Answers For Pennies, Insights For Dollars: Generative AI And The Question Economy

No one seems to know exactly where the boom in Generative AIs (like ChatGPT and Claude) will lead us, but one thing is for certain:  These tools are rapidly driving down the cost of getting a good (or, at least, good enough) answer very quickly.  Moreover, they are likely to continue to do so for quite some time.  

The data is notional
but the trend is unquestionable, I think.

To be honest, this has been a trend since at least the mid-1800's with the widespread establishment of public libraries in the US and UK.  Since then, improvements in cataloging, the professionalization of the workforce, and technology, among other things, worked to drive down the cost of getting a good answer (See chart to the right).

The quest for a less expensive but still good answer accelerated, of course, with the introduction of the World Wide Web in the mid-1990's, driving down the cost of answering even tough questions.  While misinformation, disinformation, and the unspeakable horror that social media has become will continue to lead many people astray, savvy users are better able to find consistently good answers to harder and more obscure questions than ever before.  

If the internet accelerated this historical trend of driving down the cost of getting a good answer, the roll-out of generative AI to the public in late 2022 tied a rocket to its backside and pushed it off a cliff.  Hallucinations and bias to the side, the simple truth is that generative AI is, more often than not, able to give pretty good answers to an awful lot of questions and it is free or cheap to use.  

How good is it?  Check out the chart below (Courtesy Visual Capitalist).  GPT-4, OpenAI's best, publicly available, large language model, blows away most standardized tests.  

It is important to note that this chart was made in April, 2023 and represent results from GPT-4.  OpenAI is working on GPT 5 and five months in this field is like a dozen years in any other (Truly.  I have been watching tech evolve for 50 years.  Nothing in my lifetime has ever improved as quickly as generative AIs have).  Eventually, the forces driving these improvements will reach a point of diminishing returns and growth will slow down and maybe even flatline, but that is not the trajectory today.

All this sort of begs a question, though: If answers are getting better, cheaper, and more widely available at an accelerating rate, what's left?  In other words, if no one needs to pay for my answers anymore, what can I offer?  How can I make a living?  Where is the value-added?  This is precisely the sort of thinking that led Goldman-Sachs to predict the loss of 300 million jobs worldwide due to AI.  

My take on it is a little different.  I think that as the cost of a good answer goes down, the value of a good question goes up.  

In short, the winners in the coming AI wars are going to be the ones who can ask the best questions at the most opportune times.  

There is evidence, in fact, that this is already becoming the case.  Go to Google and look for jobs for "prompt engineers."  This term barely existed a year ago.  Today, it is one of the hottest growing fields in AI.  Prompts are just a fancy name for the questions that we ask of generative AI, and a prompt engineer is someone who knows the right questions to ask to get the best possible answers.  There is even a marketplace for these "good questions" called Promptbase where you can, for aa small fee, buy a customizable prompt from someone who has already done the hard work of optimizing the question for you.

Today, earning the qualifications to become a prompt engineer is a combination of on-the-job training and art.  There are some approaches, some magical combination of words, phrases, and techniques, that can be used to get the damn machines to do what you want.  Beyond that, though, much of what works seems to have been discovered by power users who are just messing around with the various generative AIs available for public use.

None of this is a bad thing, of course.  The list of discoveries that have come about from people just messing around or mashing two things together that have not been messed with/mashed together before is both long and honorable.  At some point, though, we are going to have to do more than that.  At some point, we are going to have to start teaching people how to ask better questions of AI.

The idea that asking the right question is not only smart but essential is a old one:

“A prudent question is one-half of wisdom.” – Francis Bacon

"The art of proposing a question must be held of higher value than solving it.” – Georg Cantor

“If you do not know how to ask the right question, you discover nothing.” – W. Edwards Deming

And we often think that at least one purpose of education, certainly of higher education, is to teach students how to think critically; how, in essence to ask better questions.  

But is that really true?  Virtually our whole education system is structured around evaluating the quality of student answers.  We may think that we educate children and adults to ask probing, insightful questions but we grade, promote, and celebrate students for the number of answers they get right.  

What would a test based not on the quality of the answers given but on the quality of the questions asked even look like?  What criteria would you use to evaluate a question?  How would you create a question rubric?  

Let me give you an example.  Imagine you have told a group of students that they are going to pretend that they are about to go into a job interview.  They know, as with most interviews, that once the interview is over, they will get asked, "Do you have any questions for us?"  You task the students to come up with interesting questions to ask the interviewer.

Here is what you get from the students:

1. What are the biggest challenges that I might face in this position?
2. What are the next steps in the hiring process?
3. What’s different about working here than anywhere else you’ve ever worked?

What do you think?  Which question is the most interesting?  Which question gets the highest grade?  If you are like the vast majority of the people I have asked, you say #3.  But why?  Sure, you can come up with reasons after the fact (humans are good at that), but where is the research that indicates why an interesting question is...well, interesting?  It doesn't exist (to my knowledge anyway).  We are left, like Justice Stewart and the definition of pornography, with "I know it when I see it."

What about "hard" questions?  Or "insightful" questions?  Knowing the criteria for each of these and teaching those criteria such that students can reliably ask better questions under a variety of circumstances seems like the key to getting the most out of AI.  There is very little research, however, on what these criteria are.  There are some hypotheses to be sure, but statistically significant, peer-reviewed research is thin on the ground.

This represents an opportunity, of course, for intellectual overmatch.  If there is very little real research in this space, then any meaningful contribution is likely to move the discipline forward significantly.  If what you ask in the AI-enabled future really is going to be more important than what you know, then such an investment seems not just prudent, but an absolute no-brainer.

How To Think About The Future (Part 3--Why Are Questions About Things Outside Your Control So Difficult?)

I am writing a series of posts about how to think about the future.  In case you missed the first two parts, you can find them here:

Part 1--Questions About Questions
Part 2--What Do You Control

These posts represent my own views and do not represent the official policy or positions of the US Army or the War College, where I currently work.

*******************

Former Director of the CIA, Mike Hayden, likes to tell this story:

"Some months ago, I met with a small group of investment bankers and one of them asked me, 'On a scale of 1 to 10, how good is our intelligence today?'" recalled Hayden. "I said the first thing to understand is that anything above 7 isn't on our scale. If we're at 8, 9, or 10, we're not in the realm of intelligence—no one is asking us the questions that can yield such confidence. We only get the hard sliders on the corner of the plate. Our profession deals with subjects that are inherently ambiguous, and often deliberately hidden. Even when we're at the top of our game, we can offer policymakers insight, we can provide context, and we can give them a clearer picture of the issue at hand, but we cannot claim certainty for our judgments." (Italics mine)

I think it is important to note that the main reason Director Hayden cited for the Agency's "batting average" was not politics or funding or even a hostile operating environment.  No.  The #1 reason was the difficulty of the questions. 

Understanding why some questions are more difficult than others is incredibly important.  Difficult questions typically demand more resources--and have more consequences.  What makes it particularly interesting is that we all have an innate sense of when a question is difficult and when it is not, but we don't really understand why.  I have written about this elsewhere (here and here and here, for example), and may have become a bit like the man in the  "What makes soup, soup?" video below...




No one, however, to my knowledge, has solved the problem of reliably categorizing questions by difficulty.

I have a hypothesis, however.

I think that the AI guys might have taken a big step towards cracking the code.  When I first heard about how AI researchers categorize AI tasks by difficulty, I thought there might be some useful thinking there.  That was way back in 2011, though.  As I went looking for updates for this series of posts, I got really excited.  There has been a ton of good work done in this area (no surprise there), and I think that Russel and Norvig in their book, Artificial Intelligence:  A Modern Approach, may have gotten even closer to what is, essentially, a working definition of question difficulty.

Let me be clear here.  The AI community did not set out to figure out why some questions are more difficult than others.  They were looking to categorize AI tasks by difficulty.  My sense, however, is that, in so doing, they have inadvertently shown a light on the more general question of question difficulty.  Here is the list of eight criteria they use to categorize task environments (the interpretation of their thinking in terms of questions is mine):

• Fully observable vs. partially observable -- Questions about things that are hidden (or partially hidden) are more difficult than questions about things that are not.
• Single agent vs. multi-agent -- Questions about things involving multiple people or organizations are more difficult than questions about a single person or organization.
• Competitive vs. cooperative -- If someone is trying to stop you from getting an answer or is going to take the time to try to lead you to the wrong answer, it is a more difficult question.  Questions about enemies are inherently harder to answer than questions about allies.
• Deterministic vs. stochastic -- Is it a question about something with fairly well-defined rules (like many engineering questions) or is it a question with a large degree of uncertainty in it (like questions about the feelings of a particular audience)?  How much randomness is in the environment?
• Episodic vs. sequential -- Questions about things that happen over time are more difficult than questions about things that happen once.
• Static vs. dynamic -- It is easier to answer questions about places where nothing moves than it is to answer questions about places where everything is moving.
• Discrete vs. continuous -- Spaces that have boundaries, even notional or technical ones, make for easier questions than unbounded, "open world," spaces.
• Known vs. unknown -- Questions where you don't know how anything works are much more difficult than questions where you have a pretty good sense of how things work.  

Why is this important to questions about the future?  Two reasons.  First, it is worth noting that most questions about the future, particularly those about things that are outside our control, fall at the harder rather than easier end of each of these criteria.  Second, understanding the specific reasons why these questions are hard also gives clues as to how to make them easier to answer.  

There is one more important reason why questions can be difficult.  It doesn't come from AI research.  It comes from the person (or organization) asking the question.  All too often, people either don't ask the "real" question they want answered or are incredibly unclear in the way they phrase their questions.  If you want some solutions to these problems, I suggest you look here, here and here.  

I was a big kid who grew up in a small town.  I only played Little League ball one year, but I had a .700 batting average.  Even when I was at my best physical condition as an adult, however, I doubt that I could hit a foul tip off a major league pitcher.  Hayden is right.  Meaningful questions about things outside your control are Major League questions, hard sliders on the corner of the plate.  Understanding that, and understanding what makes these questions so challenging, is a necessary precondition to taking the next step--answering them.

Next:  How Should We Think About Answers?  

How To Think About The Future (Part 1 -- Questions About Questions)

We don't think about the future; we worry about it.

Whether it's killer robots or social media or zero-day exploits, we love to rub our preferred, future-infused worry stone between our thumb and finger until it is either a thing of shining beauty or the death of us all (and sometimes both).  

This is not a useful approach.

Worry is the antithesis of thinking.  Worry is all about jumping to the first and usually the worst possible conclusion.  It induces stress.  It narrows your focus.  It shuts down the very faculties you need to think through a problem.  Worry starts with answers; thinking begins with questions.

What Are Your Questions?

“A prudent question is one-half of wisdom.” – Francis Bacon
"The art of proposing a question must be held of higher value than solving it.” – Georg Cantor
“If you do not know how to ask the right question, you discover nothing.” – W. Edwards Deming

Given the importance of questions (and of asking the "right" ones), you would think that there would be more literature on the subject.  In fact, the question of questions is, in my experience, one of the great understudied areas.  A few years ago, Brian Manning and I took a stab at it and only managed to uncover how little we really know about how to think about, create, and evaluate questions.

For purposes of thinking about the future, however, I start with two broad categories to consider:  Speculative questions and meaningful questions.  

There is nothing wrong with a speculative question.  Wondering about the nature of things, musing on the interconnectedness of life, and even just staring off into space for a bit are time-honored ways to come up with new ideas and new answers.  We should question our assumptions, utilize methods like the Nominal Group Technique to leverage the wisdom of our collective conscious, and explore all of the other divergent thinking tools in our mental toolkits.  

Speculation does not come without risks, however.  For example, how many terrorist groups would like to strike inside the US?  Let's say 10.  How are they planning to do it?  Bombs, guns, drones, viruses, nukes?  Let's say we can come up with 10 ways they can attack.  Where will they strike?  One of the ten largest cities in the US?  Do the math--you already have 1000 possible combinations of who, what, and where.

How do we start to narrow this down?  Without some additional thinking strategies, we likely give in to cognitive biases like vividness and recency to narrow our focus.    Other aspects of the way our minds work--like working memory limitations--also get in the way.  Pretty soon, our minds, which like to be fast and certain even when they should be neither, have turned our 1 in 1000 possibility into a nice, shiny, new worry stone for us to fret over (and, of course, share on Facebook).

Meaningful questions are questions that are important to you--important to your plans, to your (or your organization's) success or failure.  Note that there are two criteria here.  First, meaningful questions are important.  Second, they are yours.  The answers to meaningful questions almost, by definition, have consequences.  The answers to these questions tend to compel decisions or, at least, further study.

It is entirely possible, however, to spend a lot of time on questions which are both of dubious relevance to you and are not particularly important.  The Brits have a lovely word for this, bikeshedding.  It captures our willingness to argue for hours about what color to paint the bikeshed while ignoring much harder and more consequential questions.  Bikeshedding, in short, allows us to distract ourselves from our speculations and our worries and feel like we are still getting something done.

Next:  What do you control?

RFI: Should Intelligence Analysis Be More Like Competitive Diving?

Quick!  Which is more difficult:  A jackknife or three and a half somersaults from a tuck position? In case you are not familiar with these dives, you can see videos of both below.





Now, here is the more difficult question: How much more difficult is a 3.5 from the tuck than a jackknife?

The answer is about 2.3 times more difficult. How do I know this? Because I checked out the handy diving tables at FINA (the international organization that regulates diving). I'm no expert but my reading of the tables says that a 3.5 from the tuck is a dive with a 3 point difficulty while a forward dive from the pike position (a jackknife?) is a 1.3 point dive.

Note that the degree of difficulty is simply a multiplier for the actual score of the dive. It is theoretically possible that a perfect jackknife would beat a lousy 3.5 somersault.

Intelligence, right now, is all about scoring the dive. Degree of difficulty? Not so much. 

I am hoping to change that...

We spend a good bit of time in intelligence talking about forecasting accuracy and we should.  Saying accurate things about the future is arguably much more valuable to decisionmakers than saying accurate things about the present or past.  It is also inherently more difficult.

Even when we are trying to say accurate things about the future, though, some questions are easier to answer than others.  Quick!  Which is more difficult to answer:  Is there likely to be a war somewhere in the Middle East in the next 100 years or is there likely to be a war between Israel and Egypt within the next 24 months?  I am no Middle East expert but it seems to me that the first question is much easier than the second.  I am guessing that most readers of this blog would say the same thing.

Why?   What are the essential elements of a question that make it obviously more or less difficult to answer?  How do we generalize these criteria across all questions?

I am not the only person to recognize the inherent difficulties in different kinds of questions.  Michael Hayden, the former Director of the CIA and NSA, likes to tell this story:

"Some months ago, I met with a small group of investment bankers and one of them asked me, 'On a scale of 1 to 10, how good is our intelligence today?'  I said the first thing to understand is that anything above 7 isn't on our scale. If we're at 8, 9, or 10, we're not in the realm of intelligence—no one is asking us the questions that can yield such confidence. We only get the hard sliders on the corner of the plate."

Note that Hayden highlighted the degree of difficulty of the questions (not the difficulty of obtaining the information or the complications associated with supporting political appointees or the lack of area experts or anything else) as the reason for more realistic expectations for the intelligence community's analysts.

So...if degree of question difficulty is the missing half of the "evaluating intelligence" equation, shouldn't someone be working on a diving-like degree of  dfficulty table for intel analysts?

That is precisely what I, along with one of our top graduate students, Brian Manning, have set out to do this year.  This research question piqued our interest primarily because of our involvement in the DAGGRE Research Project (more on that soon). 

In that project, we are asking questions (lots of them) that all have to be resolvable.  That is, they have to all have an answer eventually ("Will Moammar Ghaddafi be President of Libya after 31 DEC 2011?" is a resolvable question -- he either will or he won't be president after that date). 

My concern was that this is not the way that most questions are actually asked by the decisionmakers that intel typically supports.  For example, I would expect that the Ghaddafi question would come at me in the form of "So, what is going to happen with Ghaddafi?"  A very different question and, intuitively, much more difficult to answer.

So far our research has turned up some interesting answers from the fields of linguistics, artificial intelligence and, from all places, marketing. We expect to find interesting answers in other fields (like philosophy) but have not yet. 

Our goal is to sort through this research and figure out if any of the existing answers to this "question about questions" makes any sense for intel professionals.  Alternatively, we might take elements from each answer and kludge them together into some steampunk-looking difficulty of question generator.  We just don't know at this point. 

What we are looking for is good ideas, in general, and, in particular, any real research into how to rank questions for difficulty.

The comments section is now open!