Introducing AI

(Music) 
At IBM, we define AI as 
anything that makes machines act more intelligently 
We like to think of AI as 
augmented intelligence. 
We believe that AI should not attempt 
to replace human experts, but rather extend human capabilities 
and accomplish tasks that neither humans 
nor machines could do on their own. 
The internet has given us access to more information, faster. 
Distributed computing and IoT have led to massive amounts of data, 
and social networking has encouraged most of that data to be unstructured. 
With Augmented Intelligence, we are putting information 
that subject matter experts need at their fingertips, 
and backing it with evidence so they can make informed decisions. 
We want experts to scale their capabilities and let the machines do the time-consuming work. 
How do we define intelligence? 
Human beings have innate intelligence, defined as the intelligence that governs every activity in our body. 
This intelligence is what causes an oak tree to grow out of a little seed, and an elephant to form from a
single-celled organism. 
How does AI learn? The only innate intelligence machines have is what we give them. 
We provide machines the ability to examine examples and create machine learning models based on the
inputs and desired outputs. 
And we do this in different ways such as Supervised Learning, Unsupervised Learning, 
and Reinforcement Learning, about which you will learn in more detail in subsequent lessons. 
Based on strength, breadth, and application, AI can be described in different ways. 
Weak or Narrow AI is AI that is applied to a specific domain. For example, language translators, virtual
assistants, 
self-driving cars, AI-powered web searches, recommendation engines, and intelligent spam filters. 
Applied AI can perform specific tasks, but not learn new ones, making decisions based on programmed
algorithms, and training data. 
Strong AI or Generalized AI is AI that can interact and operate a wide variety of independent and unrelated
tasks. 
It can learn new tasks to solve new problems, and it does this by teaching itself new strategies. 
Generalized Intelligence is the combination of many AI strategies that learn from experience and can
perform at a human level of intelligence. 
Super AI or Conscious AI is AI with human-level consciousness, which would require it to be self-aware. 
Because we are not yet able to adequately define what consciousness is, it is unlikely that we will be able
to create a conscious AI in the near future. 
AI is the fusion of many fields of study. Computer science and electrical engineering determine how AI is
implemented in software and hardware. 
Mathematics and statistics determine viable models and measure performance. 
Because AI is modeled on how we believe the brain works, psychology and linguistics play an essential role
in understanding how AI might work. 
And philosophy provides guidance on intelligence and ethical considerations. 
While the science fiction version of AI may be a distant possibility, 
we already see more and more AI involved in the decisions we make every day. 
Over the years, AI has proven to be useful in different domains, impacting the lives of people and our
society in meaningful ways.

There's a lot of talk about artificial intelligence 

these days. How do you define or what does AI mean for you? There is a lot of talk and 
there's a lot of definitions for what artificial intelligence says. So one of them is about 
teaching the machines to learn, and act, and think as humans would. Another dimension is 
really about how do we get the machines to- how do we impart more of a cognitive capabilities 
on the machines and sensory capabilities. So 
it's about analyzing images and videos about natural language processing and understanding 
speech. It's about pattern recognition, and so on, and so forth. So the third axis 
is more around creating a technology that's able to, 
in some cases, replace what humans do. I'd like to think of this as augment what humans 
do. To me personally, the most important part of definition for artificial intelligence 
is about imparting the ability to think and learn on the machines. To me that's what defines 
artificial intelligence. AI is the application of computing to solve 
problems in an intelligent way using algorithms. So what is an intelligent way? Well, an intelligent 
way may be something that mimics human intelligence. Or it may be a purely computational approach 
and optimization approach but something that manipulates data in a way to 
get not obvious results out, I think, is what I would classify 
as being artificially intelligent. I would define AI as a tool that uses computer to 
complete a task automatically with very little to no human intervention. 
To me AI is really a complex series of layers of algorithms that 
do something with the information that's coming into it. 
Artificial intelligence is a set of technologies that 
allows us to extract knowledge from data. So it's any kind of system that learns or 
understands patterns within that data, and can identify them, and then reproduce them 
on new information. Artificial intelligence is not the kind of 
simulating human intelligence that people think it is. 
It's really not about intelligence at all. But I think another word that describes AI 
more accurately today is machine learning. The reason I say that is because machine learning 
technology is all about using essentially mathematics on computers in order to find 
patterns in data. Now this data can be structured or unstructured. 
The only difference between machine learning and 
the technologies that came before it is instead of us, as humans, having to manually hard 
code these patterns, and these conditions into computers. They're 
able to find these patterns on their own by using math. That's really the only difference 
here. So what I'd say artificial intelligence is, is it's a set of mathematical algorithms 
that enable us to have computers find, very deep and patterns that we may not have even 
known exist, without us having to hard code them manually. 
(Music)

Hi Tanmay, welcome and tell us how old are you? 

I'm 15 years old. Great. How did you get started in technology and AI? Sure. So I've been working 
with technology for over 10 years now. It all started back when I was around five years 
old because my dad used to work as a computer programmer, and watching him program almost 
all day were so fascinating to me, that I really wanted to find out more. I wanted 
to find out how computers could do really anything that they did? 
Whether that be displaying my name on the screen, or adding two numbers, or really anything 
of that sort, it was like magic to me at that age. So my 
dad introduced me to the world of programming, and I've been working with code. I submitted 
my first iOS application and more. But when I was around 10 years old, I started 
to feel like technology wasn't as fun as it used to be for me. 
Technology wasn't as exciting as it used to be for one simple reason, it's because technology 
was very rigid, you code something in and it immediately starts to become obsolete, 
it never adapts, it never changes new data, new users, new circumstances they create. 
But when I was 11 years old, I actually stumbled upon a documentary on IBM Watson playing and 
winning the Jeopardy game show back in 2011. So of course, that immediately fascinated 
me, as to how a computer can play Jeopardy, and so I went ahead do a little bit more research, 
found out that IBM provides Watson's individual components as APIs on the cloud. I did a little 
bit more research, started to create my very first sort of cognitive applications, and 
I also created tutorials on my YouTube channel, on how others can also leverage the IBM Watson 
APIs. So really ever since I was 11 years old, I've been working with machine learning 
technology through numerous different services like IBM Watson. 
That's awesome. So what does AI mean for you? Really what AI means to me, before I get to 
that, before I can explain what AI is to me. I think it's first important to understand 
really what others like to think of AI as. Now, a lot of people have this very sort of 
split, this sort of very sort of bipolar response to machine learning or AI, as people call 
it. Some people are like, yes, it's the greatest technology of all time, 
and some people are like, this will be the downfall of humanity. I'd say that neither 
of those responses would be correct. The reason I say that is because 
machine learning technology is technology. It's very advanced technology, it helps us 
do things that we never could have done before, but it's just that, it's technology. Artificial 
intelligence and machine learning is something that people have been working on mathematically 
since even before computers were a thing. Machine learning 
technology is not new at all, in fact, it existed the very fundamentals at least, for 
many decades before I was even born. But the thing is machine learning technology 
or at least, for example, the basic perceptron and these sorts of mathematical techniques 
have existed since even before computers or calculators became 
popular. So when we were creating these sorts of machine-learning concepts and AI, and we 
started to create literature and movies on the future of technology and computers, 
we barely had any idea of not only where technology would go in the future, but also what technology 
really is. Because of that, people have this very common 
misconception of artificial intelligence being the human mind within a computer, the human 
intelligence simulated wholly within a computer. But that couldn't 
be farther from the truth. Machine learning or AI is not simulating a human mind, but 
what it does try and do, is it tries to open up new doors for computers. 
It tries to enable computers to understand certain kinds of data that they couldn't have 
understood before. So for example, if you take a look at what 
we as humans are so special at, the fact that we can understand natural language, in fact, 
we are the only animal to have such a complex ability of being able to communicate in natural 
language, even if through something that I have not directly 
witnessed or seen or heard evidence for, I can still describe that imaginative concept 
to you, that is actually really wonderful. We're also great at understanding at raw auditory 
data. I mean, imagine your brain is actually taking vibrations of air molecules and converting 
that to thoughts, that's really amazing. We're also great at processing visual data, like 
when you look at someone's face, the fact that you can instantly recognize them. 
When you look at someone's eyes, you can tell exactly where they're looking, that's really 
an amazing ability. These are things that computers cannot do because they are fundamentally 
limited to mathematics. They can only understand numbers and mathematical operations. 
But by using machine learning technology, you can actually take these mathematics, and 
use them to understand patterns in vast amounts of both structured 
and unstructured human data. The only difference here is that before, we as humans would manually 
construct these patterns and these conditions, whereas now it's done automatically for us 
at least mostly automatically by techniques like gradient descent and calculus. So machine 
learning technology is more accurate term for what AI really is today 
and will be in the future. Of course, artificial intelligence isn't meant to replace us because 
on a fundamental level, it is a completely different thing than a 
human brain.

Impact and Examples of AI

[MUSIC] 
AI is here to stay, with the promise of transforming the way the world works. 
According to a study by PWC, $16 trillion of GDP will 
be added between now and 2030 on the basis of AI. 
This is a never before seen scale of economic impact, and it is not just 
in the IT industry, it impacts virtually every industry and aspect of our lives. 
AI means different things to different people. 
For a videogame designer, AI means writing the code that affects how bots play, 
and how the environment reacts to the player. 
For a screenwriter, AI means a character that acts like a human, 
with some trope of computer features mixed in. 
For a data scientist, AI is a way of exploring and 
classifying data to meet specific goals. 
AI algorithms that learn by example are the reason we can talk to Watson, 
Alexa, Siri, Cortana, and Google Assistant, and they can talk back to us. 
The natural language processing and 
natural language generation capabilities of AI are not only enabling machines and 
humans to understand and interact with each other, but 
are creating new opportunities and new ways of doing business. 
Chatbots powered by natural language processing capabilities, are being used in 
healthcare to question patients and run basic diagnoses like real doctors. 
In education, they are providing students with 
easy to learn conversational interfaces and on-demand online tutors. 
Customer service chatbots are improving customer experience by resolving 
queries on the spot and freeing up agents time for conversations that add value. 
AI-powered advances in speech-to-text technology have made real time 
transcription a reality. 
Advances in speech synthesis are the reason companies are using AI-powered 
voice to enhance customer experience, and give their brand its unique voice. 
In the field of medicine, it's helping patients with Lou Gehrig's disease, 
for example, to regain their real voice in place of using a computerized voice. 
It is due to advances in AI that the field of computer vision has been able to 
surpass humans in tasks related to detecting and labeling objects. 
Computer vision is one of the reasons why cars can steer their way on streets and 
highways and avoid hitting obstacles. 
Computer vision algorithms detect facial features and images and 
compare them with databases of face profiles. 
This is what allows consumer devices to authenticate the identities of their 
owners through facial recognition, social media apps to detect and tag users, and 
law enforcement agencies to identify criminals in video feeds. 
Computer vision algorithms are helping automate tasks. 
Such as detecting cancerous moles in skin images or 
finding symptoms in x-ray and MRI scan. 
AI is impacting the quality of our lives on a daily basis. 
There's AI in our Netflix queue, our navigation apps, 
keeping spam out of our inboxes and reminding us of important events. 
AI is working behind the scenes 
monitoring our investments, detecting fraudulent transactions, 
identifying credit card fraud, and preventing financial crimes. 
AI is impacting healthcare in significant ways, 
by helping doctors arrive at more accurate preliminary diagnoses, 
reading medical imaging, finding appropriate clinical trials for patients. 
It is not just influencing patient outcomes But 
also making operational processes less expensive. 
AI has the potential to access enormous amounts of information, imitate humans, 
even specific humans, make life-changing recommendations about health and 
finances, correlate data that may invade privacy, and much more. 
[MUSIC]
Can you talk about some of the applications of AI? 
>> There's all kind of different applications, obviously, 
there's healthcare, there's finance. 
The one that's closest to my heart, of course, is robotics and automation. 
Where the AI technologies really help us to improve our abilities to perceive 
the environment around the robot and 
to make plans in unpredictable environments as they're changing. 
>> There's a great book out by an author, it was Kelvin Kelly, and 
he is an editor for the Wired magazine, he's written a great book 
about technologies that are going to be changing shaping our world, 
specifically 12 technologies. 
And he's got a fantastic definition in the book about specifically how AI is going 
to permeate our everyday life and it's all summarized in one excellent quote. 
So he says that the business cases for 
the next 10,000 startups are easy to predict, 
I have x and I will add AI to my x. 
The way I understand that is it's basically a notion 
that AI in one shape, way or form, in any shape or form, 
is going to permeate every aspect of human endeavor. 
Everything we do, everything we touch is going to be enhanced by AI. 
We have great benefits from taking any device, any machine, and 
make it just a little bit smarter. 
The benefit of that is just adding a bit of smarts to it, 
a bit of intelligence to it is exponential in its benefit. 
>> So we work a lot with some really fun applications of AI. 
We do a couple of different things in the lab that I run. 
We work on self-driving vehicles as one aspect, so autonomy for self-driving. 
Which requires a lot of AI for the vision systems, for the navigational 
intelligence, for the planning and control aspects of the car, we do that. 
And we also have a large research program in what are called collaborative robotics, 
or cobots. 
So robots that are designed to work in and around and with people. 
And that presents a lot of challenges, because we want the robots to 
act intelligently and to interface with humans in a way that is natural. 
And that requires understanding how people behave, which requires intelligence. 
In addition to those, there are a myriad of other applications, 
drug discovery, medical treatments for cancer and other diseases. 
So a bunch of extremely exciting applications. 
>> I mean I think the general use of AI so 
far has been taking large data sets, and making sense of them. 
And doing some sort of processing with that data in real time. 
That's what we've been doing, and 
that's what we've seen most effective, in terms of creating some sort 
of larger scale impact in healthcare beyond just having a siloed device. 
And we've been seeing that across the board, 
across the whole healthcare spectrum. 
>> We use AI all the time, and a lot of the times we're not even aware of it. 
We use AI every time we type in a search query on a search engine, 
every time we use our GPS. 
Or every time we use some kind of voice recognition system. 
>> I like to focus on a particular segment of AI, 
if that's okay, around computer vision. 
Because it's just particularly fascinating to me. 
Now, when we think of computer vision, they're looking at AI in ways to help 
augment, or to help automate or to help train computers to do something that's 
already very difficult to train humans to do. 
Like when it comes to the airport, trying to find weapons within luggage 
through the X-ray scanner, now that could be difficult to do. 
No matter how much you train someone that can be very difficult to identify. 
But with computer vision that can help to automate, help to augment, 
help to flag, certain X-ray images so that maybe even humans can just 
take a look at a filtered set of images, not all of the images right? 
So computer vision is very disruptive. 
And so there's many ways in which computer vision can really help to augment 
the capabilities of humans in lots of different industries. 
>> Now I mean applications of AI are really all around us. 
There's no limit to really what we're doing with artificial intelligence. 
When you do practically anything on any technology, you're most probably using 
some form of what we call machine learning or artificial intelligence. 
For example, when you check your email. 
Doing something as simple as checking your email. 
Spam filtering has been done for years with machine learning technology. 
More recently, 
Google came out with their features that enable you to do smart email compositions. 
So you can actually have text written for 
you on the fly as you're writing your email. 
Your subjects are automatically written as well, it'll recommend to you who you 
should be sending the email to, see if you missed someone. 
All of these things are powered by machine learning. 
But some of the main areas where I believe machine learning technology can make 
an impact are the fields of health care and education. 
[MUSIC]

Can you give 

Can you give some specific examples 
of applications of AI? 
Certainly. So we have 
a fairly large collaborative robotics program. 
So the cobots we work on 
are primarily targeted at the moment at 
manufacturing applications, 
manufacturing, warehousing, logistics, 
these types of applications where normally you may have 
a person doing a job that can be dull, 
it can be dangerous, 
and having robotic support or 
having a robot actually do 
the job may make it much safer, 
more efficient, and more effective overall. 
So we work on a lot of those types of applications, 
particularly where the robots are trying 
to interface directly with people, as I said. 
So the robot may 
help an individual to lift a heavy container, 
or help to move items on a stocking, 
on a shelf stocking purposes, 
so all these kinds of applications, 
where I think we'll see collaborative robots move first, 
and then hopefully one day 
and maybe into your home to help you with 
the laundry and dishes in the kitchen. Hopefully. 
For example, in oil and gas, there's a company, 
a pretty large oil and gas company 
called the Abu Dhabi National Oil Company, 
and one of the problems 
that any kind of oil company has to deal with is, 
where's the best place for them to drill for oil? 
So they have to find 
these rock samples of all these different places, 
for this place and in this place, and that place, 
and maybe hundreds of 
different places for them to drill oil. 
From these rock samples, 
now you have all these fine sheets 
of rock in maybe hundreds or thousands of them, 
and it's up to these oil companies to be able to classify 
these using they're trained and expert geologists. 
But to train geologists to properly 
classify these sheets of rock can be quite difficult, 
it could be time-consuming, 
could cost a lot of money as well. 
So one way to help augment 
the capabilities of humans 
is to be able to use computer vision, 
to classify these rocks samples to be able to 
identify which of these locations are 
the best to drill for oil? 
That's in oil and gas. 
Imagine before this, if there was a very, 
very rare form of cancer 
experienced by a doctor in Dubai, 
and if there were another case in New Zealand, 
how do you think they would have 
actually figured out that, "Hey, 
we're both dealing with this very 
rare case since we work together." 
That wouldn't have been possible in the past, 
but now with machine learning technology 
being able to aggregate knowledge from 
so many different sources into 
one centralized Cloud and understand it, 
and provide that information 
inaccessible, intuitive, implicit way. 
Now, that New Zealand doctor can actually go 
ahead and use this machine 
learning technique to say, "Hey, 
just a few days ago there 
was a doctor with a very similar 
case," even though it may not be the exact same thing. 
Sure. So we work with a number of 
startups and the number of enterprises, 
and I'll just bring a couple of examples. 
So what they like to talk quite a bit about 
is company out in California called Echo Devices. 
What they've done is they've taken 
a simple device which is stethoscope, 
something we see around the neck of every physician, 
nurse, and the health care professional, 
and they taken that device and 
basically have transformed that into first, 
into a digital device by cutting the tube on stethoscope, 
inserting a digitizer into it that takes an analog sound, 
transforms it into a digital signal, 
amplifies it in the process, 
makes it a lot easier for people to hear, 
it's amplified sound, the sound of your heart, 
or your lungs working. 
But what it also allows us 
to do is that allows us to take 
the digital signal and sent it 
via Bluetooth to a smart phone. 
Once it's on a smart phone, 
they're able to graph it, 
which allows the physician to better understand, 
not just through audio data but 
through an actual graph of how your heart is working. 
But because the information 
is now captured in the digital world, 
it can now be central machine-learning algorithm, 
and that's what they do. 
A machine-learning algorithm can 
actually learn from that, 
apply your previous learnings from the human doctors, 
cardiologist, and now assist 
a physician who is using 
the device in their current diagnosis. 
So it basically not 
replacing a physician in any way, shape, or form, 
it is assistive technology which is taking the learnings 
of the previous generations of human cardiologist, 
and helping in the diagnosis in the current state. 
To me, that's a perfect example of taking the X, 
which is in this particular case as a stethoscope, 
and then adding AI to that X. I 
have a really nifty name for that, 
they call it Shazam for Heartbeats. 
(Music)

Can you talk about AI in action. 

To give an example of say 
machine learning in action today, 
how companies have actually implemented it, 
there's one example that I always love to go back to, 
and it is the example of Woodside Energy, 
a company in the Australia New Zealand region. 
Now originally, they actually 
contacted IBM because they wanted 
the IBM to be able to create 
essentially a system that can understand 
the different documents and 
the research that they're engineers come up with, 
and have Watson and understand that, 
and essentially replace some 
of the engineers on their team. 
IBM actually went ahead 
and build the application that worked to 
Watson was able to understand that unstructured content, 
but they never ended up replacing any of their engineers. 
Instead, they actually ended up hiring more engineers, 
because now they realized that two things. 
First of all, the barrier of 
entry for each engineer is now 
lower and knowledge can now be 
shared more effectively among the teams. 
Because now instead of research being written 
and put into an archive drawer 
where it's never seen again, 
Watson's ingesting that data, 
understanding it and providing it to whoever needs it, 
whoever Watson thinks would need that data. 
So if you imagine in these TV shows and 
in these movie scenes as well, you have sometimes, 
if someone's looking for a particular suspect in 
this particular traffic intersection or whatnot, 
if passed through this intersection, 
and there's of course some cameras around. 
So we have the security guard maybe, 
trying to look through hours and hours, 
dozens and hundreds of hours of footage, 
maybe at 10x speed and find 
that particular black SUV or that green car. 
Then as soon as they find it at 
the end of the episode or whatnot, 
then say aha, we found that person. 
But if you had some sort of computer vision algorithm 
running on this video for just the entire time, 
then you wouldn't have a need for some person to 
have to manually watch 
through hours and hours of footage. 
Our specific use case is 
actually triggering 
new neural pathways in the brain to form. 
As you can imagine, 
there's a lot of information that happens there between 
the connection of how 
your body functions and how your brain functions, 
and what parts of the brain are damage, 
what parts of the brain aren't damaged, 
and how you're actually moving the person or how you're 
triggering certain things in the human body 
to happen in order for new neural pathways to form. 
So what we've done is actually, 
we've created massive data sets 
of information of how people move, 
and how that responds to different areas of the brain. 
Through that information, we're able to 
trigger specific movements with a robotic device, 
which in turn creates 
these neural pathways to form in the brain, 
and therefore recovering the person 
who suffered a neurological trauma. 
(Music)

I remember that morning going to the lab and I was thinking this is it, 
this is the last Jeopardy game. It became real to me when 
the music played and Johnny Gilbert said from IBM Research in Yorktown Heights, New York, 
this is Jeopardy and I just went. Hear it is one day. 
This is the culmination of all this work. To be honest with you I was emotional. 
Watson. What is Shoe? 
You are right. We actually took the lead. We were ahead of 
them, but then we start getting some questions wrong. 
Watson? What is Leg? 
No, I'm sorry. I can't accept that. What is 1920's? 
No. What is Chic? 
No, sorry, Brad. What is Class? 
Class. You got it. Watson. 
What is Sauron? Sauron is right and that puts 
you into a tie for the lead with Brad. The double Jeopardy round of 
the first game I thought was phenomenal. Watson went on a terror. 
Watson, who is Franz List? You are right. 
What is Violin? Who is the Church Lady? 
Yes. Watson. What is Narcolepsy? 
You are right and with that you move to $36,681. Now, we come to Watson. Who is Bram Stoker 
and the wager, hello $17,973 and a two day total of $77,147. 
We won Jeopardy. They are very justifiably proud of what they've 
done. I would've thought that technology 
like this was years away but it's here now. I have the bruised ego 
to prove it. I think we saw something important today. 
Wow, wait a second. This is history. The 60th Annual Grammy Awards, 
powered by IBM Watson. There's a tremendous amount of unstructured 
data that we process on Grammy Sunday. 
Our partnership with the recording Academy really is focused on 
helping them with some of their workflows for their digital production. 
My content team is responsible not only for taking all this raw stuff that's 
coming in, but curating it and publishing it. 
You're talking about five hours of red carpet coverage with 5,000 artists, 
making that trip down the carpet with a 100,000 photos being shot. 
For the last five hours, Watson has been using AI to analyze the colors, 
patterns, and silhouettes of every single outfit that has passed through. 
So we've been able to see all the dominant styles and 
compare them to Grammy shows in the past. Watson's also analyzing the emotions 
of Grammy nominated song lyrics over the last 60 years. 
Get this, it can actually identify the emotional themes in 
music and categorize them as joy, sadness, and everything else in between. It's 
very cool. Fantasy sports are an incredibly 
important and fun way that we serve sports fans. 
Our fantasy games drive tremendous consumption across ESPN digital 
properties, and they drive tune-in to our events and studio 
shows. But our users have a lot of 
different ways they can spend their time. So we have to continuously improve our game 
so they choose to spend that time with us. This year, ESPN teamed up with 
IBM to add a powerful new feature to their fantasy football platform. 
Fantasy football generates a huge volume of content - 
articles, blogs, videos, podcasts. We call it unstructured data - 
data that doesn't fit neatly into spreadsheets or databases. 
Watson was built to analyze that kind of information and turn it into usable insights. 
We train Watson on millions of fantasy football stories, 
blog posts, and videos. We taught it to develop a scoring 
range for thousands of players, their upsides and their downsides, 
and we taught it to estimate the chances of player will 
exceed their upside or fall below the downside. Watson even assesses a player's media buzz 
and their likelihood to play. This is a big win for our fantasy football 
players. It's one more tool to help them decide which 
running back or QB to start each week. It's a great complement to 
the award-winning analysts our fans rely on. As with any Machine Learning, 
the system gets smarter all the time. That means the insights are better, 
which means are you just can make better decisions and have a better chance to 
win their matchup every week. The more successful our fantasy players are, 
the more time they'll spend with us. The ESPN and IBM partnership 
is a great vehicle to demonstrate the power of enterprise-grade AI to millions 
of people, and it's not hard to see how 
the same technology applies to real life. There are thousands of business scenarios 
where you're assessing value and making trade-offs. 
This is what the future of decision-making is going to look like. 
Man and machine working together, assessing risk and reward, 
working through difficult decisions. This is the same technology 
IBM uses to help doctors mine millions of pages of medical research and 
investment banks fund market moving insights. 
(Music)
Lesson Summary
In this lesson, you have learned:
AI-powered applications are creating an impact in diverse areas such as Healthcare, Education,
Transcription, Law Enforcement, Customer Service, Mobile and Social Media Apps, Financial Fraud
Prevention, Patient Diagnoses, Clinical Trials, and more.

Some of these applications include:

 Robotics and Automation, where AI is making it possible for robots to perceive unpredictable
environments around them in order to decide on the next steps.
 Airport Security, where AI is making it possible for X-ray scanners to flag images that may look
suspicious.
 Oil and Gas, where AI is helping companies analyze and classify thousands of rock samples to help
identify the best locations to drill for oil?
Some famous applications of AI from IBM include:

 Watson playing Jeopardy to win against two of its greatest champions, Ken Jennings and Brad
Rutter.
 Watson teaming up with the Academy to deliver an amplified Grammy experience for millions of
fans.
 Watson collaborating with ESPN to serve 10 million users of the ESPN Fantasy App sharing insights
that help them make better decisions to win their weekly matchups.

To learn more about the applications of AI from IBM:

 Expert Insights: AI fast forwards video for sports highlights
 IBM Watson creates the first movie trailer for 20th Century Fox
 USTA uses IBM Watson to enhance player performance

Week 2
1.Cognitive Computing (Perception, Learning,
Reasoning)
 
AI is at the forefront of a new era of computing, Cognitive Computing. 
It's a radically new kind of computing, very different from the programmable systems 
that preceded it, as different as those systems 
were from the tabulating machines of a century ago. 
Conventional computing solutions, based on the mathematical principles that emanate from 
the 1940's, are programmed based on rules and logic 
intended to derive mathematically precise answers, 
often following a rigid decision tree approach. But with today's wealth of big data and 
the need for more complex evidence-based decisions, such a rigid approach often breaks or 
fails to keep up with available information. Cognitive Computing enables people to 
create a profoundly new kind of value, finding answers and insights 
locked away in volumes of data. Whether we consider a doctor diagnosing a 
patient, a wealth manager advising 
a client on their retirement portfolio, or even a chef creating a new recipe, 
they need new approaches to put into context the volume of 
information they deal with on a daily basis in order to derive value from 
it. These processes serve to enhance human expertise. 
Cognitive Computing mirrors some of the key cognitive elements of human expertise, 
systems that reason about problems like a human does. 
When we as humans seek to understand something and to make a decision, 
we go through four key steps. First, we observe visible phenomena 
and bodies of evidence. Second, we draw on what we know to interpret 
what we are seeing to generate hypotheses about what 
it means. Third, we evaluate which hypotheses are right 
or wrong. Finally, we decide, choosing 
the option that seems best and acting accordingly. Just as humans become experts by going through 
the process of observation, evaluation, and decision-making, 
cognitive systems use similar processes to reason about the information they read, 
and they can do this at massive speed and scale. 
Unlike conventional computing solutions, which can only handle 
neatly organized structured data such as what is stored in a database, 
cognitive computing solutions can understand unstructured data, 
which is 80 percent of data today. All of the information that is produced primarily 
by humans for other humans to consume. This includes everything from literature, 
articles, research reports to blogs, posts, and tweets. 
While structured data is governed by well-defined fields that 
contain well-specified information, cognitive systems rely on natural language, 
which is governed by rules of grammar, context, and culture. 
It is implicit, ambiguous, complex, and a challenge to process. 
While all human language is difficult to parse, certain idioms can be particularly challenging. 
In English for instance, we can feel blue because it's raining cats 
and dogs, while we're filling in a form, 
someone asked us to fill out. Cognitive systems read and interpret text 
like a person. They do this by breaking down 
a sentence grammatically, relationally, and structurally, discerning meaning 
from the semantics of the written material. Cognitive systems understand context. 
This is very different from simple speech recognition, 
which is how a computer translates human speech into a set of words. 
Cognitive systems try to understand the real intent of the users language, 
and use that understanding to draw inferences through 
a broad array of linguistic models and algorithms. Cognitive systems learn, adapt, 
and keep getting smarter. They do this by learning from their interactions 
with us, and from their own successes and failures, 
just like humans do.
In this lesson, you have learned:

Cognitive computing systems differ from conventional computing systems in that they can:

 Read and interpret unstructured data, understanding not just the meaning of words
but also the intent and context in which they are used.
 Reason about problems in a way that humans reason and make decisions.
 Learn over time from their interactions with humans and keep getting smarter.

Machine Learning, Deep Learning, Neural Networks

Terminology and Related Concepts
Before we deep dive into how AI works, 
and its various use cases and applications, 
let's differentiate some of the closely related terms and 
concepts of AI: artificial intelligence, 
machine learning, deep learning, and neural networks. 
These terms are sometimes used interchangeably, 
but they do not refer to the same thing. 
Artificial intelligence is a branch of 
computer science dealing with 
a simulation of intelligent behavior. 
AI systems will 
typically demonstrate behaviors associated 
with human intelligence such 
as planning, learning, reasoning, 
problem-solving, knowledge representation, 
perception, motion, 
and manipulation, and to 
a lesser extent social intelligence and creativity. 
Machine learning is a subset of 
AI that uses computer algorithms 
to analyze data and make 
intelligent decisions based on what it has learned, 
without being explicitly programmed. 
Machine learning algorithms are trained with 
large sets of data and they learn from examples. 
They do not follow rules-based algorithms. 
Machine learning is what enables 
machines to solve problems on their 
own and make accurate predictions 
using the provided data. 
Deep learning is a specialized subset 
of Machine Learning that 
uses layered neural networks 
to simulate human decision-making. 
Deep learning algorithms can label and 
categorize information and identify patterns. 
It is what enables AI systems 
to continuously learn on the job, 
and improve the quality and accuracy of 
results by determining whether decisions were correct. 
Artificial neural networks often referred to simply as 
neural networks take inspiration 
from biological neural networks, 
although they work quite a bit differently. 
A neural network in AI is 
a collection of small computing units called 
neurons that take incoming data 
and learn to make decisions over time. 
Neural networks are often layered deep and are the reason 
deep learning algorithms become more 
efficient as the datasets increase in volume, 
as opposed to other machine learning algorithms 
that may plateau as data increases. 
Now that you have a broad understanding of 
the differences between some key AI concepts, 
there is one more differentiation 
that is important to understand, 
that between artificial intelligence and data science. 
Data science is the process and method for extracting 
knowledge and insights from 
large volumes of disparate data. 
It's an interdisciplinary field involving mathematics, 
statistical analysis, data visualization, 
machine learning, and more. 
It's what makes it possible for us to 
appropriate information, see patterns, 
find meaning from large volumes of data, 
and use it to make decisions that drive business. 
Data Science can use many of 
the AI techniques to derive insight from data. 
For example, it could use machine learning algorithms and 
even deep learning models to extract 
meaning and draw inferences from data. 
There is some intersection between AI and data science, 
but one is not a subset of the other. 
Rather, data science is a broad term that 
encompasses the entire data processing methodology. 
Well, AI includes everything that allows computers 
to learn how to solve 
problems and make intelligent decisions. 
Both AI and Data Science can involve the use of 
big data that is significantly large volumes of data. 
In the next few lessons, 
the terms machine learning, 
deep learning, and neural networks 
will be discussed in more detail. 
(Music)

Machine Learning
Machine Learning, a subset of AI, 
uses computer algorithms to analyze data and 
make intelligent decisions based on what it has learned. 
Instead of following rules-based algorithms, 
machine learning builds models to 
classify and make predictions from data. 
Let's understand this by exploring a problem 
we may be able to tackle with Machine Learning. 
What if we want to determine whether a heart can fail, 
is this something we can solve with Machine Learning? 
The answer is, Yes. 
Let's say we are given data such as 
beats per minute, body mass index, 
age, sex, and the result 
whether the heart has failed or not. 
With Machine Learning given this dataset, 
we are able to learn and create a model 
that given inputs, will predict results. 
So what is the difference between this and using 
statistical analysis to create an algorithm? 
An algorithm is a mathematical technique. 
With traditional programming, we take data and rules, 
and use these to develop 
an algorithm that will give us an answer. 
In the previous example, 
if we were using a traditional algorithm, 
we would take the data such as beats per minute and BMI, 
and use this data to create an algorithm that 
will determine whether the heart will fail or not. 
Essentially, it would be an if-then-else statement. 
When we submit inputs, 
we get answers based on 
what the algorithm we determined is, 
and this algorithm will not change. 
Machine Learning, on the other hand, 
takes data and answers and creates the algorithm. 
Instead of getting answers in the end, 
we already have the answers. 
What we get is a set of rules that 
determine what the machine learning model will be. 
The model determines the rules, 
and the if-then-else statement when it gets the inputs. 
Essentially, what the model does is determine 
what the parameters are in a traditional algorithm, 
and instead of deciding arbitrarily that 
beats per minute plus BMI equals a certain result, 
we use the model to determine what the logic will be. 
This model, unlike a traditional algorithm, 
can be continuously trained and 
be used in the future to predict values. 
Machine Learning relies on defining behavioral rules by 
examining and comparing large datasets 
to find common patterns. 
For instance, we can provide 
a machine learning program with 
a large volume of pictures of 
birds and train the model to return 
the label "bird" whenever 
it has provided a picture of a bird. 
We can also create a label for "cat" 
and provide pictures of cats to train on. 
When the machine model is shown 
a picture of a cat or a bird, 
it will label the picture with some level of confidence. 
This type of Machine Learning 
is called Supervised Learning, 
where an algorithm is trained on human-labeled data. 
The more samples you provide 
a supervised learning algorithm, 
the more precise it becomes in classifying new data. 
Unsupervised Learning, another type of machine language, 
relies on giving the algorithm 
unlabeled data and letting it find patterns by itself. 
You provide the input but not labels, 
and let the machine infer qualities that 
algorithm ingests unlabeled data, 
draws inferences, and finds patterns. 
This type of learning can be useful for clustering data, 
where data is grouped according to how similar it is 
to its neighbors and dissimilar to everything else. 
Once the data is clustered, 
different techniques can be used to 
explore that data and look for patterns. 
For instance, you provide 
a machine learning algorithm with 
a constant stream of network traffic and 
let it independently learn the baseline, 
normal network activity, as well as the outlier and 
possibly malicious behavior happening on the network. 
The third type of machine learning algorithm, 
Reinforcement Learning, 
relies on providing a machine learning algorithm 
with a set of rules and constraints, 
and letting it learn how to achieve its goals. 
You define the state, 
the desired goal, allowed actions, and constraints. 
The algorithm figures out how to achieve 
the goal by trying different combinations 
of allowed actions, 
and is rewarded or punished 
depending on whether the decision was a good one. 
The algorithm tries its best to maximize 
its rewards within the constraints provided. 
You could use Reinforcement Learning to teach 
a machine to play chess or navigate an obstacle course. 
A machine learning model is a algorithm used 
to find patterns in the data without 
the programmer having to 
explicitly program these patterns.

Machine Learning Techniques and Training

Machine Learning is a broad field and we can split it up into three different categories, 
Supervised Learning, Unsupervised Learning, and Reinforcement Learning. 
There are many different tasks we can solve with these. 
Supervised Learning refers to when we have class labels in 
the dataset and we use these to build the classification model. 
What this means is when we receive data, it has labels that say what the data represents. 
In a previous example, we had a table with labels such as age or 
sex. With Unsupervised Learning, we don't have 
class labels and we must discover class labels from unstructured 
data. This could involve things such as 
deep learning looking at pictures to train models. 
Things like this are typically done with something called clustering. 
Reinforcement Learning is a different subset, and what this does is it uses a reward function 
to penalize bad actions or reward good actions. 
Breaking down Supervised Learning, we can split it up into three categories, 
Regression, Classification and Neural Networks. Regression models are built by 
looking at the relationships between features x and the result 
y where y is a continuous variable. Essentially, Regression estimates continuous 
values. Neural Networks refer to structures 
that imitate the structure of the human brain. Classification on the other hand, 
focuses on discrete values it identifies. We can assign discrete class labels 
y based on many input features x. In a previous example, 
given a set of features x, like beats per minute, 
body mass index, age and sex, the algorithm classifies the output y 
as two categories, True or False, predicting whether the heart will fail or 
not. In other Classification models, 
we can classify results into more than two categories. 
For example, predicting whether a recipe is for an Indian, 
Chinese, Japanese, or Thai dish. Some forms of classification include decision 
trees, support vector machines, 
logistic regression, and random forests. With Classification, we can 
extract features from the data. The features in this example would 
be beats per minute or age. Features are distinctive properties 
of input patterns that help in determining the output categories 
or classes of output. Each column is a feature and each row is a 
data point. Classification is the process of 
predicting the class of given data points. Our classifier uses some training data to 
understand how given input variables relate to that class. 
What exactly do we mean by training? Training refers to using 
a learning algorithm to determine and develop the parameters of your model. 
While there are many algorithms to do this, in layman's terms, 
if you're training a model to predict whether the heart will fail or not, 
that is True or False values, you will be showing the algorithm 
some real-life data labeled True, then showing the algorithm again, 
some data labeled False, and you will be repeating this process 
with data having True or False values, that is whether the heart actually failed 
or not. The algorithm modifies internal values 
until it has learned to tell from data that indicates heart failure that 
is True or not, that is False. With Machine Learning, we typically take 
a dataset and split it into three sets, Training, Validation and Test sets. 
The Training subset is the data used to train the algorithm. 
The Validation subset is used to validate our results and fine-tune the algorithm's 
parameters. The Testing data is the data the model has 
never seen before and used to evaluate how good our model 
is. We can then indicate how good 
the model is using terms like, accuracy, precision and recall.

Deep Learning
While Machine Learning is a subset of Artificial Intelligence, 
Deep Learning is a specialized subset of Machine Learning. 
Deep Learning layers algorithms to create a Neural Network, 
an artificial replication of the structure and functionality of the brain, 
enabling AI systems to continuously learn on the job and improve the quality and 
accuracy of results. 
This is what enables these systems to learn from unstructured data such as 
photos, videos, and audio files. 
Deep Learning, for example, 
enables natural language understanding capabilities of AI systems, 
and allows them to work out the context and intent of what is being conveyed. 
Deep learning algorithms do not directly map input to output. 
Instead, they rely on several layers of processing units. 
Each layer passes its output to the next layer, which processes it and 
passes it to the next. 
The many layers is why it’s called deep learning. 
When creating deep learning algorithms, developers and 
engineers configure the number of layers and the type of functions that connect 
the outputs of each layer to the inputs of the next. 
Then they train the model by providing it with lots of annotated examples. 
For instance, you give a deep learning algorithm thousands of images and 
labels that correspond to the content of each image. 
The algorithm will run the those examples through its layered neural network, 
and adjust the weights of the variables in each layer of the neural network to be 
able to detect the common patterns that define the images with similar labels. 
Deep Learning fixes one of the major problems present in 
older generations of learning algorithms. 
While the efficiency and 
performance of machine learning algorithms plateau as the datasets grow, 
deep learning algorithms continue to improve as they are fed more data. 
Deep Learning has proven to be very efficient at various tasks, 
including image captioning, voice recognition and transcription, 
facial recognition, medical imaging, and language translation. 
Deep Learning is also one of the main components of driverless cars. 
[MUSIC]

Neural Networks
An artificial neural network is 
a collection of smaller units called neurons, 
which are computing units modeled on the way 
the human brain processes information. 
Artificial neural networks borrow 
some ideas from the biological 
neural network of the brain, 
in order to approximate some of its processing results. 
These units or neurons take incoming data like 
the biological neural networks 
and learn to make decisions over time. 
Neural networks learn through 
a process called backpropagation. 
Backpropagation uses a set of training data that 
match known inputs to desired outputs. 
First, the inputs are plugged 
into the network and outputs are determined. 
Then, an error function determines how far 
the given output is from the desired output. 
Finally, adjustments are made in order to reduce errors. 
A collection of neurons is called a layer, 
and a layer takes in an input and provides an output. 
Any neural network will have 
one input layer and one output layer. 
It will also have one or more hidden layers which 
simulate the types of activity 
that goes on in the human brain. 
Hidden layers take in a set of weighted inputs 
and produce an output through an activation function. 
A neural network having 
more than one hidden layer is 
referred to as a deep neural network. 
Perceptrons are the simplest 
and oldest types of neural networks. 
They are single-layered neural networks consisting 
of input nodes connected directly to an output node. 
Input layers forward the input values to the next layer, 
by means of multiplying by 
a weight and summing the results. 
Hidden layers receive input from other nodes 
and forward their output to other nodes. 
Hidden and output nodes have a property called bias, 
which is a special type of weight that applies to 
a node after the other inputs are considered. 
Finally, an activation function 
determines how a node responds to its inputs. 
The function is run against 
the sum of the inputs and bias, 
and then the result is forwarded as an output. 
Activation functions can take different forms, 
and choosing them is a critical component 
to the success of a neural network. 
Convolutional neural networks or CNNs are 
multilayer neural networks that take 
inspiration from the animal visual cortex. 
CNNs are useful in applications such as image processing, 
video recognition, and natural language processing. 
A convolution is a mathematical operation, 
where a function is applied to another function 
and the result is a mixture of the two functions. 
Convolutions are good at detecting 
simple structures in an image, 
and putting those simple features 
together to construct more complex features. 
In a convolutional network, 
this process occurs over a series of layers, 
each of which conducts 
a convolution on the output of the previous layer. 
CNNs are adept at building 
complex features from less complex ones. 
Recurrent neural networks or RNNs, 
are recurrent because they perform 
the same task for every element of a sequence, 
with prior outputs feeding subsequent stage inputs. 
In a general neural network, 
an input is processed through 
a number of layers and an output is 
produced with an assumption that 
the two successive inputs are independent of each other, 
but that may not hold true in certain scenarios. 
For example, when we need to consider 
the context in which a word has been spoken, 
in such scenarios, dependence on 
previous observations has to 
be considered to produce the output. 
RNNs can make use of information in long sequences, 
each layer of the network representing 
the observation at a certain time. 
(Music)

Lesson Summary
In this lesson, you have learned:
 Machine Learning, a subset of AI, uses computer algorithms to analyze data and make intelligent
decisions based on what it has learned. The three main categories of machine learning algorithms
include Supervised Learning, Unsupervised Learning, and Reinforcement learning.
 Deep Learning, a specialized subset of Machine Learning, layers algorithms to create a neural
network enabling AI systems to learn from unstructured data and continue learning on the job.
 Neural Networks, a collection of computing units modeled on biological neurons, take incoming
data and learn to make decisions over time. The different types of neural networks include
Perceptrons, Convolutional Neural Networks or CNNs, and Recurrent Neural Networks or RNNs.

In the Machine Learning Techniques and Training topic, you have learned:
 Supervised Learning is when we have class labels in the data set and use these to build the
classification model.
 Supervised Learning is split into three categories – Regression, Classification, and Neural
Networks.
 Machine learning algorithms are trained using data sets split into training data, validation data,
and test data.

[Optional] To learn more about Machine Learning, Deep Learning, and Neural Networks, read
these articles:
 Models for Machine Learning
 Applications of Deep Learning
 A Neural Networks deep dive
AI Application Areas

Key Fields of Application in AI

So can you talk 
about the different areas 
or categories of artificial intelligence? 
Now, there are lots of different fields that AI works in. 
But if I were to on 
a very very high level group some of 
the major areas where artificial intelligence is applied, 
I'd like to start off with natural language. 
Because natural language is, 
I'd say, the most complex data 
for machine learning to work with. 
If you see all sorts of data, 
whether that be a sequence to genome, 
whether that be audio, 
whether that be images. 
There's some sort of discernible pattern. 
There's some sort of yes, 
this is what a car sounds like or yes, 
this is what human voice sounds like. 
But natural language is fundamentally, 
a very human task. 
It's very human data source. 
We as humans invented it for humans to understand. 
If I were to, for example, 
give you a book title, there's 
actually a very very famous book, 
and the title of the book is there 
are two mistakes in the the title of this book. 
Now, there's actually only one mistake, the two the's. 
The human brain doesn't realize that. 
What's the second mistake? 
That there was only one mistake. 
So this is a sort of natural language complexity 
that's involved here. 
Humans we don't view natural language literally. 
We view it conceptually. 
If I were to write a three instead of an E, 
you will understand it because we don't 
mean the three in a literal sense. 
We mean that in a symbolic sense to represent the concept 
of E and you can 
contextualize that three to figure out that, "Yeah. 
It means in E" and not an actual three. 
These are things that computers aren't capable of. 
So natural languages that number 
one field that I'm most 
interested in when it comes to machine learning. 
Second, I'd say the most popular would be visual. 
Visual data understanding, computer vision. 
Because it enables us to do so many things. 
As humans, our primary sense is vision. 
In fact, a vast majority of 
your brain's processing power at any given moment, 
goes to understanding what it is that you're seeing. 
Whether it be a person's face, 
or whether it be a computer or some texts, 
or anything of that sort. 
Third, I would say audio-based data. 
So text-to-speech, 
speech-to-text these are very very complex. 
The reason it's complex is because it 
combines a lot of challenges into one. 
First of all, you've got to support many languages. 
You can't just support English and call it a day. 
You've got to support other languages. 
You've got to support other demographics. 
Another challenge is that even within languages, 
there are absolutely infinite number 
of ways that any human could represent a language. 
Everyone's going to have a different accent. 
Everyone's going to have 
a different way of pronouncing certain words. 
There's no standardized way that every human will 
pronounce ice cube exactly 
like ice cube. That doesn't exist. 
If you take a look at another challenge, 
it's that audio data is 
fundamentally very very difficult to work with. 
Because the thing is, audio data 
exists in the natural world. What is audio? 
It's vibrations of air molecules, 
and vibrations of air molecules are fast. 
Audio is recorded at overpay say 44 kilohertz. 
That's a lot of data, 
44,000 data points every single second. 
There are usually only 44,000 data points in 
an individual low-resolution image. 
So of course, there are lots of 
challenges to work around when it comes to audio. 
But companies like IBM, Google, 
Microsoft have actually worked around 
these challenges and they're working 
towards creating different services 
to make it easier for developers. 
So again, on a very very high level, 
there's natural language understanding, 
there's computer vision, 
there's audio data and of course, 
there's the traditional set 
of tabular data understanding. 
Which is essentially, structured data understanding. 
(Music)

Natural Language Processing, Speech,

Computer Vision
Some of the most common application areas of AI include natural language processing, 
speech, and computer vision. Now, let's look at each of these in turn. 
Humans have the most advanced method of communication which is known as natural language. 
While humans can use computers to send voice and text messages to each other, 
computers do not innately know how to process natural language. 
Natural language processing is a subset of artificial intelligence that enables 
computers to understand the meaning of human language. 
Natural language processing uses machine learning and 
deep learning algorithms to discern a word's semantic meaning. 
It does this by deconstructing sentences grammatically, relationally, and structurally and 
understanding the context of use. For instance, based on the context of a conversation, 
NLP can determine if the word "Cloud" is a reference to cloud 
computing or the mass of condensed water vapor floating in the sky. 
NLP systems might also be able to understand intent and emotion, 
such as whether you're asking a question out of 
frustration, confusion, or irritation. Understanding the real intent of the user's 
language, NLP systems draw inferences through 
a broad array of linguistic models and algorithms. Natural language processing is broken down 
into many subcategories related to audio and visual 
tasks. For computers to communicate in natural language, 
they need to be able to convert speech into text, 
so communication is more natural and easy to process. 
They also need to be able to convert text-to-speech, so users can interact with computers 
without the requirement to stare at a screen. The older iterations of speech-to-text 
technology require programmers to go through tedious process of discovering and 
codifying the rules of classifying and 
converting voice samples into text. With neural networks, instead of coding the 
rules, you provide voice samples and their corresponding 
text. The neural network finds 
the common patterns among the pronunciation of words 
and then learns to map new voice recordings to their corresponding 
texts. These advances in speech-to-text technology 
are the reason we have real time transcription. Google uses AI-powered speech-to-text 
in there Call Screen feature to handle scam calls and show you 
the text of the person speaking in real time. YouTube uses this to provide automatic closed 
captioning. The flip side of speech-to-text is 
text-to-speech also known as speech synthesis. In the past, the creation of 
a voice model required hundreds of hours of coding. 
Now, with the help of neural networks, synthesizing human voice has become possible. 
First, a neural network ingests numerous samples of 
a person's voice until it can tell whether a new voice sample belongs to the same person. 
Then, a second neural network generates audio data and runs it 
through the first network to see if it validates it as belonging to the subject. 
If it does not, the generator corrects its sample and reruns it through the classifier. 
The two networks repeat the process until they generate samples that sound natural. 
Companies use AI-powered voice synthesis to enhance 
customer experience and give their brands their unique voice. 
In the medical field, this technology is helping ALS patients regain 
their true voice instead of using a computerized voice. 
The field of computer vision focuses on replicating parts of the complexity 
of the human visual system, and enabling computers to identify and 
process objects in images and videos, in the same way humans do. 
Computer vision is one of the technologies that enables 
the digital world to interact with the physical world. 
The field of computer vision has taken great leaps in recent years and 
surpasses humans in tasks related to detecting and labeling objects, 
thanks to advances in deep learning and neural networks. 
This technology enables self-driving cars to make sense of their surroundings. 
It plays a vital role in facial recognition applications allowing 
computers to match images of people's faces to their identities. 
It also plays a crucial role in augmented and mixed reality. 
The technology that allows computing devices such as smartphones, tablets, 
and smart glasses to overlay and embed virtual objects on real-world imagery. 
Online photo libraries like Google Photos, use computer vision to detect objects and 
classify images by the type of content they contain. 
(Music)

Self Driving Cars

Can you tell us a little bit about the work you're doing with self-driving cars. 
>> I've been working on self-driving cars for the last few years. 
It's a domain that's exploded, obviously, in interest since 
early competitions back in the 2005 domain. 
And what we've been working on really is putting together our 
own self-driving vehicle that was able to drive on public roads in 
the regional Waterloo last August. 
With the self-driving cars area, 
one of our key research domains is in 3D object detection. 
So this remains a challenging task for algorithms to perform automatically. 
Trying to identify every vehicle, every pedestrian, 
every sign that's in a driving environment. 
So that the vehicle can make the correct decisions about how it should move and 
interact with those vehicles. 
And so we work extensively on how we take in laser data and 
vision data and radar data. 
And then fuse that into a complete view of the world around the vehicle. 
>> When we think of computer vision, 
we usually think immediately of self-driving cars, and why is that? 
Well, it's because it's hard to pay attention when driving on the road, right? 
You can't both be looking at your smartphone and 
also be looking at the road at the same time. 
Of course, it's sometimes hard to predict what people are going to be doing on 
the street, as well. 
When they're crossing the street with their bike or skateboard, or whatnot. 
So it's great when we have some sort of camera or sensor that can help us 
detect these things and prevent accidents before they could potentially occur. 
And that's one of the limitations of human vision, is attention, is visual attention. 
So I could be looking at you, Rav, but 
behind you could be this delicious slice of pizza. 
But I can only pay attention to one or 
just some limited number of things at a time. 
But I can't attend to everything in my visual field all at once at the same time 
like a camera could. 
Or like how computer vision could potentially do so. 
And so that's one of the great things that cameras and computer vision is good for. 
Helping us pay attention to the whole world around us without having us to look 
around and make sure that we're paying attention to everything. 
And that's just in self-driving cars, so I think we all kind of have a good sense of 
how AI and computer vision shapes the driving and transportation industry. 
>> Well, self-driving cars are certainly the future. 
And there's tremendous interest right now in self-driving vehicles. 
In part because of their potential to really change the way 
our society works and operates. 
I'm very excited about being able to get into a self-driving car and 
read or sit on the phone on the way to work. 
Instead of having to pilot through Toronto traffic. 
So I think they represent a really exciting step forward, but 
there's still lots to do. 
We still have lots of 
interesting challenges to solve in the self-driving space. 
Before we have really robust and safe cars 
that are able to drive themselves 100% of the time autonomously on our roads.
Play video starting at :3:18 and follow transcript3:18
>> We've just launched our own self-driving car specialization 
on Coursera. 
And we'd be really happy to see students in this specialization also come and 
learn more about self-driving. 
It's a wonderful starting point, it gives you a really nice perspective 
on the different components of the self-driving software stack and 
how it actually works. 
So everywhere from how it perceives the environment, how it makes decisions and 
plans its way through that environment. 
To how it controls the vehicle and makes sure it executes those plans safely. 
So you'll get a nice broad sweep of all of those things from that specialization. 
And from there you then want to become really good and 
really deep in one particular area, if you want to work in this domain. 
Because again, there's so many layers behind this. 
There's so much foundational knowledge you need to start contributing that you 
can't go wrong. 
If you find something interesting, just go after it. 
And I am sure there'll be companies that'll need you for this. 
[MUSIC]

Lesson Summary
In this lesson, you have learned:
 Natural Language Processing (NLP) is a subset of artificial intelligence that enables computers to
understand the meaning of human language, including the intent and context of use.
 Speech-to-text enables machines to convert speech to text by identifying common patterns in the
different pronunciations of a word, mapping new voice samples to corresponding words.
 Speech Synthesis enables machines to create natural sounding voice models, including the voice
of particular individuals.
 Computer Vision enables machines to identify and differentiate objects in images the same way
humans do.
 Self-driving cars is an application of AI that can utilize NLP, speech, and most importantly,
computer vision.

To learn more about Natural Language Processing, read this article:

 A beginner's guide to Natural Language Processing

Week 3
Exploring Today's AI Concerns
Welcome to exploring today's AI concerns. 
In this video, you will learn about some of today's hot topics in AI. 
First, you will hear about why trustworthy AI is the hot topic in AI. 
Then, you will hear about how AI is used in facial recognition technologies, 
in hiring in marketing on social media and in healthcare. 
People frequently ask me what our current hot topics in AI and 
I will tell you that whatever I answer today is likely to be different 
next week or even by tomorrow. 
The world of AI is extremely dynamic, which is a good thing. 
It's an emerging technology with an amazing amount of possibilities and 
the potential to solve so many problems, so 
much faster than what we thought was before possible. 
Now, as we've seen in some cases it can have harmful consequences. 
And so, I would say that the hot topic in AI is how do we do this responsibly? 
And IBM has come up with five pillars to address this issue, 
kind of summarizing the idea of responsible AI. 
That is explain ability, transparency, robustness, privacy and fairness. 
And we can go into those topics in more depth, but 
I want to emphasize two things here and 
one is that this is not a one and done sport. 
If you're going to use AI, if we're going to put it into use in society, 
this is not something you just address at the beginning or at the end. 
This is something you have to address throughout the entire lifecycle of AI. 
These are questions you have to ask whether you're at the drawing board, 
whether you're designing the AI, you're training the AI or 
you have put it into use or you are the end user who's interacting with the AI. 
And so, those five pillars or 
things you want to constantly think about throughout the entire lifecycle of AI. 
And then second and I think even more importantly is this is a team sport, 
we all need to be aware of both the potential good and 
the potential harm that comes from AI. 
And encourage everybody to ask questions. 
Make room for people to be curious about how AI works and what it's doing. 
And with that I think we can really use it to address good problems and 
have some great results and mitigate any of the potential harm. 
So stay curious. 
In designing solutions around Artificial Intelligence, call it AI, 
facial recognition has become a permanent use case. 
There are really three typical examples of categories of models and 
algorithms that are being designed. Facial detection that is simply 
detecting whether it is a human face versus a or a dog or cat. 
This type of facial recognition happens without uniquely identifying who 
that face might belong to. 
In facial authentication, you might use this type of facial 
recognition to open up your iPhone or your android device. 
In this case, we provide a one-on-one authentication by 
comparing the features of a face image. 
What they previously stored, single up image, meaning that you are really only 
comparing the images with the distinct image of the owner of the iPhone or 
android device. Facial matching, 
in this case, we compare the image with a database of other images or photos. 
Just as different from the previous in that, the model is trying to 
determine a facial match of an individual against the database for 
images below it or photos belonging to other humans. 
There are many different examples of facial recognition. 
Many of them you have no doubt experienced in your day to day activity. 
Some have proven to be helpful while others have shown to be not so helpful and 
then there are others that have proven to be direct criminal in nature. 
Where certain demographics of people have been harmed because of the use of these 
facial recognition systems. 
We've seen facial recognitions and solutions in AI systems 
provide significant value in scenarios like navigating through an airport or 
going through security or security align. 
Or even using previous previous examples like the one we talked 
about earlier where facial recognition to unlock your iPhone or 
possibly to unlock your home or down lock the door in your automobile. 
These are all helpful uses of facial recognition technologies but 
there are also some clear examples and use cases that must be off-limits. 
These might include identifying a person and the crowd without the sole permission 
of that person or doing mass surveillance on a single or group of people. 
These types of uses of technology raises important privacy, 
civil rights, and human rights concerns. 
When used the wrong way by the wrong people in facial recognition 
technologies no doubt can be used to suppress, dissent, or 
infringe upon the rights of minorities or 
to simply just erase your basic expectations of having privacy. 
AI is being increasingly introduced into each stage of workforce 
progression, hiring, onboarding, career progression, 
including promotions and awards handling, attrition etc. 
Stock board hiring: 
Consider an organization that receives thousands of job applications. 
People applying for all kinds of jobs, front office, 
back office, seasonal, permanent. 
Instead of having large teams of people sit and 
sift through all these applications, AI helps you rank and 
prioritize applicants against targeted job openings, 
presenting a list of the top candidates to the hiring managers. 
AI solutions can process text in resumes and 
combine that with other structured data to help in decision making.
Play video starting at :6:43 and follow transcript6:43
Now, we need to be careful to have guardrails in place. 
We need to ensure the use of AI in hiring is not biased across 
sensitive attributes like age, general ethnicity and the like. 
Even when those attributes are not directly used by the AI but 
maybe creeping in, coming in from proxy attributes like zip code or 
type of job previously held. 
One of the hot topics in AI today is its application and 
marketing on social media. 
It has completely transformed how brands interact 
with their audiences on social media platforms like 
TikTok, LinkedIn, Twitter, Instagram, Facebook. 
AI today can create ads for you, it can create social media posts for you. 
It can help you target those ads appropriately. 
It can use sentiment analysis to identify new audiences for you. 
All of this drives incredible results for a marketeer. 
It improves the effectiveness of the marketing campaigns while dramatically 
reducing the cost of running those campaigns.
Play video starting at :8:6 and follow transcript8:06
Now, the same techniques and capabilities that AI produces for 
doing marketing on social media platforms also raises some ethical questions.
Play video starting at :8:21 and follow transcript8:21
The marketing is successful because of all of the data and 
social media platforms collect from their users. 
Ostensibly, this data is collected to deliver more 
personalized experiences for end users. 
It's not always explicit what data is being collected and 
if you are providing your consent for them to use as data. 
Now, the same techniques that are so effective for 
marketing campaigns for brands can also be applied for 
generating misinformation, conspiracy theories. 
Whether it's political or scientific misinformation and 
this has horrendous implications for our communities at large.
Play video starting at :9:15 and follow transcript9:15
This is why it is absolutely critical that all enterprises 
adhere to some clear principles around transparency, 
explain ability, trust, privacy in terms of how they use AI or 
build AI into their solutions into their platforms.
Play video starting at :9:40 and follow transcript9:40
The use of AI is increasing across all healthcare segments, 
healthcare providers, pairs, life sciences etc.
Play video starting at :9:50 and follow transcript9:50
Pair organizations are using AI and 
machine learning solutions that tap into claims data, 
often combining it with other data sets like social determinants of health. 
A top use case is disease prediction for coordinating care. 
For example, predicting who in the member population is likely to have 
an adverse condition, maybe like an ER visit in the next three months and 
then providing the right forms of intervention and prevention.
Play video starting at :10:22 and follow transcript10:22
Equitable care becomes very important in this context. 
We need to make sure the AI is not biased across sensitive attributes like age, 
gender, ethnicity, etc. 
Across all of these of course, conversational AI where virtual agents 
as well as systems that help humans better service the member population.
Play video starting at :10:45 and follow transcript10:45
That has become table stakes. 
Across all of these use cases of AI in health care, we see a few common things. 
Being able to unlock insights from the rich sets of data the organization owns, 
improving the member or patient experience, and 
having guardrails in place to ensure AI is trustworthy. 
[MUSIC]
Exploring AI and Ethics
Welcome to exploring AI and ethics. 
In this video, you will learn about what AI ethics is, and why it matters. 
You will find out, what makes AI ethics a socio-technical challenge, 
what it means to build and use AI ethically, 
and how organizations can put AI ethics into action. 
AI, Artificial Intelligence, is a very pervasive in everybody's life. 
Even if often we don't realize it, we use it when we use a credit 
card to buy something online, when we search something on a web, 
when we post or like or follow somebody on a social platform. 
And even when we drive with the navigation support, and 
the driver assistance capabilities of the car based on AI. 
This pervasiveness generates very fast a significant transformations in our life, 
and also in the structure and equilibrium of our society. 
This is why AI besides being a technical and scientific discipline, 
has also a very significant social impact. 
This raises a lot of ethical questions about how AI should be designed, 
developed, deployed, used, and also regulated. 
The social technical dimensions of AI requires efforts to identify 
all stakeholders, that go well beyond technical experts, and 
include also sociologists, philosophers, economists, policymakers. 
And all the communities that are impacted by the deployment of this technology. 
Inclusiveness is necessary in defining the ecosystem, 
in all the phases of the AI development and deployment, and 
also in the impact of AI in the deployment scenario. 
Without it, we risk of creating AI only for some, 
and leave many others behind in a disadvantaged position. 
Everybody needs to be involved in defining the vision of the future that we 
want to build using AI and other technology as a means and not as an end. 
To achieve this, appropriate guidelines are necessary to drive the creation and 
use of AI in the right direction. 
Technical tools are necessary and useful, but they should be complemented by 
principles guardrails, well-defined processes, and effective governance. 
We should not think that all these slows down innovation. 
Think about traffic rules, it may seem that traffic lights, 
precedents, and stop signals, and speed limits are slowing us down. 
However, without them, we will not drive faster, but 
actually we would drive much slower, because we would be always in 
a complete state of uncertainty about other vehicles and pedestrians. 
AI ethics identifies and 
addresses the socio-technical issues raised by this technology, 
and makes sure that the right kind of innovation is supported, 
and facilitated, so that the path to the future we want is faster. 
As our CEO at IBM States, "Trust is our license to operate." 
We've earned this trust through our policies, programs, 
partnerships and advocacy of the responsible use of technology. 
For over 100 years, IBM has been at the forefront of 
innovation that brings benefits to our clients and society. 
This approach most definitely applies to the development, use, and 
deployment of AI. 
Therefore, ethics should be embedded into the lifecycle of the design and 
development process. 
Ethical decision making is not just a technical problem solving approach. 
Rather an ethical, sociological, technical and 
human centered approach, should be embarked upon, 
based on principles, value standards, laws and benefits to society. 
So having this foundation is important, even necessary, but where to start? 
A good place to start is with a set of guiding principles, 
at IBM we call our principles, the principles of trust and transparency. 
of which there are three. 
The purpose of AI is to augment not replace human intelligence. 
Data and insights belong to their creator and 
new technology including AI systems must be transparent and explainable. 
This last principle is built upon our pillars, of which there are five. 
We just mentioned transparency which reinforces trust, 
by sharing the what, and the how, the AI is being used for. 
It must be explainable and also fair. 
So when it's properly calibrated, it can assist in making better choices. 
Should be robust, which means it should be secure, and 
as well as privacy preserving, safeguarding privacy and rights. 
We know having principles and pillars are not enough. 
We have an extensive set of tools and talented practitioners that can help 
diagnose, monitor, promote all of our pillars and continuous monitoring, 
to mitigate against drift and unintended consequences. 
The first step to putting AI ethics into action, just like with anything else, 
is about building understanding and awareness. 
This is about equipping your teams to think about AI ethics, what it means 
to put it into action, whatever solution you're building and deploying. 
Let's take an example, if you're building a learning solution and 
deploying that within a company, 
your HR team leader who is doing that should be thinking about, 
is this solution designed with users in mind? 
Have you co-created the solution with users? 
How does it enable equal access to opportunity to 
all the employees across diverse groups. 
A keen understanding of AI ethics, and 
reflecting on these issues continuously, 
is critical as a foundation to putting AI ethics into action. 
The second step in putting AI ethics into action, 
once you build that understanding and awareness and 
everybody is reflecting on this topic, is to put in place in a governance structure. 
And the critical point here is, 
it's a governance structure to scale AI ethics in action. 
It's not about doing it in one isolated instance in a market, 
or in a business unit, it's about a governance structure that works at scale. 
We talked about understanding and awareness as a foundation, second 
governance, which is the responsibility of leaders to put in place structures. 
Once you've got these two elements, the third step is operationaIizing. 
How do you make sure a developer, or a data scientist, or a vendor, 
who's in Malaysia or Poland knows how to put AI ethics into action? 
What does it mean for them? 
Right? 
It's one thing to put structures in place at the global level, but 
how do you make sure it's operationalized at scale in the markets and 
every user, every data scientist, every developer knows what they need to do? 
This is all about having clarity of the pillars of trustworthy AI for 
IBM, it is transparency. 
Let's go back to our learning example, are you designing it with users? 
Think about, what we think of as best in class, transparent recommendation systems, 
your favorite movie streaming service, or your cab hailing service? 
It's transparent, is it explainable? 
Is it telling you what recommendations are, and why they're being made, but 
also telling you as a user, it's your choice to make the final decision? 
Fairness, is it giving equal access to opportunity to everyone by ensuring 
adoption, not just of the process, but also the outcome across different groups. 
Robustness, privacy, every data scientist and developer and 
every vendor needs to know, what we mean by these in a very operational manner. 
[MUSIC]

Defining AI Ethics
Welcome to Defining AI Ethics. 
Humans rely on culturally agreed-upon morals and standards of action — or ethics — to 
guide their decision-making, especially for decisions that impact others. 
As AI is increasingly used to automate and augment decision-making, it is critical that 
AI is built with ethics at the core so its outcomes align with human ethics and expectations. 
AI ethics is a multidisciplinary field that investigates how to maximize AI's beneficial 
impacts while reducing risks and adverse impacts. 
It explores issues like data responsibility and privacy, inclusion, moral agency, value 
alignment, accountability, and technology misuse 
…to understand how to build and use AI in ways that 
align with human ethics and expectations. 
There are five pillars for AI ethics: 
explainability, 
fairness, 
robustness, 
transparency, and privacy. 
These pillars are focus areas that help us take action to build and use AI ethically. 
Explainability 
AI is explainable when it can show how and why it arrived at a particular outcome or 
recommendation. 
You can think of explainability as an AI system showing its work. 
Fairness 
AI is fair when it treats individuals or groups equitably. 
AI can help humans make fairer choices by counterbalancing human biases, but beware 
— bias can be present in AI too, so steps must be taken to mitigate it. 
Robustness 
AI is robust when it can effectively handle exceptional conditions, like abnormal input 
or adversarial attacks. 
Robust AI is built to withstand intentional and unintentional interference. 
Transparency 
AI is transparent when appropriate information is shared with humans about how the AI system 
was designed and developed. 
Transparency means that humans have access to information like what data was used to 
train the AI system, how the system collects and stores data, and who has access to the 
data the system collects. 
Privacy 
Because AI ingests so much data, it must be designed to prioritize and safeguard humans' 
privacy and data rights. 
AI that is built to respect privacy collects and stores only the minimum amount of data 
it needs to function, 
and collected data should never be repurposed without users' consent, among other considerations. 
In summary, together, these five pillars — 
explainability, 
fairness, 
robustness, 
transparency, and 
privacy — help us to design, develop, deploy, and use AI more ethically 
and to understand how to build and use AI in ways that 
align with human ethics and expectations.

Understanding Bias and AI

Welcome to understanding bias in AI. In this video, 
you will learn about bias in the context of AI. You will find out 
how bias can emerge in AI, how bias can impact AI's outcomes, and 
how to begin mitigating potential bias in AI. So bias and 
artificial intelligence is all about unwanted behaviors. 
Where AI systems that are used in consequential decision making tasks, 
whether it's in lending, hiring, even in criminal justice, gives 
systematic disadvantages to certain groups and individuals. 
So an example would be an AI system that's helping make decisions 
on who should receive extra preventative healthcare and 
it gives more of that to white people than black people. 
Or another example is hiring algorithm where 
the AI system actually gives more qualified men 
a chance to interview than qualified women. 
There's actually many sources of this happening. 
So when we talk about AI or machine learning systems, they're trained on 
historical decisions that human decision makers have made in the past. 
So because of that there's actually the possibility that the human decision 
makers in the past were implicitly or explicitly biased themselves and 
so that's reflected in the training data through prejudice. 
Second is through the sampling of the data. 
So it's possible that certain groups are overrepresented or 
underrepresented in a particular data set. 
Another aspect that sometimes overlooked is in the data processing or 
data preparation phase in the data science project. 
So even the fact that I do some feature engineering can lead to 
additional biases that might not have been there before. 
So an example of that is coming back to our healthcare example. 
So if I look at healthcare costs separately as a feature. 
So inpatient, outpatient or emergency room, 
then there's much less bias that's introduced against African Americans. 
Whereas if combine all of those into a single feature is actually 
much more biased against African Americans. 
And then there's also the question of how am I even posing the problem itself? 
Maybe I'm predicting the wrong thing. 
So if I'm trying to predict criminality where future 
crimes then arrests is not a good way to look at that 
because the police arrest people more often where 
they're more active in certain neighborhoods and 
being arrested is not the same as being guilty.
Play video starting at :3:14 and follow transcript3:14
So, as we talked about, there's many sources of those biases and so 
we need to take actions that help undo those sources. 
So one important aspect of it is even recognizing that there are biases. 
So having a team of people who have diverse lived experiences is 
one way to recognize those harms and other sort of biases that might be in play. 
Another thing is to search for 
data sets that actually have your your biases themselves. 
So that's another way to counteract biases. 
And then finally, there's technical approaches. 
So if we have a machine learning model that we're training on data 
that is biased, we can actually introduce extra constraints or 
other sort of statistical measures in order to mitigate biases. 
So we've developed several such algorithms and 
many of them are available in the open source toolkit in AI fairness.
Play video starting at :4:26 and follow transcript4:26
Our goal is to ensure technology makes a positive impact on society. 
I am proud to be a member of the IBM AI Ethics Board 
to focus on conscious inclusion through Good Tech. 
Our AI Ethics board represents a cross section of diverse 
IBM-ers who address in research eliminating 
bias while establishing standards in this space. 
A few ways in which IBM-ers use AI and data skills is by 
leveraging our technology to develop assets to address bias. 
Also to address inclusive language in tech terminology. 
This is around our Language Matters initiative. And 
to ensure testing of our solutions to decrease the probability and 
minimize the potential for bias. 
In D & I, 
we use AI and data analytics and insights to inform decision support and 
address as well as minimize the potential for bias. In HR, 
this means enhancing decision support that can inform pay and retention, 
hiring and promotion decisions. Using AI and 
data insights plays an important role to assist in our compliance requirements and 
proactive actions that we take around the globe.

AI Ethics and Regulations

Welcome to AI ethics and regulations. 
In this video, you will learn about what an AI regulation is, 
how AI regulations connect with AI ethics, 
and why it's important to understand AI regulations if you work with AI. 
A regulation is a government rule enforceable by law. 
The landscape of regulations around AI is evolving rapidly. 
And it's important to understand key pieces of regulation in order to design, 
develop, and deploy, and use AI legally and ethically. 
IBM's position is that we call for 
precision regulation of artificial intelligence. 
And we support targeted policies that would increase the responsibilities for 
companies to develop and operate trustworthy AI. 
Precision regulation of AI, 
refers to a regulation that aims to be risk-based, context specific, 
and which allocates responsibility to the party that is closest to the risk, 
which might shift throughout the AI lifecycle. 
Specifically, IBM has proposed a precision regulation 
framework that incorporates five policy imperatives for 
organizations that provide and or use AI systems. 
First, designate an AI ethics official, a lead official responsible for 
compliance with trustworthy AI. 
Develop different rules for different risks. 
In other words, regulate AI in context not the technology itself. 
Don't hide your AI make it transparent. 
Explain your AI. 
So, in other words, make it explainable not a black box decision. 
And test your AI for bias. 
[SOUND]

AI Ethics, Governance, and ESG

Welcome to AI ethics, governance, and ESG. 
In this video, you will learn about what AI governance is and what it accomplishes, 
what ESG is and what it accomplishes, 
and how governance and ESG connect to AI ethics? 
Governance is the organization's act of governing through its corporate 
instructions, staff, its processes and systems to direct, evaluate, and 
monitor, and to take corrective action throughout the AI lifecycle 
to provide assurance that an AI system is operating as an organization 
intends it to, 
and as stakeholders expected to, and as may be required by relevant regulation. 
The objective of governance is to deliver trustworthy AI by establishing 
requirements for accountability, responsibility and oversight. 
Governance provides many benefits, including, for 
example, trust. When AI activities are aligned with values 
organizations can build systems that are transparent, fair and 
trustworthy, boosting client satisfaction and brand reputation. 
Another benefit is efficiency, when AI activities are standardized and optimized 
development can happen with greater efficiency, accelerating time to market. 
And compliance, when AI activities are already managed and monitored, 
adjusting them to align with new and forthcoming industry regulations and 
legal requirements is really less cumbersome. 
The successful governance program takes into account people, processes, and tools. 
It clearly defines the roles and 
responsibilities of people, in building and managing trustworthy AI, 
including leaders who will set policies and establish accountability. 
It establishes processes, for building, 
managing, monitoring and communicating about AI. 
And it leverages tools to gain greater visibility and 
consistency into AI system performance throughout the AI lifecycle. 
ESG stands for environmental social governance, and 
these are factors that are used to measure all non financial risks and 
opportunities at companies. 
At IBM, that translates into IBM impact, our strategy and philosophy on ESG. 
IBM impact is comprised of three pillars, 
that we believe will create a more sustainable future. 
These pillars are environmental impact, equitable impact, and ethical impact. 
And we accomplish this by making a lasting positive impact, 
in the world in business ethics, our environment, and 
in the communities in which we work and live. 
The governance aspect of ESG pertains to creating innovations, 
policies and practices that prioritize ethics, 
trust, transparency and above all accountability. 
And AI ethics is a significant aspect of our governance initiatives. 
For example, in 2022 we're aiming to reach 1,000 
ecosystem partners trained in tech ethics. 
This goal matters because we believe the benefits of AI, 
should touch the many and not just the few. 
And developing a culture of trustworthy AI must happen everywhere, 
not just within IBM. 
We're leading the way in AI ethics as we aspire to create a more ethical future. 
[MUSIC]
Foundations of Trustworthy AI:
Operationalizing Trustworthy AI
Click to read the following article:

Foundations of Trustworthy AI: Operationalizing Trustworthy AI

Precision Regulation for Artificial Intelligence

Click to read the following article:

Precision Regulation for Artificial Intelligence

Lesson Summary
In this lesson, you have learned:

 Today's AI issues and concerns include how to build and use AI responsibly and how AI is used in
healthcare, facial recognition, social media and marketing, and hiring
 AI ethics is a multidisciplinary field that investigates how to optimize AI's beneficial impact while
reducing risks and adverse impacts
 AI ethics is a "socio-technical" challenge, meaning that it cannot be solved by tools alone
 AI ethics principles are:
o The purpose of AI is to augment — not replace — human intelligence
o Data and insights belong to their creator
o New technology must be transparent and explainable
 AI ethics pillars are:
 Explainability: The ability of an AI system to provide insights that humans can use to understand
the causes of the system's predictions.
o Fairness: The equitable treatment of individuals or groups of individuals.
o Robustness: The ability of an AI system to effectively handle exceptional conditions, such
as abnormal input or adversarial attacks.
o Transparency: Sharing appropriate information with stakeholders on how an AI system
has been designed and developed.
o Privacy: AI systems must prioritize and safeguard consumers’ privacy and data rights.
 In AI, bias gives systematic disadvantages to certain groups or individuals
 One potential cause of bias in AI is the implicit or explicit biases of the people who design and
develop AI
 One way to mitigate bias is to assemble diverse teams
 Regulations are government rules enforceable by law.
 AI governance is an organization's act of governing, through its corporate instructions, staff,
processes and systems
To learn more about AI ethics, visit:
 IBM AI Ethics homepage
 IBM Trustworthy AI homepage
 IBM Policy Lab
 Good Tech IBM