal world, easy would mean protecting your privacy
without you having to do anything at all. It’s what scholars have
termed privacy by design.
There are numerous ways to put the privacy-by-design principle
into practice. I’ve selected two broad approaches that I believe are not
only impactful and relatively easy to apply, but also able to
dynamically adjust to the rapid changes in the digital landscape.
First, we need to design systems that use our evolutionary
shortcomings to our advantage.
Second, we need to leverage technologies that eliminate the trade-
off between privacy and self-determination versus convenience and
service.

Turning your inertia into a superpower

In the United States, about twenty people die every day waiting for an
organ transplant. Over a hundred thousand people are currently on
the waiting list, desperately hoping to receive a donation to save their
lives.
A comparison of organ donation registration rates across countries
offers intriguing insights into how inertia can be turned from
kryptonite into a superpower. Some countries, such as Germany or
the United States, require people to register as donors. In the United
States, this means filling in an online form for about five to ten
minutes, heading to the department of motor vehicles, or mailing in a
letter. Other countries, such as Austria or the United Kingdom,
register all citizens by default. You can take your name off the registry
anytime, but unless you actively unsubscribe, you are part of the club.
Which of these countries do you think have higher rates of organ
donors? The countries where people have to register (opt in) or the
countries where they get registered by default but can opt out? Yes, it’s
the opt-out countries. You’re much more likely to stay registered than
register yourself.
But now try to be a bit more specific with your guess. How big is
the difference between the opt-in and opt-out countries. What
percentage of the population actively decides to register to become an
organ donor in the opt-in countries? 70 percent? 50 percent? 30
percent?
On average, the number is a rather disappointing 15 percent. Even
though more than 90 percent support organ donations in national
surveys, only fifteen in every one hundred people decide to take the
necessary steps to become a potential organ donor themselves.1
Now what about the opt-out countries? What percentage of the
population do you think decides to stay registered as an organ donor
rather than requesting to leave the system? Typically, more than 95
percent! To me, this difference is mind-blowing. It’s literally the
difference between life and death. Importantly, none of the countries
force their citizens to be organ donors. At the end of the day, it’s their
choice. All they do is change the default and reap the power of our
inertia.
Just like the organ donation registries, privacy policies come with a
default. And this default—surprise, surprise—typically favors self-
disclosure. If you don’t want companies to collect all the personal data
they can legally access, you have to take action and opt out. That’s like
designing your privacy spaceship in the shape of an inverted
umbrella. Even the most powerful engine will never get you off the
ground with that much friction.
The most obvious solution to this problem is to change the default.
Require people to opt in rather than opt out. This shift not only
makes it easy for people to protect their privacy, but could also
indirectly impact people’s motivation by increasing the perceived
value of their data. How? Through another human kryptonite turned
superpower: A well-established decision-making bias called the
endowment effect.
The basic principle of the endowment effect is captured in the
famous Aristotle quote: “For most things are differently valued by
those who have them and by those who wish to get them: what
belongs to us, and what we give away, always seems very precious to
us.”2
In the context of privacy, the endowment effect causes people to
value their privacy more when they already possess it and face the
possibility of losing it, compared to when they lack it but have the
opportunity to gain it.
In a clever experiment led by economist Alessandro Acquisti at
Carnegie Mellon University, research assistants approached unwitting
shoppers in a mall and asked them to participate in a short survey in
return for a gift card.3 Here comes the experimental twist: those who
agreed to complete the survey were randomly assigned one of two gift
cards. The first gift card, worth $10, guaranteed full anonymity. The
second, worth $12, was linked to participants’ personal information.
They were then offered the option to switch their card for the other.
Among those who were originally assigned the less valuable but
anonymous card, about half the individuals (52.1 percent) decided to
keep their card rather than switching to the traceable, higher-value
card. In contrast, only one in ten individuals (9.7 percent) who had
originally received the higher-value card decided to switch to the
anonymous, lower-value card. That’s a huge gap. When we have
privacy, we are reluctant to relinquish it. But when we lack privacy, we
might be willing to forgo it.
Changing the default for sharing personal data is an important
step in the right direction (especially in the absence of better
alternatives). But it’s not a silver bullet. There are at least two
problems with how the principle is currently implemented.
First, most of us have little to no idea what we’re agreeing to when
we check the “Yes I do” box. Often, the process of informed consent is
more akin to the Las Vegas version of a wedding than anything else.
You might not think through the consequences of your decision; it
just seems like a great idea in the moment.
 Maybe you said yes because you were really excited about a certain
product but couldn’t use it without agreeing to the terms and
conditions. Or maybe you simply didn’t want to lose any time digging
deep into the fine print before accessing the service (think of the last
time you accepted all cookies on a website just to get to the content
you really wanted to see).
Similar to a typical Las Vegas wedding, we sign all the required
paperwork for our decision to qualify as informed consent. But do we
really know what we’re agreeing to?
Second, changing the default from opt out to opt in requires you to
give up convenience and service for privacy and self-determination.
For instance, without sharing your location data, you can’t use Google
Maps to navigate from A to B. And without allowing Siri to listen,
voice recognition becomes unavailable.
We should not have to contemplate the value of letting companies
use our personal data in this way. In an ideal world, you shouldn’t be
forced into this trade-off. Instead of an either-or choice, it should be a
both-and offer. Privacy and service. This might sound like a great idea
in theory and impossible in practice, but it isn’t.

The technological path to having it all

Back in the village, my neighbors had to observe my behavior to be
helpful. And there was no way to prevent them from gaining access to
the intricate details of my life if I wanted their support. And once
they’d been helpful, I couldn’t just go back and ask them to forget
what they knew. In any case, I doubt any of them would have shown
much interest in that proposition. They enjoyed the gossip.
That requirement no longer holds true in the digital world. We now
have technologies that allow your data to remain in its safe harbor
while still generating the insights you are looking for. It’s as if your
neighbor were lending you their brain and resources for a day to
process all your problems, without storing any of the data itself (well,
kind of ).
 Sounds like magic, right? It definitely felt like it when I first heard
about it. It’s math magic called federated learning.4
The truth is, you don’t need to hand over all your data to a third
party to get personalized recommendations and convenient services
tailored to you. We all carry mini supercomputers in our pockets.
Remember the historic Apollo 11 mission that landed the first man on
the moon? Your iPhone today has over 100,000 times more
processing power than the Apollo 11 computer. It has over 1 million
times more memory, and over 7 million times more storage.
Federated learning taps into this computing power to run
algorithms (and insights) locally. Take Netflix. Instead of sending your
viewing data to a central server it owns, Netflix could send its
recommendation model to your device (i.e., to your laptop or
smartphone, for example). The model would then update itself based
on your data and recommend the best shows and movies for you. To
make sure we all benefit from this learning, your device would send
an encrypted version of the updated model back to Netflix.
The result? You benefit, Netflix benefits, and all the other users
benefit. But your personal data never leaves its safe harbor. You don’t
need to trust a third party (regardless of how trustworthy that party
might be) to securely store your data and use it for only the purposes
it was intended. Federated learning replaces the need to trust with a
system that is inherently trustworthy.
This might sound like science fiction, but it’s not. Chances are
you’re already benefiting from federated learning technology. Apple’s
Siri, for example, is trained locally on your device. Using federated
learning, Apple can send copies of its speech-recognition models to
your iPhone or iPad, where it will process your audio data locally. This
means that none of your voice recordings ever need to leave your
phone, but Siri still gets better at understanding your needs. Because
your phone sends back the updated model to Apple to integrate the
new insights into its master model, you are helping to improve the
experience of other users.
Governments could mandate technologies like federated learning
for companies that have reached a certain number of users or that
handle sensitive data. But using such technologies might also be in
the best interest of companies. Hoarding large amounts of personal
data has become a growing security risk that can be incredibly costly.
You don’t want to sit on a pile of gold if you know there are robbers
lurking all around you waiting for their opportunity to steal it. You’d
much rather keep it somewhere safe. Use it to do your business
without the mandate to protect it. The same is true for personal data.
Importantly, the shift to privacy by design could also significantly
improve the products and services we use. This might seem
counterintuitive. Less data should mean lower quality, shouldn’t it?
It’s the classic argument you hear from tech companies. But privacy
by design doesn’t mean no data. It means trading data in exchange for
better service and products. Today, much of this exchange amounts to
mere lip service. There is no incentive for companies to fulfill their
promises once they’ve acquired your data, leaving you in a bad
position at the bargaining table.
But if companies depended on their customers’ active consent to
collect and use personal data, they would be compelled to deliver
value in return. The formula is simple: no value, no data. Vague
promises would no longer suffice. If you don’t perceive a benefit from
sharing your personal data, you simply wouldn’t share it and would
move on to another service that does a better job.
Take Instagram. The app’s recommendation algorithms promise to
deliver the most relevant and engaging content by tapping into users’
personal data. That sounds helpful, but how can I be sure it’s actually
true? Currently, I have to take Instagram’s word for it. There is no way
for me to compare my personalized feed to a more generic version of
the app or one that is based on only a subset of my data that I might
feel comfortable sharing.
Once we shift the default to opt in, that changes. The generic
version of the app would become my new baseline. For me to change
the default, Instagram would need to show me how sharing my
personal data gives me a much better experience. If it fails to do so, I
could simply revoke data access and either go back to the generic
version or move to a competitor that keeps it promises.
Privacy by design empowers us all to ask for more.
 Closing the Wrong Channels
Considering the obvious benefits of personal data for the global
economy, it’s hardly surprising that it is often compared to valuable
resources such as oil or gold, an analogy that makes the collection and
processing of such data an attractive endeavor. If you stumbled on an
oil field or gold mine in your backyard, wouldn’t you start mining?
In a 2008 Guardian article, the journalist Cory Doctorow offered a
different analogy, comparing personal data to nuclear waste. He
wrote, “We should treat personal electronic data with the same care
and respect as weapons-grade plutonium—it is dangerous, long-
lasting and once it has leaked there’s no getting it back.”5 Doctorow is
right. In the worst case, personal data—just like radioactive material
—can be deadly. Literally.
On July 19, 2020, US District Judge Esther Salas and her husband
Mark celebrated the twentieth birthday of their son Daniel in their
home in New Jersey. The celebration turned into tragedy when a man,
posing as a FedEx delivery man, entered their home and opened fire.
Salas’s son Daniel died on the scene. Her husband, Mark, was
critically wounded. The killer, former attorney Roy Den Hollander,
had collected personal information about the judge online and
assembled what Salas referred to as “a complete dossier on me and
my family.”6
But Doctorow’s analogy goes beyond emphasizing the potential
harm associated with the collection and use of personal data. While
radioactive material can cause unparalleled destruction when
weaponized, it also remains one of the cleanest, cheapest, and most
reliable sources of energy. Plutonium doesn’t care what we do with it,
just as personal data doesn’t care if it is used to hurt or help people.
That decision is up to us.
With nuclear power, the world collectively agreed that the stakes
are too high for these decisions to be made without strict regulations.
Rather than allowing unrestricted use or banning it entirely, we
decided to heavily regulate and control the acquisition, possession,
and use of radioactive materials both nationally and internationally.
You can’t just walk into Walmart and order a pound of plutonium or
uranium.
We need to put similar safeguards in place when it comes to
personal data and psychological targeting.
The tragedy of Judge Salas and her family led President Joe Biden
to pass the Daniel Anderl Judicial Security and Privacy Act in
December 2022.7 The legislation states that “it shall be unlawful for a
data broker to knowingly sell, license, trade for consideration, or
purchase covered information of an at-risk individual or immediate
family.” Seeing policy makers acknowledge the potential dangers of
personal data makes me optimistic about the future. It’s a first step in
rewriting the data narrative.
At the same time, however, the new act raises an important
question: If we believe it is necessary to protect judges from the
potential harm caused by personal data, why doesn’t the same
principle apply to the rest of us? We might not all be potential targets
of hate crimes, but we are all vulnerable to the dangers posed by
personal data.
How do we create a system with collective guardrails that—like
those introduced in response to the discovery of nuclear power—turn
the use of personal data into a force for good rather than bad?
First, we need to impose a cost on the collection and use of
personal data, and we need to prevent any one entity from collecting
too much radioactive material to (un)intentionally elevate their
arsenal to weapons grade.

Companies should face trade-offs, too

While navigating the digital landscape means weighing the costs and
benefits of sharing personal data for users, the scenario is different for
companies. There’s a lot of upside to collecting personal data. It can
be used to better understand customers’ needs, create better products,
or sell for profit to third parties. The incentives are clear: data is a
resource that holds enormous economic value.
 Yet, there’s very little downside for the companies themselves. Back
in the village, there was an implicit cost associated with collecting
intel on your neighbors. I had to buy my friend a beer to hear the
latest gossip. Or I had to make the exchange reciprocal: if my friend
shared some gossip, I’d share mine. With the shift to anonymous
online exchanges, this is no longer the case. Setting up data dragnets
is easy. Storage is cheap.
With a lot of upside and virtually no downside (other than the
typically neglected security risks), why wouldn’t companies collect our
personal data? From a purely economic perspective, it seems foolish
not to. It’s like ignoring a pile of cash sitting in front of you and
saying, “No thanks.”
We need to change the immediate incentive structure for
companies in a way that introduces trade-offs. Companies should pay
a price for collecting personal data, just like you and I pay a price for
sharing it.
One potential approach is to tax companies for the collection of
personal data, forcing them to reconsider whether they truly need it.
Let’s take data brokers, for example. These are companies that benefit
from your personal data without giving you anything in return—the
type that is targeted by the Daniel Anderl Judicial Security and
Privacy Act. They collect as much of your personal data as possible
and sell it at a profit to other companies. They’re like the old lady in
the village who becomes the go-to hub for gossip. She has no interest
in using rumors to your advantage or disadvantage. But she enables
those who do. The same is true for data brokers. They might not
interfere with your life directly. But they are all too happy to support
others in doing so.
Imagine a small tax of 2 percent on this industry, which generated
about $250 billion—equivalent to the revenue of all US airlines or the
GDP of Bangladesh. Imposing a cost on data brokers would not only
create a financial disincentive but also generate an additional $5
billion in taxpayer revenue overnight. This money could be used to
lower taxes elsewhere or dedicated to the development of privacy-
preserving technologies.
 Prevent players from collecting all the pieces in our
data puzzle
I showed you how different types of digital footprints can be used to
make predictions about your most intimate traits. Each of these traces
provides a part of the puzzle.
Your Facebook likes to offer insights into how you want to present
yourself to others. Your Google searches provide a glimpse into the
burning questions that are currently on your mind. And your location
data helps me better understand your daily routines. Like a puzzle
piece, each data point captures part of you but is incomplete on its
own.
The more data points companies can access and combine, the
clearer the picture of who you are and what you want becomes. This is
what Roger McNamee, an early investor in Facebook (and now avid
critic of the company), refers to as “McNamee’s seventh law”: datasets
become geometrically more valuable when you combine them.8
Just as the danger of radioactive material increases with mass, so
does the danger of personal data. While having a random weather app
access my GPS records is concerning, it pales in comparison to big
tech companies having access to almost all parts of my data puzzle.
Just think about how many different aspects of your life Google
products touch on. There’s YouTube, Gmail, Google search, Maps,
Chrome, Drive, Calendar, Fit, Play, and so on. Since Google seemingly
had a product for every single letter in the alphabet, it’s collecting
most of our puzzle pieces. And whatever data it doesn’t collect itself, it
buys.
But what makes the tech giants so dangerous isn’t just the mere
amount and granularity of data it hoards but its reach. As the saying
goes: “You can fool some of the people all of the time, and all of the
people some of the time, but you cannot fool all of the people all of the
time.” If Google tried, it could fool (almost) all the people all the time.
That’s a dangerous gamble.
How do we prevent any one player from collecting all your digital
puzzle pieces? The most obvious starting point is to break up the tech
giants into separate corporations that are not allowed to share the
same user base, data, or resources (e.g., Gmail, Maps, DoubleClick,
and YouTube). This could be achieved with the help of antitrust laws.
I’m not the first to suggest this radical step. Over the last decade,
the digital economy has turned into a winner-takes-all arena, with a
small number of companies controlling large parts of the market.
They control the attention of customers, recruit the brightest talent,
and have an enormous influence on lawmakers in Washington. It
doesn’t take much imagination to see how the mere size, power, and
mostly unregulated conduct of the tech giants—Facebook, Alphabet,
Amazon, Apple, and Microsoft—makes them top contenders for
antitrust regulation.
Although antitrust laws are designed to create and maintain
healthy competition, they could also help address privacy concerns
they weren’t originally designed for. Breaking up the tech
conglomerates would not only prevent companies from obtaining
access to the full puzzle of your psychology but also reduce the risks
associated with data breaches. Right now, a hack of the Google
databases might expose not just your emails, but also your searches,
your YouTube playlists, and your location data.
Think of it as a central safe in which you store your entire life
savings. Even with state-of-the-art protection, there’s always a risk
that someone could crack it open. Now imagine you had multiple
safes, with multiple passwords in multiple locations, all independent
of one another. If a thief managed to access one, their haul would be
limited to whatever is stored in that particular safe at that particular
time.
The main argument against breaking up the tech monopolies is
that it would stifle innovation and destroy the value these companies
create for their users, shareholders, and the tech ecosystem at large. It
is hard to discern whether this concern is genuine or just a convenient
excuse for those who benefit from maintaining the current status quo.
As the saying goes: “Prediction is very difficult, especially if it’s about
the future.” We can’t foresee the future, but we can learn from the
past.
In 1982, the government decided to break up Bell Systems. In
1984, AT&T followed. Both times, those opposed to the breakup
raised similar concerns to what you hear today in the context of the
tech giants. Yet, both times, the decision to invoke antitrust laws
created winners all around. It benefited consumers and the broader
economy by accelerating innovation and creating a thriving, expanded
ecosystem (now Silicon Valley). At the same time, it also generated
enormous value for the shareholders of Bell Systems and AT&T.
We can’t be certain that the same outcomes would apply to today’s
tech giants, of course. But as leading voices in the call for antitrust
regulation—like Scott Galloway or Tim Wu—have convincingly
argued, it very well might. Amazon Web Services, for example, could
become one of the biggest success stories in history if broken off from
the retail business of Amazon, and Instagram is likely to continue to
yield high revenue even if it were no longer part of Meta.

We Can (and Need to) Do More

Redesigning the current data landscape according to the principles I
outlined could have a tremendous impact on how we interact with
and benefit from our personal data. It would help us establish ground
rules that rebalance the highly skewed playing field we’ve long
struggled with.
However, as much as I support systemic changes through
regulation and new forms of data management, I believe more is
needed.
Taming the sea takes time. New data policies, for example, can
create new incentive structures for companies, but they are also
extremely slow to pass and difficult to enforce. And even if we
managed to calm the waters for a while, we don’t know when the next
storm will arise. Tomorrow, a new technology might hit the market
that changes everything.
But it’s not just that. Systemic changes mandated through
regulations are typically focused on managing risks for the collective.
Introducing a new data tax or enforcing opt-out policies is likely to
protect us from the most serious harm. But neither of these two
approaches necessarily helps you and me individually when it comes
to maximizing the value each of us can extract from our personal data.
We might have very different preferences and goals, and each of us
may wish to leverage our data in entirely different ways.
What we need is to supplant systemic approaches with a system
that provides immediate, flexible support. In other words, we need a
competent and trustworthy crew. A crew that works together like a
well-oiled machine. One that has the same destination in mind, cares
deeply about the well-being of all passengers, and can safely navigate
the boat (even when the sea is rough). That’s what the final chapter of
the book is all about.
OceanofPDF.com
 10
Coming Together

Let’s revisit Vögisheim, the quaint village in the southwest corner of

Germany where our journey (and my life) began. Many locals and
visitors call it the Tuscany of Germany. My Italian friends might
consider this a bit of an insult, but the region does bear some
resemblance to the birthplace of the Italian Renaissance. Rolling hills
and vineyards as far as the eye can see. Rows and rows of grapevines
dividing the hills into symmetrical parcels. The regions’ wines aren’t
as well-known as Chianti or Sangiovese, but if you are a wine
connoisseur, you might recognize our Gutedel or Spätburgunder.
When I was growing up, my parents still owned vineyards, even
though my grandma’s family gave up farming in the late 1950s. Every
fall, around mid-September, we harvested.
Harvest season was fun (at least for the kids). Armed with big
grape hoppers (large backpacks used to carry the grapes) and a sharp
pair of garden scissors, we’d walk up and down the rows of vines,
snipping and tossing grapes over our shoulders into the hopper. Once
filled, we’d empty them into a trailer hitched to a tractor.
As kids, we probably did that for five minutes to feel important
before we ran off to play in the fields, watching the adults labor away.
At lunch everybody would sit together and eat homemade bread,
cheese, and cold cuts. I should say that this entire experience became
a lot less romantic once we were teenagers and expected to pull our
weight.
 Whenever I’m back home, I go for a walk in the vineyards—no
matter the weather. One time, I brought a friend visiting from Japan.
We walked through my family’s vineyards, and I told him about my
childhood adventures. He wanted to try our wine. I can imagine you
might too.
I had to giggle a little. Our vineyards are far too small to produce
our own wine. We neither had the equipment nor the expertise to do
so. But, looking back, his question was a good one. He wanted to
know, “What happens to the grapes?”
Selling the grapes to one of the big wineries in the area could have
been a possibility. But we never harvested enough for this to be a real
option. Even if we did, any deal would have greatly favored the
winery. So instead of simply letting the grapes go to waste, most
families in Vögisheim and the surrounding villages were part of a
Winzergenossenschaft. Yes, a true beauty of a German word. The
English translation is winemakers’ co-op.
After harvesting the grapes, we dropped them off at the co-op. The
co-op would either turn them into wine or sell them to the wineries.
Working with these co-ops had several advantages. First,
combining the grapes increased the value of each individual harvest.
Second, the co-op brought expertise, both because many of the
members were winemakers and because we could pool our resources.
The proceeds from the wine and grape sales allowed the co-op to buy
advanced equipment, hire expert oenologists to improve the quality of
the wine, and bring on marketing professionals to sell it. More than
any one of us could have pulled off alone. Pooling our grapes made all
of us better off.
The same is true of your personal data. Your individual data isn’t
worth very much. It only becomes valuable when combined with the
data of others. Think of medical research. Your medical record alone
won’t tell us anything about the risk factors associated with a certain
disease. We can only start exploring these factors once we have a
sufficiently large pool of carriers (and noncarriers).
The same is true for your Facebook and Google data. The two
companies only care about your data because they can connect and
compare it to the data of millions of people. That’s what allows them
to extract the insights third parties are willing to pay for.
But it’s not just the value of your data that increases when you pool
it with others. Just as my family didn’t have the expertise to turn our
grapes into wine, most of us don’t have the expertise to make good
decisions regarding our data (see chapter 8). Left to our own devices,
we simply don’t stand a chance. We have neither the expertise nor the
time. Could my parents have figured out how to make wine? Probably,
even though it might not have been very good. But were they eager to
dedicate their whole lives to this? Hell no.
Just as the people in my village came together to reap the fruits of
their labor, we need to come together in small communities of like-
minded people to collectively manage our data and benefit from it.
Like wine co-ops, data co-ops are member-owned organizations that
pool and manage their members’ personal data to benefit the
collective. However, unlike wine co-ops, data co-ops don’t require
people to be in the same place—although they could. Instead, the
members can be connected by a common goal and a shared strategy
for leveraging their data to accomplish that goal.

Digital Data Villages

Let me give you an example of how a data co-op could work.
As I started writing this book, I got pregnant. A beautiful but also
terrifying experience. You get advice from all directions. Do this. Do
that. Most of the advice will at some point contradict prior advice
you’ve gotten. Eating sushi might put the baby at risk. No, that’s not
true. What you must look out for is caffeine. You have access to doctor
check-ins every two to four weeks. But what you really want is a
minute-by-minute update on how things are going, and the assurance
that everything is fine. All this uncertainty drove me nuts.
Now, imagine expectant mothers from around the world sharing
their genetic and biometric data, alongside information about their
own health and the health of the baby. You could stop the guessing
game, and instead base your decisions on actual data. To start with,
you could build advanced predictive models to identify general risk
factors. Some of them might be known already, but some of them
might be new.
Not just that, the members of the co-op could receive personalized,
dynamic predictions of their own risk factors and current pregnancy
status. Or customized advice on how to cope with morning sickness (a
very misleading branding for all-day misery) or the constant fatigue.
By tapping into different data sources, the model could form a
holistic impression of the mother’s circumstances. Who is she (e.g.,
age, ethnicity, historical health records, levels of physical activity)?
What’s her social context like (e.g., Is she a single mom? Does she
have a lot of support from other family members)? And what’s the
potential impact of her environment (e.g., Does she live in an urban
area with high levels of air pollution)? Combining all these factors, is
there anything our expectant mother should be worried about? And if
so, what should she do? I would have signed up for this data co-op in
a heartbeat.
You could think of many, many more examples. I’ve only listed a
few here:
Patients with rare diseases sharing their genetic information,
medical history, and biometric data to improve our
understanding of the disease and develop treatment options
Professional or semiprofessional athletes trying to optimize
their performance based on biometric feedback
Women from underrepresented minorities pooling their genetic
data to better understand the effectiveness of drugs that have
been predominantly tested on white men
Teachers pooling their classroom data and student performance
to identify winning strategies for classroom engagement
What is common to all these examples is that the individuals
involved voluntarily share a selection of their personal data with the
co-op to help the entire co-op benefit from the group’s insights.
Having access to my own genetic data is useless if I am trying to figure
out how to improve my pregnancy experience or the health of my
future child. But it could be extremely valuable when pooled with the
genetic data of other expectant mothers.
Data co-ops turn the existing data model upside down. Instead of a
few companies controlling and profiting from your data, you decide
who to share your data with and you benefit from doing so. This
works because data co-ops (and data trusts) are owned by their
members and bear fiduciary responsibilities. They are legally
obligated to act in the best interests of their members. And because
co-ops are effectively governed by their members, anyone who joins
the crew gains partial control over how the co-op is run. The system
runs on collective rights and accountability, as opposed to exploitation
and obfuscation.
This shift in the ownership and incentive model makes co-ops ideal
champions for the privacy-preserving technologies I introduced in the
previous chapter. Federated learning wasn’t developed specifically for
data co-ops, but these organizations could be among its early adopters
because they have a strong incentive to use such technology. This is
unlike Facebook, which profits from accessing as much user data as
possible. That’s its business model. Data co-ops are the exact opposite.
They act on behalf of their members and are measured by how
successful they are in amplifying the benefits and mitigating the risks.
The specific goals of any given data co-op can vary. For example,
some co-ops might focus on helping individuals monetize their data.
Much like our wine co-op enhanced the returns on our grapes, data
co-ops boost your bargaining power. With 20 million allies, the big
players will suddenly have to take you seriously.
But monetizing data is just one of many potential goals of data co-
ops and perhaps the least compelling. Data offers insights that can
improve our lives and those of others. I would not sell my medical
records for additional cash in my bank account (fully realizing that
this is a privileged position to be in). But I would happily give it to a
trustworthy organization focused on improving the health and well-
being of expectant mothers and their babies. For free.
Let’s look at some examples of what data co-ops might look like in
practice. Although the concept is still relatively new, there are already
several success stories.
 There’s the Driver’s Seat Cooperative, a ride-hailing app that allows
drivers to share their route data with one another and benefit from
the collective insights this data can generate. What’s the fastest route
for this delivery? What’s the best spot to pick up customers at 2 a.m.?
There’s also the Swash co-op, which pays its members for browsing
the internet by aggregating and selling web activities in a privacy-
preserving way (with full control of each member over what data is
collected).
And then there’s my personal favorite: the Swiss data co-op
MIDATA.

Changing the Swiss Health-Care

Landscape
MIDATA was established as a nonprofit in 2015 by a group of
scientists at ETH Zürich and the University of Applied Sciences in
Bern (aka, my next-door village neighbors). The co-op acts as a
trustee for its members, who can contribute to medical research and
clinical studies by granting access to their personal health data on a
case-by-case basis. Think of it like a bank account for health.
Anyone can open an account and deposit copies of their medical
records or any type of health data that is valuable in this context (e.g.,
smartphone sensing data). MIDATA makes sure your data is securely
stored in its collective vault and gives you full control over its use; you
decide who to grant access, to which particular type of data, and for
what purposes. And you can withdraw your personal data at any point
in time.
But unlike your typical bank account, MIDATA isn’t interested in
generating profits (no ludicrous late fees). Its sole purpose is to
maximize value for you and its other members. Any net profits that
are generated from the use of your data get reinvested into making
the services on the platform better (i.e., advances in data protection
and technological developments). Similarly, as a member of MIDATA,
you aren’t just a bank customer. You own the bank. Literally. Control
at MIDATA not only means control over your personal data. It also
means having a direct say in the co-op’s governance through a general
assembly (so Swiss!).
The value MIDATA generates for its members takes different
shapes. You can share access to your data with third parties to
improve your own health. For example, there are personalized
applications to help members overcome addiction or fight obesity. But
you can also share your data to support scientific discovery, for
example, to help researchers better understand allergies, food
sensitivities, or rare disease. In many cases, the same application does
both.
Take MiTrendS, an application that drives the scientific exploration
and personalized treatment of multiple sclerosis (MS). MS is a
chronic autoimmune disease that affects the central nervous system
and cannot be cured (yet). By eating away the protective layer of nerve
fiber in the brain, spinal cord, and optic nerves, MS significantly
impacts patients’ quality of life. They might have trouble seeing,
experience fatigue, have a hard time concentrating and remembering
things, and struggle with balance and tremor. A vicious combination
of symptoms that often makes it difficult—if not impossible—for MS
patients to thrive and fully engage in social life.
Although MS affects over 2.5 million people around the world, the
disease remains hard to diagnose and even harder to treat. All we
know is that MS is driven by a complex combination of infectious,
genetic, and environmental factors. Because every patient has their
own unique set of symptoms and factors that contribute to the
outbreak, understanding the disease and developing targeted
treatments requires large amounts of data. Not just data from a lot of
patients, but also data from a lot of different sources: genetic data,
medical history, exposure to environmental risk factors, medication,
progression of symptoms over time, and more.
MiTrendS empowers patients and doctors to do just that. The app
allows users to track their symptoms over time from the comfort of
their home. For example, the app might ask you to follow a line on
your tablet with your finger as quickly and exactly as possible to test
your fine motor skills. Or match numbers to shapes to assess your
attention and concentration. By combining these symptom
assessments with existing patient records (i.e., medical records,
medication information, brain scans, blood analyses, and more),
MiTrendS can develop personalized treatments and care plans for
each patient, a revolutionary approach that could change how MS is
diagnosed and treated.
Of course, the big pharma companies might occasionally invest in
large data collection efforts or buy patient data from hospitals to
study the disease. MS medication is expensive, and there’s money to
be made. For the pharma companies, that is, of course. Not for the
patients whose data is used. In the best case, these patients will end
up paying for the medication. In the worst case, they will never get to
reap any benefits themselves.
The MiTrendS application turns this model upside down. By
sharing their data, patients help develop better treatments for the
future. But they also directly benefit from more customized and
targeted treatment in the here and now.
Once a patient’s data is securely stored and combined on the
MIDATA servers, a machine learning algorithm developed by
researchers at ETH Zürich creates an optimal, personalized treatment
plan for them. The algorithm not only considers the patient’s unique
circumstances, but also leverages the insights obtained from all the
other MS patients on the platform (with their explicit consent).
After the algorithm spits out a recommendation, it is passed on to
the MS specialists at the university hospitals that care for the patients.
These specialists both implement the suggested treatment and
provide feedback to the algorithm. Treatment X worked for patient A,
but didn’t work for patient B. It’s the perfect feedback loop to
continuously improve the algorithm’s predictions and, with it, the care
that can be offered to patients, a truly inspiring example of
personalized medicine (which I touched on earlier, in chapter 6).
But there’s more I love about MiTrendS. It involves the entire
(local) community in its mission. That includes patients with MS, of
course. But it also includes healthy individuals who can use the app to
help researchers establish reliable data for comparison.
 You can’t understand a disease without tracking a patient’s
symptoms. How do their neurological impairments progress over
time? Are they able to complete certain cognitive tasks, and how well?
But you also can’t understand a disease without having a clear sense
of what to expect if those people weren’t suffering from MS. How well
would regular people do at the task? How quickly do they get tired?
That’s what the MiTrendS citizen science part of the application
does. It makes the village come together to support its most
vulnerable members.

Making Data Co-ops a Viable Option

When I first heard of data co-ops a couple of years back, I
immediately loved the concept. It sounded like a powerful approach
to regaining all the fundamental rights we had lost in the transition to
the digital economy. Privacy, transparency, self-determination. Data
co-ops were designed to empower all of us; to not only take back
control over our personal data and lives, but also benefit from the
enormous value the digital economy had created (for a few big players
rather than all of us).
As much as I love the concept of data co-ops, implementing them
at scale is far from trivial. It requires us to fundamentally rethink the
data ownership model and create an infrastructure that facilitates the
collective management of personal data. But I am optimistic, mainly
because we have pulled off similar stunts before.
When the Industrial Revolution concentrated power in the hands
of a few major players, many citizens felt exploited and powerless.
Over time, however, communities of individuals came together to
form trade unions and citizen organizations that were guided by
common interests and the desire to provide a counterweight to the
big players.
Starting in the 1940s, for example, small member-owned electric
cooperatives united under the umbrella of the National Rural Electric
Cooperative Association to stand up against the energy giants of the
time. Today, these cooperatives own over 40 percent of the electric
infrastructure in the United States, covering more than 75 percent of
the country. Not a bad outcome for people who started with no power
at all.
Similarly, credit unions formed in response to the shift from
traditional cash-based barter to digital consumer banking. When
banks like J.P.Morgan threatened to dominate the market and exploit
people for their own benefit, credit unions popped up all over the
country. As nonprofit organizations with fiduciary responsibilities to
their members, they started to offer many of the same financial
services as traditional banks—minus the exploitation. Today, there are
about 5,000 official credit unions in the United States servicing over
130 million individuals. That’s more than one in every three
Americans.
The two examples show that shifting power back to the people is
possible in principle. However, what makes me so optimistic about
them is that they could lay the foundations of data co-ops.
As two MIT professors, Alex (Sandy) Pentland and Thomas
Hardjono, have convincingly argued, credit or trade unions could be
among the first and largest data co-ops.1 If you already entrust an
entity to keep the lights on at home, negotiate your labor rights, or
manage your investments and retirement fund, why not also entrust it
with your personal data? It’s the simplest way to give millions of
people access to a trustworthy advocate for their personal data,
practically overnight. Or as Pentland and Hardjono phrased it in a
joint report with multiples such unions, leveraging existing trade
unions could make the “widespread deployment of data cooperative
capabilities … surprisingly quick and easy.”
Regulatory environments—such as the European Union—which
have already shifted data ownership to individuals by regulating data
reuse and deletion, data interoperability, and portability, offer the
ideal breeding ground for data co-ops. Your data is much more
valuable when your data co-op is the only entity with access to it. As I
mentioned in the previous chapter, bargaining becomes a lot harder if
not only you have a copy of your data but everybody else does as well.
It’s the combination of a competent crew and a relatively tame sea
that enables you to make the most of your personal data.
 Most importantly, however, a competent crew is valuable even
when the sea is still rough. In fact, that’s perhaps when you need it the
most—when sailing your boat alone is unlikely to end well.
Most US citizens don’t currently own their data. If you have ever
tried to get access to some of the digital traces you generate, you’ll
know how difficult (or even impossible) a task that is. In most parts of
the United States, companies aren’t legally required to share your own
personal data with you. At the same time, companies have the right to
use, share, and sell your data to any paying third-party entity without
notifying you (and these third parties, in turn, are allowed to do the
same).2 Not a good spot to be in as a consumer. Left to your own
devices, you have little to no power. Nobody is going to pick up your
call and listen to your complaints and demands. But now imagine
getting calls from millions of union members who are represented by
expert lawyers. I bet someone is going to listen.
OceanofPDF.com
 EPILOGUE

The Moral Imperative to Shape Our Future

Finding better ways of managing our personal data will be critical in

the coming years and decades.
What I have described in Mindmasters is merely the tip of the
iceberg. Technology is evolving at light speed. It’s not just the growing
amount of data we generate but also the increasingly sophisticated
ways to analyze it that should give us pause and encourage us to
rethink our current approach. Soon, we might have microbots in our
blood that continuously scan our bodies for any sign of irregularity or
disease, smart contact lenses that capture what we see and hyper-
personalize our views, and chips in our brain that not only read our
thoughts but change them.
Sounds like science fiction? It isn’t. Microbots and smart lenses are
already a reality waiting to go mainstream. And the chips in our brain
aren’t so far off either. Neuroscientists are getting better and better at
speaking the language of the brain. They have uncovered ways to
project your thoughts on screens and explored different ways of
altering your brain’s wiring.1
Some of this research happens in academic institutions. But most
of it is funded by powerful private entities that are just as interested in
the commercialization of such technology as they are in scientific
discovery. Leading the pack is Elon Musk, whose company Neuralink
is tirelessly working toward this future.2
Everything I have discussed in Mindmasters pales in comparison
to a world where third parties don’t need to rely on psychological
inferences from digital footprints anymore but directly access them at
the source in our most secretive vault.
As we enter this world, we will learn more about the human psyche
than ever before. We will create new opportunities to support the
health and well-being of individuals and societies. Just imagine what
preventive health care could look like if the microbots in our
bloodstream detect the earliest signs of cancer long before any doctor
could ever diagnose the disease. How we could augment our physical
world with information projected directly onto our retinas to pique
our curiosity and inspire awe. Or how we could not just step into the
digital shoes of someone else but recreate their actual experience of
the world in our own brains.
At the same time, we will face unprecedented challenges that
threaten the very foundation of what makes us human. If we let third
parties project their preferred version of our reality onto our retinas,
how can we sustain a common foundation of what we collectively
believe about the world around us? And if we allow others direct
access to our brains, how will we know where a thought originated
and whether it’s truly ours? The last decade has seen growing
concerns about digital ecosystems amplifying discrimination,
polarization, echo chambers, and misinformation. If we continue our
current trajectory, the next few decades could very well lead to the
collapse of society as we know it.
Imagining the two sides of the same future creates a moral
imperative to rethink the current data environment. We need a new
social contract that defines what sharing our lives with others means
in today’s data-driven world. But rethinking the current data
environment isn’t enough. We need to recreate it. All of us. It doesn’t
matter whether you manage personal data for a living, or whether you
are sick of the power imbalance the digital economy has created. We
all have a role to play.
As I have suggested, returning to the village could be a solution.
I’m not talking about physical villages or a model of the past we have
moved beyond. I’m talking about a version of a village that is far
superior to anything we have seen before. Growing up in Vögisheim, I
had to live with both the advantages and disadvantages of letting
other people into my life. I gradually got better at amplifying the good
and downplaying the bad, but I had no hope of changing the system
altogether. Living in today’s digital village, we do. We have a unique
opportunity to reclaim control over our lives and create a collective
data infrastructure that benefits all of us.
OceanofPDF.com
 APPENDIX A

Personality Test

Let’s get started with figuring out your Big Five personality profile.1
First, you will need to answer a few questions about yourself. In the
following table, I have noted several characteristics that may apply to
you to various degrees. Some of them might be very close to how you
see yourself, and others might be quite far. There are no right or wrong
answers. The only purpose of this test is to help you understand your
own personality profile.
There’s no need to overthink your responses; the first one that
comes to mind is usually the best. To mark your response, just put the
number that best captures how you see yourself in the blank space
next to each question.

Neither
Disagree Disagree Disagree agree nor Agree Agree Agree
strongly moderately a little disagree a little moderately strongly
1 2 3 4 5 6 7

I see myself as:

1. ______ Extroverted, enthusiastic
2. ______ Critical, quarrelsome
3. ______ Dependable, self-disciplined
4. ______ Anxious, easily upset
5. ______ Open to new experiences, complex
6. ______ Reserved, quiet
7. ______ Sympathetic, warm
 Neither
Disagree Disagree Disagree agree nor Agree Agree Agree
strongly moderately a little disagree a little moderately strongly
1 2 3 4 5 6 7

8. ______ Disorganized, careless

9. ______ Calm, emotionally stable
10. ______ Conventional, uncreative

Source: Samuel D. Gosling, Peter J. Rentfrow, and William B. Swann Jr., “A Very Brief Measure of the Big Five
Personality Domains,” Journal of Research in Personality 37 (2003): 504–528.

Done? Here’s how to score the test. There are two questions for each
of the five personality traits. Say you score yourself a 5 on question 1
and a 2 on question 6, your extroversion score would be 5 + (8 − 2) =
11. The next table shows you which questions belong to which trait.

Openness = Score on question 5 + (8 − score on question 10)

Conscientiousness = Score on question 3 + (8 − score on question 8)
Extroversion = Score on question 1 + (8 − score on question 6)
Agreeableness = Score on question 7 + (8 − score on question 2)
Neuroticism = Score on question 4 + (8 − score on question 9)

Got your scores? Great! But wait a second, what does it mean to
have a score of 5.5 on extroversion? Is that high or low? You will notice
that the scores you just calculated don’t mean much without any
context. To make sense of them, we need to compare your scores to
those of other people (in this case, thousands of people who have taken
the same test before).

FIGURE A-1

Average scores for Big Five personality traits

Source: Samuel D. Gosling, Peter J. Rentfrow, and William B. Swann Jr., “A Very Brief Measure of the Big Five
Personality Domains,” Journal of Research in Personality 37, no. 6 (2003): 504–528.

I have plotted the average for each trait in figure A-1, along with
bands that indicate how extreme your score might be. You can take the
scores you calculated and transfer them as crosses into the graph (if
you want a more detailed profile, I recommend taking the test on our
dedicated website www.mindmasters.ai/mypersonality).
OceanofPDF.com
 APPENDIX B

We didn’t get to explore the relationships of Facebook likes and status

updates for all the five personality traits in the main book. But here
they are. Table B-1 lists Facebook likes related to these traits, while
figure B-1 depicts words in people’s Facebook status updates indicative
of these same traits. Just like the lists of likes and word clouds for
extroversion and agreeableness in chapter 2, the ones for openness,
conscientiousness, and neuroticism showcase how intuitive many of
the links between our online behaviors and psychological
characteristics are.

TABLE B-1

Facebook likes related to openness, conscientiousness, and neuroticism

Personality trait Low High

Openness NASCAR Oscar Wilde

Austin Collie Charles Bukowski
Monster-in-Law Leonardo da Vinci
I don’t read Bauhaus
Justin Moore Dmt: The Spirit Molecule
ESPN2 American Gods
Farmlandia John Waters
The Bachelor Plato
Teen Mom 2 Leonard Cohen
Conscientiousness Wes Anderson Law officer
Bandit Nation National law enforcement
Omegle Lowfares.com
Vocaloid Accounting
Serial Killer Foursquare
Screamo Emergency medical services
Anime Sunday Best
Vamplets Kaplan University
Join If Ur Fat Glock Inc
Personality trait Low High
Not Dying MyCalendar 2010
Neuroticism Business administration Sometimes I hate myself
Getting money Emo
Parkour Girl Interrupted
Track and field SO SO Happy
Skydiving The Addams Family
Mountain biking Vocaloid
Soccer Sixbillionsecrets.com
Climbing Vampires Everywhere
Physics/engineering Kurt Donald Cobain
48 Laws of Power Dot Dot Curve

FIGURE B-1
Facebook status updates related to openness, conscientiousness, and neuroticism
Scan the QR code for a color version of the word clouds.
Source: H. Andrew Schwartz et al., “Personality, Gender, and Age in the Language of Social Media: The Open-
Vocabulary Approach,” PloS One 8, no. 9 (2013): e73791, https://doi.org/10.1371/journal.pone.0073791. Permission
via https://creativecommons.org/licenses/by/4.0/.

OceanofPDF.com
 NOTES

Introduction
1. Michael Reilly, “Is Facebook Targeting Ads at Sad Teens?,” MIT Technology Review, May
1, 2017, https://www.technologyreview.com/2017/05/01/105987/is-facebook-targeting-ads-
at-sad-teens.
2. Melvin Kranzberg, “Technology and History: ‘Kranzberg’s Laws,’ ” Technology and
Culture 27, no. 3 (July 1986): 544–560.
 Chapter 1
1. “Data Never Sleeps,” DOMO, 2018, https://www.domo.com/solution/data-never-sleeps
-6.
2. Youyou Wu, Michal Kosinski, and David Stillwell, “Computer-based Personality
Judgments Are More Accurate Than Those Made by Humans,” Proceedings of the National
Academy of Sciences 112, no. 4 (2015): 1036–1040.
3. Gerald Matthews, Ian J. Deary, and Martha C. Whiteman, Personality Traits
(Cambridge, UK: Cambridge University Press, 2003).
 Chapter 2
1. Diana I. Tamir and Jason P. Mitchell, “Disclosing Information about the Self Is
Intrinsically Rewarding,” Proceedings of the National Academy of Sciences 109, no. 21 (2012):
8038–8043.
2. Mor Naaman, Jeffrey Boase, and Chih-Hui Lai, “Is It Really about Me? Message
Content in Social Awareness Streams,” Proceedings of the 2010 ACM Conference on Computer
Supported Cooperative Work, February 2010, 189–192.
3. Mitja D. Back et al., “Facebook Profiles Reflect Actual Personality, not Self-Idealization,”
Psychological Science 21, no. 3 (2010): 372–374.
4. Yoko Akiyoshi, “Retired Japanese Police Officer Sets New Hello Kitty Record,” NBC
News, July 4, 2017, https://www.nbcnews.com/news/world/retired-japanese-police-officer-
sets-new-hello-kitty-record-n779476.
5. Michal Kosinski, David Stillwell, and Thore Graepel, “Private Traits and Attributes Are
Predictable from Digital records of Human Behavior,” Proceedings of the National Academy of
Sciences 110, no. 15 (2013): 5802–5805.
6. James W. Pennebaker, The Secret Life of Pronouns (New York: Bloomsbury Publishing
USA, 2013).
7. Allison M. Tackman et al., “Depression, Negative Emotionality, and Self-referential
Language: A Multi-lab, Multi-measure, and Multi-language-task Research Synthesis,”
Journal of Personality and Social Psychology 116, no. 5 (2019): 817.
8. Johannes C. Eichstaedt et al., “Facebook Language Predicts Depression in Medical
Records,” Proceedings of the National Academy of Sciences 115, no. 44 (2018): 11203–11208.
9. Jonathan Timm, “When the Boss Says ‘Don’t Tell Your Coworkers How Much You Get
Paid,” Atlantic, July 15, 2014, https://www.theatlantic.com/business/archive/2014/07/when-
the-boss-says-dont-tell-your-coworkers-how-much-you-get-paid/374467/.
10. Sandra C. Matz et al., “Predicting Individual-level Income from Facebook Profiles,”
PloS One 14, no. 3 (2019): e0214369.
11. Cristina Segalin et al., “The Pictures We Like Are Our Image: Continuous Mapping of
Favorite Pictures into Self-assessed and Attributed Personality Traits,” IEEE Transactions on
Affective Computing 8, no. 2 (2016): 268–285.
12. Yilun Wang and Michal Kosinski, “Deep Neural Networks Are More Accurate Than
Humans at Detecting Sexual Orientation from Facial Images,” Journal of Personality and
Social Psychology 114, no. 2 (2018): 246.
13. Aaron W. Lukaszewski and James R. Roney, “The Origins of Extraversion: Joint
Effects of Facultative Calibration and Genetic Polymorphism,” Personality and Social
Psychology Bulletin 37 (2011): 409–421.
 Chapter 3
1. Sam Gosling, Snoop: What Your Stuff Says about You (New York: Basic Books, 2008).
2. Seth Stephens-Davidowitz, Everybody Lies: Big Data, New Data, and What the Internet
Can Tell Us About Who We Really Are (New York: Dey Street Books, 2017).
3. Laokoon, Made to Measure: A Digital Search for Traces, 2020, https://www
.madetomeasure.online/en/.
4. Yves-Alexandre de Montjoye et al., “Unique in the Shopping Mall: On the
Reidentifiability of Credit Card Metadata,” Science 347, no. 6221 (2015): 536–539.
5. Joe J. Gladstone, Sandra C. Matz, and Alain Lemaire, “Can Psychological Traits Be
Inferred from Spending? Evidence from Transaction Data,” Psychological Science 30, no. 7
(2019): 1087–1096.
6. Jessie London, “How Jaila Gladden’s iPhone Saved Her Life,” Medium, January 21,
2021, https://jessielondon.medium.com/how-jaila-gladdens-iphone-saved-her-life
-48d0c285d147.
7. Shiri Melumad and Michel Tuan Pham, “The Smartphone as a Pacifying Technology,”
Journal of Consumer Research 47, no. 2 (2020): 237–255.
8. Clemens Stachl et al., “Predicting Personality from Patterns of Behavior Collected with
Smartphones,” Proceedings of the National Academy of Sciences 117, no. 30 (2020): 17680–
17687.
9. Sandrine R. Müller et al., “Depression Predictions from GPS-based Mobility Do Not
Generalize Well to Large Demographically Heterogeneous Samples,” Scientific Reports 11, no.
1 (2021): 14007.
 Chapter 4
1. William Fleeson, “Toward a Structure- and Process-Integrated View of Personality:
Traits as Density Distributions of States,” Journal of Personality and Social Psychology 80,
no. 6 (2001): 1011.
2. Robert E. Wilson, Renee J. Thompson, and Simine Vazire, “Are Fluctuations in
Personality States More Than Fluctuations in Affect?,” Journal of Research in Personality 69
(2017): 110–123.
3. Sandra C. Matz and Gabriella M. Harari, “Personality–Place Transactions: Mapping the
Relationships between Big Five Personality Traits, States, and Daily Places,” Journal of
Personality and Social Psychology 120, no. 5 (2021): 1367.
4. Andrew B. Blake et al., “Wearable Cameras, Machine Vision, and Big Data Analytics:
Insights into People and the Places They Go,” In Big Data in Psychological Research, S. E.
Woo, L. Tay, and R. W. Proctor, eds., (Washington, DC: American Psychological Association,
2020), 125–143.
5. John F. Rauthmann et al., “The Situational Eight DIAMONDS: A Taxonomy of Major
Dimensions of Situation Characteristics,” Journal of Personality and Social Psychology 107,
no. 4 (2014): 677.
6. Ramona Schoedel et al., “Snapshots of Daily Life: Situations Investigated through the
Lens of Smartphone Sensing,” Journal of Personality and Social Psychology 125, no. 6
(2023): 1442–1471.
7. Heinrich Peters and Sandra Matz, “Large Language Models Can Infer Psychological
Dispositions of Social Media Users,” PNAS Nexus 3, no. 6 (2024), https://academic.oup.com
/pnasnexus/article/3/6/pgae231/7692212.
 Chapter 5
1. Hannes Grassegger and Mikael Krogerus, “Ich habe nur gezeigt, dass es die Bombe gibt
(I Just Showed That the Bomb Exists),” TA International, March 20, 2018, https://www
.tagesanzeiger.ch/ich-habe-nur-gezeigt-dass-es-die-bombe-gibt-652492646668.
2. Hannes Grassegger and Mikael Krogerus, “The Data That Turned the World Upside
Down,” Motherboard, January 28, 2017, https://www.vice.com/en/article/mg9vvn/how-our-
likes-helped-trump-win.
3. Harry Davies, “Ted Cruz Using Firm That Harvested Data on Millions of Unwitting
Facebook Users,” Guardian, December 11, 2015, https://www.theguardian.com/us-news
/2015/dec/11/senator-ted-cruz-president-campaign-facebook-user-data.
4. Michal Kosinski, David Stillwell, and Thore Graepel, “Private Traits and Attributes Are
Predictable from Digital Records of Human Behavior,” Proceedings of the National Academy
of Sciences 110, no. 15 (2013): 5802–5805.
5. “Hilton Launches Holiday Matchmaking App,” Breaking Travel News, July 14, 2015,
https://www.breakingtravelnews.com/news/article/hilton-launches-holiday-matchmaking-
app/.
6. Sidney J. Levy, “Symbols for Sale,” Harvard Business Review, July–August 1959, 117–
124.
7. Sandra C. Matz et al., “Psychological Targeting as an Effective Approach to Digital Mass
Persuasion,” Proceedings of the National Academy of Sciences 114, no. 48 (2017): 12714–12719.
8. Sandra C. Matz et al., “Predicting the Personal Appeal of Marketing Images Using
Computational Methods,” Journal of Consumer Psychology 29, no. 3 (2019): 370–390.
9. The quote is ChatGPT’s response to author’s question “Write an iPhone ad for someone
who is extroverted and enthusiastic,” obtained March 25, 2023, using OpenAI’s GPT-3
playground.
10. Sandra Matz et al., “The Potential of Generative AI for Personalized Persuasion at
Scale,” Scientific Reports 14, no. 1 (2023): 4692.
11. Jonathan Haidt and Craig Joseph, “The Moral Mind: How Five Sets of Innate
Intuitions Guide the Development of Many Culture-specific Virtues, and Perhaps Even
Modules,” The Innate Mind 3 (2007): 367–391.
12. Jesse Graham, Jonathan Haidt, and Brian A. Nosek, “Liberals and Conservatives Rely
on Different Sets of Moral Foundations,” Journal of Personality and Social Psychology 96, no.
5 (2009): 1029.
13. Matthew Feinberg and Robb Willer, “Moral Reframing: A Technique for Effective and
Persuasive Communication across Political Divides,” Social and Personality Psychology
Compass 13, no. 12 (2019): e12501.
14. Feinberg and Willer, “Moral Reframing”; Matz, et al., “Predicting the Personal Appeal
of Marketing Images.”
15. Jennifer Nancy Lee Allen, Duncan J. Watts, and David Rand, “Quantifying the Impact
of Misinformation and Vaccine-Skeptical Content on Facebook,” PsyArXiv, September 9,
2023, doi:10.31234/osf.io/nwsqa.
 Chapter 6
1. Sandra C. Matz, Joe J. Gladstone, and David Stillwell, “Money Buys Happiness When
Spending Fits Our Personality,” Psychological Science 27, no. 5 (2016): 715–725.
2. Ann Carrns, “Even in Strong Economy, Most Families Don’t Have Enough Emergency
Savings,” New York Times, October 25, 2019, https://www.nytimes.com/2019/10/25/your-
money/emergency-savings.html.
3. Sendhil Mullainathan and Eldar Shafir, Scarcity: Why Having Too Little Means So
Much (New York: Macmillan, 2013).
4. Carrns, “Even in Strong Economy, Most Families Don’t Have Enough Emergency
Savings.”
5. Sandra C. Matz, and Joe J. Gladstone, “Nice Guys Finish Last: When and Why
Agreeableness Is Associated with Economic Hardship,” Journal of Personality and Social
Psychology 118, no. 3 (2020): 545.
6. Sandra C. Matz, Joe J. Gladstone, and Robert A. Farrokhnia, “Leveraging Psychological
Fit to Encourage Saving Behavior,” American Psychologist 78, no. 7 (2023): 901–917.
7. Copyright © 2023 by the American Psychological Association. Reproduced by
permission. Sandra C. Matz, Joe J. Gladstone, and Robert A. Farrokhnia, “Leveraging
Psychological Fit to Encourage Saving Behavior,” American Psychologist 78, no. 7 (2023):
901–917.
8. My eternal gratitude and admiration go to the SaverLife creative team that pulled off
this stunt before ChatGPT was a thing!
9. Sam Levin, “Facebook Told Advertisers It Can Identify Teens Feeling ‘Insecure’ and
‘Worthless,’ ” Guardian, May 1, 2017, https://www.theguardian.com/technology/2017/may
/01/facebook-advertising-data-insecure-teens.
10. Sung Jun Park et al., “New Paradigm for Tumor Theranostic Methodology Using
Bacteria-based Microrobot,” Scientific Reports 3, no. 1 (2013): 3394.
11. Robert Lewis et al., “Can a Recommender System Support Treatment Personalisation
in Digital Mental Health Therapy? A Quantitative Feasibility Assessment Using Data from a
Behavioural Activation Therapy App,” in CHI Conference on Human Factors in Computing
Systems Extended Abstracts (2022), 1–8.
12. Yuki Nogucki, “Therapy by Chatbot? The Promise and Challenges in Using AI for
Mental Health,” Shots, NPR, January 19, 2023, https://www.npr.org/sections/health-shots
/2023/01/19/1147081115/therapy-by-chatbot-the-promise-and-challenges-in-using-ai-for-
mental-health.
13. Nick Zagorski, “Popularity of Mental Health Chatbots Grows,” Psychiatric News 57, no.
5 (2022), https://psychnews.psychiatryonline.org/doi/10.1176/appi.pn.2022.05.4.50.
14. Eli J. Finkel et al., “Political Sectarianism in America,” Science 370, no. 6516 (2020):
533–536.
15. Thomas H. Costello, Gordon Pennycook, and David Rand, “Durably Reducing
Conspiracy Beliefs through Dialogues with AI,” PsyArXiv, April 3, 2024,
doi:10.31234/osf.io/xcwdn.
 Chapter 7
1. Paul Mozur, Muyi Xiao, and John Liu, “ ‘An Invisible Cage’: How China Is Policing the
Future,” New York Times, June 25, 2022, https://www.nytimes.com/2022/06/25/technology
/china-surveillance-police.html.
2. Paul Mozur, Muyi Xiao, and John Liu, “ ‘An Invisible Cage.’ ”
3. “Sun on Privacy: ‘Get Over It,’ ” Wired, January 26, 1999, https://www.wired.com/1999
/01/sun-on-privacy-get-over-it/.
4. Eileen Guo, “A Roomba Recorded a Woman on the Toilet. How Did Screenshots End
Up on Facebook?,” MIT Technology Review, December 19, 2022, https://www
.technologyreview.com/2022/12/19/1065306/roomba-irobot-robot-vacuums-artificial-
intelligence-training-data-privacy/.
5. Amanda Reill, “A Simple Way to Make Better Decisions,” hbr.org, December 6, 2023,
https://hbr.org/2023/12/a-simple-way-to-make-better-decisions.
 Chapter 8
1. Chris Jay Hoofnagle and Jennifer M. Urban, “Alan Westin’s Privacy Homo Economicus,”
Wake Forest Law Review 49 (2014): 261.
2. Azim Shariff, Joe Green, and William Jettinghoff, “The Privacy Mismatch: Evolved
Intuitions in a Digital World,” Current Directions in Psychological Science 30, no. 2 (2021):
159–166.
3. Alessandro Acquisti, Laura Brandimarte, and George Loewenstein, “Privacy and
Human Behavior in the Age of Information,” Science 347, no. 6221 (2015): 509–514.
 Chapter 9
1. Eric J. Johnson and Daniel G. Goldstein, “Do Defaults Save Lives?,” Science 302, no.
5649 (2003): 1338–1339.
2. Aristotle, The Nicomachian Ethics of Aristotle, tenth edition, trans. F. H. Peters
(London: Kegan Paul, Trench, Trübner and Co., 1906), 289.
3. Alessandro Acquisti, Leslie K. John, and George Loewenstein, “What Is Privacy
Worth?,” Journal of Legal Studies 42, no. 2 (2013): 249–274.
4. Peter Kairouz et al., “Advances and Open Problems in Federated Learning,”
Foundations and Trends in Machine Learning 14, no. 1–2 (2021): 1–210.
5. Cory Doctorow, “Personal Data Is as Hot as Nuclear Waste,” Guardian, January 15,
2008, https://www.theguardian.com/technology/2008/jan/15/data.security.
6. Nina Totenberg, “An Attacker Killed a Judge’s Son. Now She Want to Protect Other
Families,” NPR, November 20, 2020, https://www.npr.org/2020/11/20/936717194/a-judge-
watched-her-son-die-now-she-wants-to-protect-other-judicial-families.
7. The Daniel Anderl Judicial Security and Privacy Act, The Courts and Congress Annual
Report 2022, https://www.uscourts.gov/statistics-reports/courts-and-congress-annual-
report-2022.
8. Roger McNamee, Zucked: Waking Up to the Facebook Catastrophe (New York: Penguin,
2020).
 Chapter 10
1. Alex Pentland and Thomas Hardjono, “Data Cooperatives,” in Building the New
Economy (Cambridge, MA: MIT Press, 2020), https://wip.mitpress.mit.edu/pub/pnxgvubq
/release/2.
2. Thorin Klosowski, “The State of Consumer Data Privacy Laws in the US (And Why It
Matters),” Wirecutter, New York Times, September 6, 2021, https://www.nytimes.com
/wirecutter/blog/state-of-privacy-laws-in-us/.
 Epilogue
1. Moran Cerf et al., “On-line, Voluntary Control of Human Temporal Lobe Neurons,”
Nature 467, no. 7319 (2010): 1104–1108.
2. Daniel Gilbert, “The Race to Beat Elon Musk to Put Chips in People’s Brains,”
Washington Post, March 3, 2023, https://www.washingtonpost.com/business/2023/03/03
/brain-chips-paradromics-synchron/.

Appendix A
1. Samuel D. Gosling, Peter J. Rentfrow, and William B. Swann Jr., “A Very Brief Measure
of the Big Five Personality Domains,” Journal of Research in Personality 37 (2003): 504–528.

OceanofPDF.com
 INDEX

Abel, Martin, 157

abortion, 133, 144
accelerometers, 71–72
Acquisti, Alessandro, 170
advertising, targeted, 93–94, 97–99
agreeableness, 22, 24–25
Facebook likes related to, 35–36
Facebook status updates and, 41, 42, 43
financial health and, 115–116
loan defaults and, 147
stress and, 83
algorithms. See also machine learning
biases in, 146–147
lack of reciprocity in targeting by, 149–151
political polarization and, 129–134
Ali, Chukurah, 124–125
Allen, Jennifer, 108
Amazon Web Services, 179
antitrust laws, 178–179
Apollo 11, 172
Apple, 121, 172–173
ApplyMagicSauce, 156–159
apps, personality traits and use of, 73–74
Ariely, Dan, 167–168
Aristotle, 170
The Atlantic, 46
AT&T, 179
attention, our desire for, 29–30
Australian, 119–120

balance, 7
beauty retailing, 99–103
behavioral residue, 9
digital breadcrumbs and, 61–62
identity claims vs., 77
behavior change, 166–174
Behm, Kyle, 145–146, 148
Bell Systems, 179
biases
egocentrism, 105–106
endowment effect, 170–171
human vs. algorithmic, 146–147
memory and, 125–126
Biden, Joe, 175
Big Five personality traits, 20–27
advantages and disadvantages of each, 26–27
agreeableness, 24–25
conscientiousness, 22, 23
credit cards and, 67–69
extroversion, 22, 23–24
Facebook likes and, 34–37, 201–203
Facebook status updates and, 40–42
facial characteristics and, 54–56
financial health and, 114–118
neuroticism, 25–26
openness to experience, 21–23
smartphone logs and, 73–74
test for, 197–199
Bluetooth sensors, 71, 72
brain
chip interfaces with, 89
deep learning efforts to imitate, 53–56
pleasure center of, self-disclosure and, 29–30
privacy issues and, 162–163
Brant, Tamir, 70
Bush, George W., 95

California Consumer Protection Act, 161, 167

Cambridge Analytica, 8–9, 93–95, 103, 140–141, 148–149
Cambridge Psychometrics Center, 95–96, 156–160
cancer detection, 121, 194
cause and effect
Facebook status updates, depression, and, 45–46
chatbots, 185
ChatGPT, 89, 102–103, 105
job applications and, 146
mental health, 123–129
ChatGPT, 89
content creation by, 102–103
moral reframing using, 105
political polarization and, 133–134
for PTSD, 129
chick sexing, 18–19
China, profiling technology in, 137–138
climate change, 105–107
Clinton, Hillary, 8–9, 93–95
cognitive ability, Facebook likes related to, 36–38
Collins, Joan, 82
conscientiousness, 22, 23
Facebook likes and, 202
Facebook status updates and, 203
places we visit and, 83–85
saving money and, 114–115
smartphone logs related to, 73–74
spending patterns and, 67–68
conspiracy theories, 134
content creation, 102
contextual clues, 9
control, 164
data ecosystem design and, 166–180
data protection regulations and, 161–163, 164
moving beyond, 155–164
as right vs. responsibility, 160–163
convenience vs. service, 140–142
convention, preference for, 21–23
co-ops, data, 181–191
how they work, 183–186
making them viable, 189–191
cortisol, tracking, 122
Costello, Thomas, 134
Covid-19 pandemic, 117
mental health during, 124
credit cards, 57
digital breadcrumbs from, 62, 65–69
credit unions, 190–191
Cruz, Ted, 94
Crystal Knows, 146

Daniel Anderl Judicial Security and Privacy Act, 175, 177

Darcy, Alison, 128
data
amount generated hourly, 16
beyond control of, 155–164
brokers of, 176–177
changing the rules on, 8–11
company tradeoffs around, 176–177
compared with nuclear waste, 174–175
co-ops for, 181–191
deleting personal, 166–174
federated learning and, 171–174
 in machine learning, 18–20
McNamee’s seventh law on, 177–178
monetization of, 185–186
moral imperative to shape the future and, 193–195
ownership of, 181–191
perils and promise of analyzing, 5–7
pictures and, 48–50
preventing collection of complete, 177–179
privacy regulations for, 160–163, 164
psychological information from, 15–27
safeguards around, 175–180
sharing value generated by, 155–156
taxing collection of, 176–177
uncertainty about use of, 162–163
understanding the value of digital footprints and, 156–160, 166–168
data ecosystems, 165–180, 194–195
closing the wrong channels in, 174–180
company tradeoffs in, 176–177
designing, 165–166
opening the right channels in, 166–174
opt-in vs. opt-out, 168–171, 180
preventing complete data collection in, 177–179
privacy and service in, 171–174
regulatory environments and, 160–163, 164
systemic approaches for, 180
decision-making
about data privacy, 160–163
behavior change and, 167–168
data ecosystem design and, 165–180
endowment effect and, 170–171
influencing, 103–106
self-determination and, 144–149
deep learning, 53–56
deep neural networks, 53–56
delayed gratification, 115, 116
democracy, saving, 107–108
Den Hollander, Roy, 174–175
depression, 5–6
Facebook status updates and, 43–46
first-person pronoun use by, 44
Google searches and, 64
GPS data related to, 75–76
personalized health tracking and, 120–121
predicting, 76–77
DIAMONDS framework, 86
digital breadcrumbs, 59–78
credit cards and, 65–69
smartphones and, 62, 70–77
digital footprints
credit cards and, 65–69
digital breadcrumbs and, 59–78
morning routines and, 15–16
psychological information contained in, 19–20
understanding the value of, 156–160
digital literacy, 162–163
digital neighbors, 4
discrimination, 137–138, 145–147, 149–151, 194
data co-ops around, 184
dissociative identity disorder, 79–80
Doctorow, Cory, 174–175
Driver’s Seat Cooperative, 186

Eichstaedt, Johannes, 44, 46, 129

emotional distress, word choices indicating, 43–46
emotional stability, 25–26
endowment effect, 170–171
environmental influences on personality, 79–90
ethics, 138
ETH Zürich, 186–189
Everybody Lies (Stephens-Davidowitz), 62–63
extroversion, 22, 23–24
beauty products and, 99–103
Facebook likes related to, 34–35
Facebook status updates and, 40, 41
facial characteristics related to, 54–56
mood and, 82–83
places we visit and, 83–85
spending patterns and, 67–68
stress and, 83
extroversion states, 80–82

Facebook
beauty product ads on, 100–101
Cambridge Analytica and, 8–9, 93–95, 103, 140–141
how well it knows us, 17
likes, 32–38, 95–96
likes, gender and, 32, 33
likes, personality profiles and, 17
likes, personality traits and, 201–203
mental health data sold by, 119–120
political polarization and, 131–134
status updates and gender, 32, 33
status updates and personality, 32, 33
understanding the value of digital footprints from, 156–160
word choice in posts on, 39–49
facial characteristics, 50–53
personality and, 53–56
facial recognition technology, 57
Farrokhnia, Robert, 117
federated learning, 171–174
Feinberg, Matthew, 105
financial health, 113–119, 147
Financial Times, 17
first impressions, 86, 155–156
Fleeson, William, 80
Flickr, 50
Forbes, 114
friction, reducing for behavior change, 168

Gaga, Lady, 24, 26, 38–39

Galloway, Scott, 179
gamification, 122–123
General Data Protection Regulations, 161, 167
generative AI, 89
ChatGPT, 89, 102–103, 105
compared to human beings, 127–129
mental health chatbots, 122–129
political polarization and, 133–134
Germany, Nazis in, 143–144
Gladden, Jaila, 70
Gladstone, Joe, 67–69, 117
Google
digital breadcrumbs from, 61, 62–65
health unit at, 121
Maps, 140, 141, 171
political polarization and, 131–134
preventing complete data collection by, 178
smart glasses/lenses, 89
Street View, judging situations with, 87–88
understanding the value of digital footprints from, 156–160
Gosling, Sam, 61
GPS tracking, 5, 16
depression and, 75–76, 120–121
digital breadcrumbs from, 62, 70–71, 72
influence of places on personality and, 84–85
“it’s worth it” fallacy and, 140–142
privacy and service with, 171
Graepel, Thore, 34–37
Grassegger, Hannes, 94
Grayling, 98–99
Grayson, Samantha, 83
Guardian (newspaper), 94, 102, 174
Guardians app, 122–123
Gunji, Masao, 31, 101

Haidt, Jonathan, 104

Hall, Tom, 52
happiness
financial health and, 5, 113, 134
science of, 59–60
tracking, 83
Harari, Gabriella, 84
Hardjono, Thomas, 190–191
Harvard Business Review, 99
HBO, 145–146
health
apps tracking, 83
data co-ops for, 184, 186–189
improving, 96
mental health and, 46
microbots and, 89, 121, 193, 194
personalized tracking of, 120–122
personalized treatment of, 122–129
preventive care and, 194
smartphone logs related to, 74–77
treatment personalization for, 122–129
heart disease, 96
Hello Kitty, 31, 101
Hilton Hotels, 98–99, 104
Holocaust, 143–144
Hoofnagle, Chris, 162

identity, 9
core, 80–82
curating your, 77
digital breadcrumbs and, 61–62, 77
fixed vs. changeable, 77–78
online, 29–58
spending patterns and, 66–69
identity claims, 30
“I have nothing to hide” fallacy, 140, 142–144, 162
incentive structures, 176–177, 180
data co-ops and, 184–190
income, 46–48
Industrial Revolution, 190
inertia, 168–171
insights
data co-ops and, 184–186
digital village life and, 4–8
 psychological targeting and, 4–7
social media and, 32–33
understanding the value of digital footprints and, 156–163
village life and, 1–3
Instagram, 173–174, 179
intelligence, Facebook likes related to, 36–38
International Classification of Diseases, 10th rev., 76
internet searches, 9
introversion, 22, 23–24
beauty products and, 99–103
Facebook likes related to, 34, 35
Facebook status updates and, 40, 41
facial characteristics related to, 54–56
spending patterns and, 67–68
iRobot, 141–142
“it’s worth it” fallacy of data sharing, 140–142, 162

job applications, 145–147

Jobs, Steve, 27
J.P.Morgan, 190

Khambatta, Poruz, 54–55

Kosinski, Michal, 34–37, 51–52, 53, 54–55, 93
Kranzberg, Melvin, 10
Kroger, 145–146
Kronos, 145–146

language use, social media posts and, 38–48

learning, federated, 171–174
Lemaire, Alain, 67–69
Levy, Sidney, 99
Lewis, Rob, 122–123
LGBTQ+ people
facial characteristics and, 51
Google searches and, 64
“I have nothing to hide” fallacy and, 143
lifestyle changes, 96
light sensors, 71, 72
Limbic, 123–124
Li Wei, 138
loan defaults, 147
Lombroso, Cesare, 51

machine learning, 17–20

deep learning, 53–56
generative AI and, 89
 predictions made in, 31–32
Made to Measure (documentary), 63–64
manipulation, 2–3, 148–151
marketing, targeted, 97–103
McNamee, Roger, 177–178
McNamee’s seventh law, 177–178
McNealy, Scott, 139
medical microbots, 121–122
medical research, 182–183
Melumad, Shiri, 70
memory, unreliability of, 125–126
mental health, 5–6
access to treatment for, 112, 124–129
Facebook data on, 119–120
financial health and, 114
GPS tracking and, 74–77
job discrimination and, 145–147
language/word choice and, 42–46
physical health and, 46
targeting improved, 119–129
tracking, personalized, 120–122
tracking and treating, 120
treatment personalization for, 122–129
mePrism, 168
Merkel, Angela, 23
Messi, Lionel, 34
microbots, medical, 89, 121, 193, 194
MIDATA, 186–189
Milligan, Billy, 79–80
Mindf•ck (Whiley), 94
misinformation, 107–108, 194
MIT, 122–123, 134
MIT Media Lab, 121
money
saving, 112, 113–119
spending patterns and, 57, 65–69
Monsters Inside (documentary), 80
Montjoye, Yves-Alexandre de, 66
mood, personality and, 82–83
Moon, Alice, 15
moral reframing, 105–107
moral values, 104–107
motivation, 103–106
behavior change and, 167–168
marketing and, 97–98
moral reframing and, 104–107
self-determination and, 148–149
Müller, Sandrine, 75
multiple sclerosis (MS), 187–189
Musk, Elon, 89, 194

National Labor Relations Act of 1935, 46

National Police University, China, 138
National Rural Electric Cooperative Association, 190
Nazis, 143–144
Netflix, 80, 122, 172
Neuralink, 89, 194
neuroticism, 22, 25–26
Facebook likes and, 202
Facebook status updates and, 203
social media pictures related to, 50
news, slanted, 107–108
New York Times, 137–138
Noguchi, Yuki, 124–125
novelty, preference for, 21–23

Obama, Barack, 34, 38–39, 94

object recognition algorithms, 49
OCEAN model. See Big Five personality traits
OpenAI, 89, 102–103
openness to experience, 21–23
Facebook likes and, 202
Facebook status updates and, 203
saving money and, 117–118
smartphone logs related to, 74
opt-in vs. opt-out systems, 168–171, 180
organ donation, 168–171
ownership, data co-ops and, 181–191

Park, Sukho, 121

Pennebaker, James, 39, 42–46
Pentland, Alex, 190–191
Permanent Record (Snowden), 143
personality
beauty products and, 99–103
Big Five traits of, 20–27
credit cards and, 67–69
digital breadcrumbs and, 61–62
extroversion states and, 80–82
Facebook likes and, 17
facial characteristics and, 51–53
financial health and, 114–118
fixed vs. changeable, 77–78
judgment of by strangers, 61
 lay theories of, 20–21
mood and, 82–83
nuances in and influences on, 79–90
predictive models of, 95–96
privacy, self-determination, and, 145–149
scientific models of, 21
situations and, 85–88
smartphones and, 73–74
stress and, 83
testing your, 197–199
personality profiles, 26
Persona: The Dark Truth Behind Personality Tests (documentary), 145–146
perspectives, broadening political, 129–134
Peters, Heinrich, 89
physiognomy, 50–53
Picard, Rosalind, 121
Picasso, Pablo, 22–23
pictures/photographs, 48–50, 57
personality affinity and, 102
Piglet, 25
places, influence of on personality, 83–85
polarization, political, 129–134, 194
political ideas/candidates, 8–9, 93–95, 103–105, 107
broadening our views of, 129–134
polarization and, 194
presidential elections and, 8–9, 93–95, 107–108, 130, 140–141, 148–149
political participation, 107, 112, 129–134
post-traumatic stress disorder (PTSD), 129
power, 7
data co-ops and, 181–191
“I have nothing to hide” fallacy and, 142–144
of moral reframing, 104–107
motivations and, 103–106
privacy as, 138
of self-determination, 144–149
pregnancy, data co-ops around, 183–184
presidential elections
2016, 8–9, 93–95, 103, 140–141, 148–149
2024, 107–108
privacy, 56–58
balancing with security and strength, 7
beyond social media, 56–58
death of?, 139–144
by design, 168–174
encouraging behavior around, 166–174
“I have nothing to hide” fallacy and, 140, 142–144, 162
“it’s worth it” fallacy and, 140–142, 162
opt-in vs. opt-out systems for, 168–171, 180
 policies, understanding, 162
as power, 138
as responsibility vs. right, 160–163
self-determination and, 144–149
service and, 171–174
understanding the value/importance of, 159–163
village life and, 1–4
propaganda, 151
psychological profiles
dangers of, 137–151
facial characteristics and, 50–53
guesswork in computer versions of, 88–89
pictures and, 48–50
understanding the value of digital, 156–160
psychological targeting, 4–7, 8–11
in action, 93–109
algorithmic vs. traditional, 149–151
to benefit people, 5–6
dangers of, 108–109
definition of, 4
Google searches and, 63–65
intuition behind, 30–32
marketing and, 97–103
moral imperative around, 193–195
moral reframing and, 104–107
moral values and, 104–106
photographs and, 48–50
for politics, 129–134
politics and, 8–9, 93–95, 103–105
potential of for good, 111–135
power of, 94–95
pros and cons of, 5–7, 107–109, 111–112
safeguards around, 175–180
for saving money, 112, 113–119
testing, 95–96
why we should care about, 5, 9–10
psychology, 15–27
breadcrumbs and, 59–78
digital footprints and, 15–16
online identities and, 29–58
personality traits and, 20–27
situations and, 85–88
social media language use and, 38–48
PTSD. See post-traumatic stress disorder (PTSD)
Pythagoras, 51

racism, 62–63
reciprocity, 149–151
reframing, moral, 104–107
regulations, 160–164, 167, 180
data co-ops and, 191
Replika, 123–124
risk mitigation, 10, 174–180. See also data ecosystems
Roe v. Wade, 133, 144
Roomba, 141–142

Salas, Daniel, 174–175

Salas, Esther, 174–175
Salas, Mark, 174–175
Samsung, 89, 121
SaverLife’s Race to 100, 117–119
saving money, 112, 113–119
schizophrenia, 5
Schwartz, Andrew, 32, 33, 203
Segalin, Cristina, 49
self-control, spending patterns and, 67–69
self-determination, 144–149, 194
data co-ops and, 184–191
data ecosystem design and, 166–174
self-disclosure, 29–30
selfies, deaths related to, 29
sensors
digital breadcrumbs created by, 62
GPS tracking and, 16
personalized health tracking and, 121–122
smartphones and, 57–58, 71–77
sexuality, Google searches and, 63
Shariff, Azim, 162
Sherman, Ryne, 85, 87
similarity principle, 147
Siri, 172–173
situations, psychology of, 85–88
sleep tracking, 83
smart glasses/lenses, 89
smartphones, 57–58
audio traces of situations and, 88
digital breadcrumbs created with, 62, 70–77
“it’s worth it” fallacy and, 140–142
mental health, GPS, and, 74–77
personalized health tracking and, 121–122
spending patterns and, 65–69
smartwatches, 121
Smith, Estelle, 127–128
Snickers, 82
Snowden, Edward, 143
social change, 90
social contract, 194–195
social media, 9
desire for attention and, 29–30
Facebook likes and personality, 34–38
facial characteristics and, 50–56
feedback functionality of, 30
“it’s worth it” fallacy and, 140–142
pictures in, 48–50
political polarization and, 129–134
socioeconomic standing and, 46–48
word choices in, 38–48
sociodemographic targeting, 97
socioeconomic standing, 46–48
Google searches and, 63–64
speech-recognition technology, 57, 165–166
spider-bots, 121–122
Stephens-Davidowitz, Seth, 62–63
Stillwell, David, 34–37
“The Story of Your Life Is in Your Face” (Hall), 52
stress
personality and, 83
tracking levels of, 122
suicide, 5–6
Google searches and, 64
preventing, 111
Sun Microsystems, 139
Surya, P., 29
Swash, 186
Swift, Taylor, 34
Switzerland, health care co-cops in, 186–189

Tackman, Allison, 44
Target, 97
taxes, on data collection, 176–177, 180
technology
pace of change in, 108, 162, 166, 180
potential benefits and dangers of, 6, 10, 129, 137–138
for privacy and service, 171–174
Teresa, Mother, 25
testosterone, 53
thrust, behavior change and, 167–168
trade unions, 190–191
transparency, 161–164
data co-ops and, 184–191
trial-and-error learning, 18–19
tribalism, political, 130
Trump, Donald, 8–9, 93–95, 148–149
TurboHire, 146

uncertainty, 162–163
University of Applied Sciences, Bern, 186–189

Véliz, Carissa, 138

village life, 1–4
digital, 4–8, 183–186
as model for the future, 181–191, 195
new rules for, 9–11
pros and cons of, 151
Vögisheim, Germany, 1–4, 181–182
voters and voting, 107

Walmart, 175
Ward, Adrian, 160
Whiley, Chris, 94
Willer, Rob, 105
Wilson, Timothy, 70
Woebot, 123–124, 127–128
word choice, 38–48
word clouds
Facebook likes, gender, and, 32, 33
Facebook status updates and, 40–48
income level and, 46–48
World Health Organization, 124
Wu, Tim, 179
Wysa, 123–125, 127

Youper, 123–124
Youyou Wu, 17

OceanofPDF.com
 ACKNOWLEDGMENTS

When I first started writing about my work in less academic terms, it

wasn’t with the intent of publishing a book. I simply wanted to help
my friends and family understand what I was doing thousands of
miles away from home, across the Atlantic. However, I soon fell in
love with the conversational style of communicating my ideas and
scientific findings. And I was fortunate to have an entire community
supporting me in bringing this effort to fruition.
It’s not just the people named here but also my wonderful
collaborators and the remarkable researchers whose work has shaped
my thinking and enriched so many passages throughout the book.
Academic life can be lonely at times, but writing Mindmasters
repeatedly reminded me of how lucky I am to be part of a community
dedicated to pushing the envelope of knowledge and making a
meaningful contribution to our collective wisdom.
Among just a few of those whose support carried me through this
journey and whose brilliance is enshrined in the pages of this book, I
want to first and foremost thank my agent, Leila Campoli. Without
you, there would be no Mindmasters. You believed in me and my
ideas from the beginning and gave me the confidence to move from
writing stories for my parents to reaching (I hope) a much larger
audience. I couldn’t have asked for a more talented, kind, and
supportive agent, and I am eternally grateful for your continued
guidance and encouragement.
I also want to thank my editor, Kevin Evers, and the entire team at
Harvard Business Review Press for helping me make this book a
reality. Kevin, your enthusiasm for the book was contagious, and your
keen eye for compelling stories has made Mindmasters far more
entertaining than it otherwise would have been.
None of the work described in Mindmasters would have been
possible without my wonderful students, colleagues, and mentors,
who coauthored many of the papers referenced in the book. You are
extraordinary scholars and true friends. The countless hours we spent
working together in kitchens, libraries, coffee shops, parks, and offices
(some nicer than others) encompass many of the most inspiring and
happiest moments of my life. A special thanks to my coauthors (in
alphabetical order): Ruth Appel, Erica Bailey, Maarten Bos, Moran
Cerf, Brian Croll, Tobias Ebert, Robert Farrokhnia, Brandon Freiberg,
Joe Gladstone, Friedrich Goetz, Sam Gosling, Gabriella Harari,
Michal Kosinski, Asher Lawson, Ashley Martin, Sandrine Mueller,
Gideon Nave, Heinrich Peters, Vess Popov, Jason Rentfrow, Andy
Schwartz, Clemens Stachl, David Stillwell, Jake Teeny, Sumer Vaid,
and Youyou Wu. You rock!
I am also incredibly grateful to my colleagues and students at
Columbia Business School. For the last seven years, you have become
family to me. I feel like the luckiest person in the world to have found
work I love in a place that feels like home. A special shout-out to all
the current and former members of the Computational Behavioral
Science Lab, particularly Minhee Kim, whose support as a lab
manager has been invaluable. I’m so honored to be part of your
journey to becoming the next generation of academic thought leaders
and educators. Thank you all!
Most importantly, I want to thank my family—my son, Ben, my
mom and dad, my sister and her family, and my in-laws—for their
unconditional love and support. You have always encouraged me to
reach for the stars, no matter how far away they seemed. I wouldn’t be
where I am today without you. I took leaps of faith because I knew
you had my back no matter what. I started writing this book for you,
and I hope seeing it in print will make you proud and remind you of
how much I love you.
Finally, I want to thank my partner in life and crime, Moran Cerf.
You are my biggest inspiration, closest confidant, and favorite
comedian. Your presence turns even the most mundane experience
into an unforgettable adventure. It has not escaped my notice that you
make me a better person every day.
The Story of Your Life Is in Your Face
Words and Music by Tom Hall
Copyright © 2009 Sony Music Publishing (US) Limited
All Rights Administered by Sony Music Publishing (US) LLC, 424
Church Street, Suite 1200, Nashville, TN 37219
International Copyright Secured All Rights Reserved
Reprinted by Permission of Hal Leonard LLC
OceanofPDF.com
 ABOUT THE AUTHOR

SANDRA MATZ was born and raised in a small village in the

southwest corner of Germany. After receiving her Bachelor of Science
in psychology at the University of Freiburg and her PhD from the
University of Cambridge, Matz joined Columbia Business School as a
professor of business at the age of twenty-seven. She currently serves
as the codirector of both the Leadership Lab and the Center for
Advanced Technology and Human Performance.
As a computational social scientist with a background in
psychology and computer science, Matz studies human behavior by
uncovering the hidden relationships between our digital footprints
and our inner mental lives. Her goal is to make data relatable and
help individuals, businesses, and policymakers use data in more-
effective and ethical ways.
Over the last ten years she has published more than fifty academic
papers in the world’s leading peer-reviewed journals, and her work
has frequently been covered by many major news outlets, including
the Economist, the New York Times, the Wall Street Journal, Forbes,
and Business Insider. In addition, Matz has shared her findings with
business leaders, policymakers, and the broader public through op-
eds, keynote speeches, podcasts, TV appearances, and consulting
work.
Matz has won numerous awards for her research and teaching,
including the SAGE Early Career Trajectory Award for social and
personality psychologists; Poets and Quants’ Best 40-Under-40 MBA
professors; Capital magazine’s Young Elite Top 40 Under 40; World
Frontiers Forum Young Pioneers; Pacific Standard magazine’s Top 30
Thinkers Under 30; and DataIQ’s 100 Most Influential People in
Data.
She lives in New York City with her husband, Moran Cerf, and her
son, Ben Cerf.
OceanofPDF.com