In this episode of Motley Fool Rule Breaker Investing, Motley Fool co-founder David Gardner sits down with Vasant Dhar to ask not merely what machines can do, but how humans can think with them. They discuss these questions:

• When should we trust AI?
• When does it sharpen our thinking -- and when does it make our brains lazy?
• If intelligence itself is becoming a commodity, where will tomorrow's value be created?
• And in a future where machines become extraordinarily capable, what are humans for?

To catch full episodes of all The Motley Fool's free podcasts, check out our podcast center. When you're ready to invest, check out this top 10 list of stocks to buy.

A full transcript is below.

This podcast was recorded on Aug. 12, 2026.

David Gardner: August means authors on Rule Breaker Investing. Amazingly, this is our ninth year of Authors in August. Last week, philosopher Thi Nguyen joined us to ask, what happens when the scores around us start shaping what we value? This week we turn to the technology shaping just about everything else these days, and that's artificial intelligence. My guest Vasant Dhar has been working in AI since long before most of us had heard the term, teaching it, building companies around it, bringing machine learning to Wall Street, and now writing his book, Thinking with Machines. What has somebody with a 40-year front-row seat learned about where AI came from, where it's going, and most importantly, how we humans should think alongside it without surrendering our own judgment? Thinking with Machines, with Vasant Dhar, only on this week's Rule Breaker Investing.

Welcome back to Rule Breaker Investing. I'm excited to be joined by Vasant Dhar in just a minute. Let me mention this is Authors in August, and what are we doing next week? Well, our third and final author for August is an author of a book I've loved and returned to for years, and that's The Why of Work by Dave and his wife, Wendy Ulrich. We began this month by asking Thi Nguyen how the scores around us shape what we value, and Vasant Dhar is now going to help us think alongside increasingly intelligent machines. Next week, perhaps the most personal question of all, why do we work? Not just why we need a paycheck, but why some workplaces give us energy, identity, connection, and purpose, while others drain them away. If you lead people, if you work with people, or simply spend a substantial portion of your waking life working, tune in next week for my conversation with Dave Ulrich on The Why of Work.

Vasant Dhar is the Robert A. Miller Professor at the Stern School of Business and professor of Data Science at New York University. Vasant refers to himself as an artificial intelligence researcher and data scientist, but given his long and deep history with AI, I'm going to add my own word here to characterize him: pioneer. Vasant Dhar is an AI pioneer and host of the podcast Brave New World, which explores how AI and other technologies are transforming humanity. I was privileged, I should mention, to appear on Brave New World this summer, so if you'd like to hear what happens when questions get turned on me instead of vice versa, do listen in to my Foolish talk with Vasant on Brave New World, wherever you find and listen to podcasts. Our episode came out just a few weeks ago on July 16. Vasant Dhar, great to be with you again, and welcome to Rule Breaker Investing.

Vasant Dhar: David, delighted to be on the show, very much looking forward to our conversation.

David Gardner: This is going to be a fascinating conversation, and I know that, not even knowing where we're going exactly, because one of my favorite lines comes from Lord Peter Wimsey, which was Dorothy Sayers answer to Sherlock Holmes. A century ago, her hero was Lord Peter Wimsey, her protagonist, and his family crest said, where my whimsy takes me. Vasant, I think that's very apropos of our conversation this week.

Vasant Dhar: I like that.

David Gardner: Thank you. Let me just start by saying you were there before it was cool. Vasant, you've taught your first AI course, if I have this right in 1984. You founded four AI companies. You brought machine learning to Wall Street in the 1990s. Now four decades later, suddenly everybody else is now using AI as well. Let me start by asking you, Vasant, what does the AI revolution look like to somebody who's actually watched almost the whole thing happen?

Vasant Dhar: Well, it's almost like I've watched it in slow motion, and I don't mean for that to sound like a train wreck. It's actually been quite a ride in slow motion. I say slow motion because it's been 40-plus years. It's been 40, I don't know, seven years since I got into the field in 1979, and little did I realize at the time what I was getting into. As I sometimes refer to the famous Jerry Garcia line, "What a long, strange trip it's been." It really has. But it didn't answer your question specifically, which is that what I've really seen is this progression, what I call these paradigm shifts, where we've gone from the paradigm of specification, which is what we used to do in the late ‘70s, the ‘80s, where we specified knowledge by talking to humans, by eliciting what they knew and then encoding it into the computer. It sounds so arcane now, but that was it, and we thought that that would take us to the promised land. Little did we know that those tools really were not sufficient. Our aspirations were very high, though. The language we used to describe AI was things like thinking, reasoning, understanding, planning; that was the vocabulary of AI in the ‘70s when I got into it.

By the time the late ‘80s and ‘90s rolled around, we'd gotten a little frustrated. Things had stalled. Progress had stalled somewhat because it's very difficult to do what I just said you were trying to do. It was just very difficult for people to articulate everything they know. It's hard to disentangle expertise from common sense. All that was difficult. People said, well, let's put this on hold for a moment. Data has become available. Machine learning came to the fore, to the rescue in a sense, and we said, let's just learn to predict from data. For the next 20, 30 years, and I'd argue even until now, the emphasis was prediction. There were still some roadblocks there that you had to do something, all feature engineering to massage the data, to help the computer to find patterns, because the algorithms of the time were still very weak. They needed a lot of human assistance. We solve that problem with deep learning because deep learning was all about perception. The machine sees the world, hears the world, reads the world, touches the world, and now smells the world. These are sensory inputs. Intelligence moved upstream, and that was tremendously exciting because now you expected the machine itself to do feature engineering.

Then the rails paradigm shift is to general intelligence, where the machine knows something about everything, and that something is getting deeper and everything is getting wider. Now here we are in this new paradigm. Now, the aspirations are even higher. I hear terms like recursive improvement, where the machine is able to improve itself automatically without any supervision. The rails are in place for that, and so it's a tremendously exciting time in AI.

David Gardner: One thing you do so well in the book, Vasant, and again, I am not a data scientist. I am somebody who enjoys technology, and I was delighted that The Motley Fool got to start right as the Internet seemed to start. It was very fortuitous timing for us. I love technology, but I wasn't following the progression of artificial intelligence. In your book, Vasant, you do a really nice job showing the four eras, as I think you might say, of artificial intelligence. It started with expert systems, and then there was machine learning, and then deep learning. Then you just mentioned it, general intelligence. For the rest of us, and I include myself, could you just briefly summarize that progression from expert systems through to general intelligence? Maybe just a few sentences to characterize each so the general listener understands the progression.

Vasant Dhar: The progression in a nutshell is just specification to learning from curated data, but there was still a lot of effort required to curate the data. The next progression was learning from original data as opposed to curated data. That was the shift to deep learning. Then the latest shift is learning about anything from all of the data out there, completely uncurated. By all the data, I mean, truths, lies, falsehoods, emotions, everything on the Internet that's available, you learn from that. The thing about that is that you learn everything. You learn truths, you learn falsehoods, you learn manipulation, and arguably, computers. AIs these days are very capable of manipulation, as well. They've learned all of these things from humans, our good side, our bad side, they've learned everything. That's where we are at the moment. That's the progression in a nutshell.

David Gardner: Thank you. A lot of it was just human-centered in terms of our giving the data, our telling the machines what we wanted from them, into increasingly machines using video, for example, and sound, and you talk about maybe even smell in time, starting to sense itself the AI data and then not needing human intercessors to provide any filtering or middleman effort. You end up with just AI learning on its own and us observing it and trying to figure out what it's thinking. Could you give an example? Internist was a really interesting system that you have a lot of familiarity with, and early on in the book, you talk some about early efforts to help us understand what's going on with our health. This is more like 1980s AI, but could you just paint that picture a little bit?

Vasant Dhar: Yes, so, Internist is really what got me into AI. I had no idea what it was. I'd gone with a bunch of PhD students to ask this professor to offer a course in AI. While we were waiting there, I was watching this interaction between the system called Internist and a legendary physician called Jack Meyers, whose brains they had picked to actually create the knowledge base of Internist. Jack Meyers was sitting, puffing a cigar, talking to Internist through his assistant because he couldn't type; no one could type in those days. They were interacting. He entered a bunch of symptoms about a case. Internist asked him a bunch of questions. They went back and forth. At one point, Internist asked a question, and Jack Meyers said, "Why are you asking me the question?" The response blew my mind. The Internist said, because the evidence you've given me so far is consistent with the following hypotheses, this question will help me discriminate between the top two. I was like, just floored. I was like, how the hell is a machine doing this? That was my aha moment.

I was like, this is what I want to do with my life. That's what really got me into AI, and medicine was a key poster child, really, for AI at the moment, because it was one of those areas where you could reasonably circumscribe the boundary of knowledge or medical knowledge and say, this is medical knowledge, and as long as we can stick to it, a system will do quite well. But, of course, the trouble is that we often go beyond that. The way you walk into a physician's office tells him or her a lot about you, the state of your health. There are all these subtle, certain symptoms, cues that people get a lot of information out of. That's what we bled into now is that this distinction between expertise and common sense has completely dissolved, and that was the biggest barrier, in my opinion, to progress in AI, because we drew those boundaries artificially, and, of course, humans don't draw any boundaries; even when we get expertise in some subject, we don't forget our common sense. We actually tend to use it maybe even more. That was essential. That's been the big deal of general intelligence is this dissolution of the boundary between expertise and common sense. Now, it doesn't matter. The AI doesn't care whether you're talking to it about, let's say, a soccer game in FIFA or whether you're talking about some deep concept in quantum mechanics or medicine. It is equally knowledgeable about all of those things and doesn't even know the distinction between them.

David Gardner: Vasant, I don't want to give short shrift to your backstory, because I'd love for you to share a little bit about how you got to this country, Pittsburgh, ground zero. I want to make sure our listeners hear some about the man and how he got here. I would say the remarkable in addition, Jack Meyers, the remarkable people that you got to meet as a young person and who obviously set you on your way and influenced your career, Herbert Simon, for example.

Vasant Dhar: Simon was a huge influence in my life. The reason I met Simon is because the person who designed the medical diagnosis system, Harry Pople, had been a student of Herb Simon. I met Harry, and Harry introduced me to Simon. Simon, for some reason, took a liking to me and would allocate a half hour every week of his time if I wanted it, which I usually took. He was incredibly generous and had a huge impact.

David Gardner: Nobel Prize winner, right?

Vasant Dhar: Nobel Prize winner in economics, a Turing Award winner, part of that 10-person committee in 1956 that coined the term AI. Simon convinced me that AI was around the corner. He was a trailblazer created the field and made all kinds of very bold predictions. In the late 70s, when I met him, he had made a prediction that by the end of the next decade, the AI would be the chess champion. It took a little longer, but a lot of his predictions have actually turned out to be correct. He thought things would happen in 10 years. They took 30 or 40, but in the larger scheme of things, what's 10 or 20 years here or there. He was just a huge influence on my life. Taught me how to think, taught me about the foundations of AI, along with my other co-mentor, Harry Pople. These two people, real AI pioneers, had a huge impact on setting the direction of my life.

David Gardner: What was the progression then from Pittsburgh to New York?

Vasant Dhar: Well, so I was looking for a job, and NYU was hiring, Wharton was hiring, a few other schools were hiring. I came to NYU for my interview, and I remember walking down Washington Square Park and thinking to myself, for someone in their 20s, I said, wow, this would be such a cool place to live. Of course, after my talk, they took me to one of the restaurants in Greenwich Village, and I was like, wow, this is where I want to live. I told them, I said, look, if you offer me a job, I'm coming, and I did, and the rest is history. That was the only school I really looked at seriously because I wanted to be in New York City.

David Gardner: Early on in your book, you modestly, one might even for the fun of it, say, immodestly, give us Dhar's conjecture. I want to make sure we talk about this because it's really good. Here it is in a nutshell, if I'm quoting myself accurately. It's patterns often emerge before the reasons for them become apparent. I'd like for you to tell us what that means. Maybe after a lifetime spent with machines, do you now trust a strong pattern even when you can't explain why it's there?

Vasant Dhar: Let me start with where that conjecture came from because my very first experience with machine learning was actually with ACNielsen, the media company. They had a division in Long Island, Port Washington, and the gentleman there handled their analytics, and they were selling information to companies about how their products were doing. He came to me and said, hey, I believe you're doing something in machine learning. We've got all this data, and we sell information, we sell knowledge. Do you think you can extract interesting patterns from this data? I don't know what interesting means, but do you think you can do that? I was like, sure, I'll give it a shot. I've been working on this genetic rule learning algorithm, so I cranked it on the data and out came a bunch of things and I'll never forget this. He came to NYU, and he said, so Vasant, have you found anything? I said, yeah, a lot of older women in the Northeast do most of their shopping on Thursdays. He said, Yeah, that's coupon day. What else did you find? I was like, wow, this is amazing. I hadn’t told the machine anything except to find me unusual consumption patterns, and this was an unusual one, like several others. That's when I realized like, wow, patterns emerge before reasons for them become apparent.

This is something I encountered like a year later, I was on Wall Street and it was deja vu all over again. I went to work for the prop trading group. They didn't want to tell me anything they knew, but they wanted to know everything I knew, so I proposed an experiment. I said, give me all your traits, and I'll tell you if you could have done better. They said, well, you don't need to know anything about the strategy and I said, no, just give me the trades. But they gave me the trades and I applied the same trick I'd applied at Nielsen. I took the trades. I attached market conditions to each trade, cracked it through the algorithm, and showed up for the meeting, and again we used to meet on Friday afternoons after market close, and Kevin Parker, who was running the technology and the trading group comes to me and says, so Vasant, did you find anything? I said, yes, I did, but I have no idea what it means. I said, take it from the top. I said, when the 30-day volatility is in the lowest quartile last year, your trades are three times as profitable as they are otherwise. Silence around the room, and then they start cursing each other saying, how long have I been telling you to look at volatility, and this guy who knows nothing about markets is telling us that it matters. I said, well can anyone explain what's going on? They said, no, not really. But we feel the pain whenever volatility spikes. It's interesting that you're telling us this.

Of course, I found the reasons for it much later. I scoured all the literature in finance, and I did find the reasons for the pattern that I'd found, and that was another data point that, wow, this is so true. I've encountered this dozens of times over my career is that if you look at the data creatively that it screams out at you. Now, of course, the challenge is which of these is real and which of them is ephemeral and I've spent my entire career answering that question, as you can imagine. Because my approach to financial market-to-market prediction is systematic. That is, get the machine to learn and to predict, and you better be sure that those patterns that the machine has picked up on are real, as opposed to phantom, because if they're phantom, you're going to lose money, and there's no better scorecard to tell you how good your science is other than P&L. I mean, the quality of your thinking shows up in the P&L. Of course, I should qualify that by saying there are times when you get lucky, there are times you get unlucky. That's the nature of markets. As my friend Kevin Parker used to say, don't confuse brains with the bull market that anyone can make money in the bull market, right? It's the ability to make money in all kinds of markets that really distinguishes great investors, and that's always been my holy grail, as far as AI is concerned, is that it should perform well in all kinds of markets. Of course, that's a question I get all the time when I talk to investors and show them models. The first thing they want to they invariably ask me is under what conditions will this do poorly? That's like a core question that I've faced dozens of times, and I've gotten very used to answering.

David Gardner: I know we talked about this on your podcast, but part of your background and part of what was being asked of you by Wall Street is, as you just said, Vasant, trying to make money in all markets. I will just say that is not the route that I've taken as an investor, and at least those who probably are listening to this podcast know that, generally, I'm ready to lose money one year and three, along with the market, but at least for rulebreaker investing, it's not just making money in a bull market. It's beating up on the averages in bull markets so that you so far exceed them with one hopes rulebreaker stocks that you don't mind when the tide recedes and you tend to have even worse downtimes more volatile than others. I do there is a Holy Grail around lack of volatility or predictability, as you mentioned earlier, Vasant. I definitely understand, especially institutionally, that's how Wall Street rolls, but I find myself marching to the beat of a different drummer a little bit when it comes to how to beat the market.

Vasant Dhar: I totally get that. By the way, I really enjoyed our conversation than reading your book because it clarified something for me. The reason I break down the investment space into three horizons, like, high frequency, short term, long term. Your approach and philosophy applies to long-term investing, which is invest in quality companies and don't sell too early. Wait for those to become five baggers, 10 baggers, 100 baggers, even, and those will more than make up for the ones that go to zero. Great advice. I started in a different space, which was the short-term and high-frequency and short-term space because I approached markets as a scientist. I said, Where do I have enough data to build credible statistical models from? Is there enough data in long-term investing? No, there isn't, or rather, that kind of data is not amenable to a statistical or a machine learning approach.

David Gardner: Yes.

Vasant Dhar: On the other hand, when your investment horizons get shorter and shorter, you essentially have many more trials per unit of time.

David Gardner: Absolutely.

Vasant Dhar: Your models can become more robust, and you take more swings at the plate. What I mean by that is that these algorithms have very little edge, like a minute edge. You're looking for those home runs for 1,000 baggers, right? I'm looking for the singles and doubles, and I just want to, like, take them all day. It's what a market maker does all day, right? Yes. You know, they're buying at the bid and selling at the ask and making money all day, just like pennies. That was really the approach that I gravitated to, not because I thought it was philosophically better, but only because that was amenable to the scientific method. That's what took me to the short-term space. As you well know, I've now extended myself into the long-term space through my Damodaran bot project, which attempts to simulate the thinking of my valuation colleague Aswath Damodaran. I've now taken AI to the long-term space because that became possible with modern AI that became possible when ChatGPT came out, because now we had this general intelligence at our fingertips and, like I said, it knows something about everything, and it actually knows something about financial markets as well.

It's already been pretrained to some degree to think about finance. If you talk to Claude, in fact, I talked to Claude and ChatGPT yesterday, and I posed a question. I posed two questions. One is, which companies are the biggest beneficiaries from increased European spending, defense spending? The other one was very open ended, which is, give me the top 10 stocks I should buy now. Now, with respect to the first one, it did a pretty good job of telling me which ones are exposed to European defense spending, although, even though they expose it doesn't mean that they're going to make money, but it did provide that answer. I can trust that.

On the other hand, when it answers this larger question about which 10 stocks should I buy now, I found the interaction fascinating because it took me in all these directions. But I had to do a lot of heavy lifting, a lot of work, a lot of pushback. I even berated Claude for even considering leveraged ETFs, which I always tell people to avoid and it's no shrinking violet, by the way. It comes back at you and says I only mentioned it for completeness. Like don't be offended. It's amazing the personality these machines have taken on. But come back to your question, that these are very different styles of investing. One where you're just taking quick hits, and you want to take as many of them as you can. Huge amount of data. Yeah, because that's what amplifies your edge.

It's what Roger Federer talks about in tennis. Because tennis sports and investing, I see is two sides of the same coin in one sense very adversarial. You're competing with the best, you're competing with the most motivated, very similar kinds of problems, and as Federer says, he only won 54% of his points, but he won 80% of his matches. Similar kind of logic applies in finance. You can have that slide edge that you're winning like 51% of the time, but you keep taking lots of swings at the plate, guess what? You're not going to have that many losing days. In fact, when I did high frequency trading, I used to have, like, one losing day a month. It was amazing. On the other hand, I couldn't make a whole lot of money because the capacity of the market is limited. I can only make so many beds per unit time of a certain size, because if I start betting too high, I'm going to move the market against me. That's been my journey more in the short-term space, but I'm very interested in the long-term space. I'm very interested in your thinking and actually trying to infuse that into AI, as well.

David Gardner: Well, thank you. Truly, we did go there some on your podcast. I do want to just remind our listeners that if you'd like, first of all, Vasant's podcast is fantastic. You should be listening to it, dear listener. Anyway, Brave New World, and if you want to hear us talk more about rulebreaker investing, I would highly encourage you to tune in. Trust is such an important thing in life as George Schultz, the former Secretary of State, once wrote, entitled a influential essay. I remember reading Trust is the coin of the realm and I know that you are of this school, as well, Vasant. Would you tell the story of jumping into a swimming pool next to your dad as a kid? Then let's think together about when can we trust AI and when can we can't and how can we tell the difference?

Vasant Dhar: I have this chapter on trust in my book, and I start that off with this vignette, where we'd gone to the swimming pool. It was in one of these big colonial clubs in India, big swimming pool, and I was with my dad, he jumped in, and he said, Hey, come on, jump in. I didn't know how to swim, but I just jumped in.

David Gardner: How old were you?

Vasant Dhar: I don't know, 6?

David Gardner: Wow.

Vasant Dhar: I don't know how to swim, but he was there, and I jumped right in. I went to the bottom. I came up, and I felt him propping me up, pushing me a little bit away, seeing if I could swim. I managed. Then he said, see, now you can swim, you know? The reason I jumped in was because there was just complete trust. You know, I knew that nothing bad was going to happen. You know, if I went to the bottom, he'd, like, drag me up, so complete trust. I open the chapter with that incident and then talk about trust in the broader sense and whether we should trust AI. Now, I wrote this article in the Harvard Business Review, like 12-ish years ago. I'm forgetting called when to trust robots with decisions and not to. That was based on about 10 or 15 years of real implementing algorithms in finance and sports and medicine and the question I came up with in each of those domains was, why should I trust the algorithm? Why do I trust this algorithm in finance? That's wrong almost half the time. Then when I ask myself, why don't trust a driverless car, that's rarely wrong. The answer became so obvious, which is that it depends on the cost of error. If the cost of error is really high, you're going to be reticent to trust something. If the cost of error is low, not much to be lost, trust it, especially if it doesn't make mistakes that often.

That's what I realized is that trust really depends on how often an answer will be wrong and the consequences of the error. That is the consequence of being wrong. When I looked at the world, in that sense, it all made sense. That is my finance algorithms are wrong a lot, but the cost of error was low because I take very little risk in each position. I've got hundreds of positions on during a day. Even if the position blows up, it's not going to blow up my portfolio. There isn't that huge systemic risk. On the other hand, in a driverless car, and by the way, I have had a near-death experience, a little over two years ago, so the costs of error there are very high, like death. I would hesitate before I trust an algorithm in a situation where the cost of error is death.

David Gardner: Part of my own experience, and I use ChatGPT and have every day for three years now, but not 30 years. But part of my experience has been that, yeah, it hallucinates, and you certainly go right there in your book. By the way, let me mention your book, Thinking with Machines, 2026. I've read it in full because I always do before an author in August podcast, but this is really fresh. You are reacting to things that were just months ago, in some cases, Vasant, although you're able to look back decades ago. But one of the points that you make in and around trust, and I always feel this with ChatGPT or as you mentioned, Claude, is when you can get it not just to identify an answer or give a pattern, but explain explain itself. For example, your list of 10 stocks for Claude yesterday, I know that you are smart with prompting, and so much of AI seems to be who's smart with prompting? But asking the AI actually to explain its decisions or choices, maybe not in every circumstance is that possible, but doesn't that greatly increase the trust that we have?

Vasant Dhar: It does. As you mentioned sense making a little bit earlier, that I've actually spent most of my career, most of my time actually making sense of the outputs of the AI, you know? That is critical because that's what people always want to know. No. 1, what's the story here? Like, what's it really doing at a high level? Like, explain the story. The story really matters. The connection between the story and the numbers is critical. What's the story? Then they want to know, why does it work? Why does it fail? When does it work? When does it fail, and so it’s just a constant process of probing and sense-making, and very often, it results in failure. That is you just cannot explain it, and if you just can't explain it and find a, a logic behind it, then you're much more uncomfortable with the question about whether this is real or ephemeral, and that's true in every domain, but it's particularly challenging in finance, because finance is such a noisy problem, such a noisy problem. You can have two situations that seem virtually identical and yet the outcomes are very different. That confuses the hell out of the machine and it also makes it much harder to explain something.

In finance, I find myself explaining things in two ways. One is, in drawing analogies of a new system with systems that we already understand. For example, people understand trend following. They understand momentum. You can explain a system by saying, well, it behaves like long-term momentum when volatility is low, but it has a very short-term momentum orientation when volatility increases. An explanation like that starts to make sense to investors. Someone can nod along and say, I can buy that, that volatility is low, you go with the trend. When it's high, be careful because things will revert more often, and then you will also show that the system actually changes positions more often when things get volatile versus when they're quiescent. It's those kinds of things that I find useful in trying to explain the behavioral system is drawing an analogy between that and things that we already know.

David Gardner: Yeah, scaffolding that we have as investors or just thinkers. It makes me smile when we hear the stock market was up or down this day for X reason. The glib headlines, the quick assignment of why the market did what it did.

Vasant Dhar: Yeah.

David Gardner: Usually a single factor or story explaining a million different directions and moves and positions.

Vasant Dhar: I've always been amazed by that kind of a statement. I was how do they know this? Where's the data? Where's the evidence?

David Gardner: Yes, well said.

Vasant Dhar: Actually, to your point, it's actually also important not to make up the wrong story. The story has to be the right one, and one of the traps that we can fall into is making ourselves believe a story that we shouldn't believe in. That is one of the traps that we sometimes fall into when we really want to believe something. As Richard Feynman said, the easiest person to fool is yourself, so you have to really take pains to avoid that at all cost.

David Gardner: Let's talk just briefly about writing your book, Vasant. I mean, you've lived this subject for more than 40 years. Why was thinking with machines the book you wanted to write now? Who did you imagine your reader is? Who's sitting across the table from you as you write?

Vasant Dhar: I was very ambitious in who I wanted to read this book. I felt that this is a pivotal technology at a critical moment, and I wanted to write a book that was accessible to everyone. Students, parents, teachers, scientists, policymakers, grandma, everyone. I wanted to make this book accessible to everyone, and so I had to write it in a way that was faithful to the concepts but was understandable. That was one of my objectives. The other objective I had was that these different constituents would get something out of the book, each one different. This is also a book for my colleagues. That is people who know a lot about AI, know everything about AI. Even for them, there's something interesting in there about how to think about AI in a new way. That was my motivation was to write a book with broad-based appeal that was accessible to everyone and yet had sufficient intellectual heft that even my colleagues and expert scientists in the field would say, hey, this is an interesting way. It's an interesting way to look at AI and where it's going. For policymakers, again, it provides some blueprint for how to think about AI. Because one of the questions that I raised in the book is, are laws sufficient for this era of AI? Or do we need to be thinking in terms of new laws, new policies? Short answer, I was very ambitious. I wrote it for everyone, and I wanted there to be something in it for everyone.

David Gardner: It really delivers. In my Foolish opinion, I would say it's an 11-chapter book, and it's about 200 pages long. It has some appendices after. But, Vasant, what you do is you're not telling AI through technology, but through stories, through your own career, medicine, markets that we've already talked some about, of course, great interest to Rule Breaker Investing listeners. Decision-making, we've talked about trust. I so appreciate that. I have two reactions back for you. One of them is a joke question that you'll handle very well. But my first reaction is Chapter 11, your final chapter on its own is worth the price more than the price of the book because at that point, having told the backstory, having shared a lot of thoughts around these things we've discussed in this conversation, for instance, you then get to the what does the future look like? How does this all play out? How do laws change? Who owns financial assets? Rather than give my money to my children, I could give them to an AI that continues to embody me and my choices. How do you tax that? How is that run? Those are just a couple of silly examples, but of interest to people listening to us right now.

But Chapter 11 is a phenomenal and stark confrontation of how this technology challenges so many of the received mores that you and I and many of us have grown up with, and there are a whole new constructs that need to happen. We can get there a little bit later, but in the meantime, I just wanted to ask you my joke question, which was, well, you wrote it yourself in the book, Vasant. You said, some people suggested that I have ChatGPT write this book, but I still write better than an AI machine you wrote, although those days are numbered. By the way, you do write better than an AI machine. Any thoughts back on that?

Vasant Dhar: There's a lot in there, David, and I'll just stop by saying that I have to thank my podcast producer for telling me, because we were at this conference, and he said, Vasant, if you write another book on AI, I'm not going to read it. But if you write your story, I will. He said, because you got a really interesting story. He knows my story. He says, you've seen the field. You had a front row seat. If you write it and make it personal, I'll read it. I have him to thank for that, and I should really thank him because everyone who reads it tells me that they really enjoyed the personal stories more than anything else. They were useful in weaving together my story of AI.

David Gardner: I'm curious, did writing the book change any of your own views?

Vasant Dhar: That's a great question because in a sense, I felt almost like an LLM when I was writing the book. I had this near-death experience on April 15, 2024, which I described in the book, and I was in complete shock. But what I did during that week just after the accident is I thought about what's important to you in life? You could have been dead. One of the things I thought to myself is, I will regret not having written this book because I've been thinking about it for five years. It's been in the works for a long time, and it would be a shame if I didn't write it. In that week, I sat down and wrote the first two chapters of the book. That was like, I have the first two chapters. I know the roadmap. I know which way it's going. But it reminds me of a judge. I don't know whether it was Sam Alito or someone else who once said, Supreme Court judge who said, sometimes the writing cooperates, and sometimes it doesn't. A lot of things become clear in the process of writing.

I'm going to go off on a bit of a tangent here, if you don't mind, which is that I will never use the AI to write or edit anything, and I tell my students the same thing, develop the writing muscle because writing is not just writing. It's a process of gaining clarity. That is, the process of writing gives you clarity, and if you just tell the machine to write it, you'll never experience that clarity that emerges from the process of writing. That's really important to me is to write myself, because that clarifies things in my mind, and it was the same thing for my book, even though I had a reasonable idea about the roadmap of the book, I did not have Chapter 11 planned when I wrote the first two chapters. I knew that I would write a chapter on governance, but by the time I had written all the chapters preceding it, the boomer and boot and truth and trust, and by the time I got to governance, I knew exactly what I wanted to say. But writing the book was a process of achieving clarity, as well, even though I knew what I was going to write.

David Gardner: How will I know what I think until I write it down? I can't remember who said that, but that's something I've always felt as a writer myself. I do want to push back slightly on one aspect. Again, we all have our different standards and codes of conduct. For me, having written my original manuscript for Rule Breaker Investing, I asked AI to do what Mr. Craig, my 11th-grade composition teacher, did on my compositions back in high school, which was to strike through with a red line words that didn’t add. At least for my own experience, having an editor, much like a human editor, was, I think, very helpful for my book, help me save. It turns out one in every eight words. Therefore, I could add content back that I wouldn't have had otherwise. I've said, and you may disagree with this, and if so, I support that. But I've said to a lot, I wish every writer would actually take the time to do that because it would be a lot more efficient for many of us to read through without turns out me being so chatty in my original draft with my readers. I also want to say, Vasant, that anytime we consult a Thesaurus, we're essentially saying, help me think third party aid, improve this text for my reader. I personally don't have any problems. In fact, I encourage people to use ChatGPT as a copy editor, much as I also encourage people to use Claude or ChatGPT to challenge our assumptions and our dearly held beliefs. But I don't know if you have any thoughts back on that.

Vasant Dhar: I do. I completely see where you're coming from, but this is a very fuzzy line. When you say copyedit this thing for me, I guess where I would push back in turn is to say that it's OK to use it for grammar, for spelling, for specific kinds of things that you're concerned about. But the moment you ask it to edit stuff for you, I think you're entering murky territory because it can make the slightest edits that could completely change the sense or the spirit of what you're trying to say. It might actually be better than what you want to say, but it's not yours.

David Gardner: Yes.

Vasant Dhar: It's not yours. It came from the AI. To me, the line gets a little bit murky.

David Gardner: Let me make sure I hasten to add that I would never and did never allow the computer, the AI to simply do the work. I would say suggest edits for this paragraph, suggest ways to save or be more economical with this page. Of course, I use judgment with every single one, and I accepted probably the majority of them, because in many cases, I'm that is actually a more efficient or elegant turn of phrase. In so doing, I I was learning. I was becoming a better writer in some senses, because I was seeing.

Vasant Dhar: Yeah.

David Gardner: It's not very different from handing my manuscript into in my case, at Harton House my very talented editor, Craig, who also gave me additional thoughts. In the end, I made choices on every single word and sentence in the book, but I don't even want to get carried away here with a pedantic discussion around what edits are legit or not. But I do very much agree that you shouldn't have had ChatGPT write this book. I'm glad not. I know you hold yourself to a very high standard, where at the bottom of your newsletters, you say, nothing here was written or edited by AI. I know that you also are a proponent of that for your students. I understand that.

Vasant Dhar: Yeah. Fair enough. Like you said, we could say a lot about this, but I think the marginal value would be quite low.

David Gardner: Well said, marginal value is a good concept from economics. Let's keep going. Vasant has graciously accepted to play our game buy, sell, or hold in just a little while. But before we go, there are a few more questions. Some of maybe our most important or deepest questions, even though we'll probably be economical with our conversation at this point. I've so enjoyed this conversation, Vasant, and I have to have you back just to shoot the breeze. You don't have to write another book, but we need to have another conversation in the year ahead.

Vasant Dhar: Done.

David Gardner: Let's go with will AI exercise my brain or make it lazy? In some ways, we just had that conversation. But you wrote, at one point in the book, the key question before invoking AI is, will the answer exercise my brain muscle or make it lazy? Yeah, I use it every day myself, Vasant. I love it. But I don't want to outsource the very thinking that makes me, me. How do you personally distinguish between AI as an amplifier of human intelligence and AI as an anesthetic for it?

Vasant Dhar: As I say in my book, there's this potential for AI to bifurcate humanity, that it'll amplify people who already know a lot. When I was talking to Claude and OpenAI about what are good investments, it got really deep, but I was really scraping the bottom of my own barrel to talk to it. It was really challenging me, and like I said, it's no pushover. It comes back at you. It was taking me on, as someone who has depth in AI and in finance, it wasn't shrinking. It was exercising my brain. On the other hand, I can imagine that if I didn't know anything about finance and I ask it, like for top 10 stocks, and I don't have the chops to really tangle with it and really get into its thinking and push it back and stuff like that, I could end up just trusting it blindly and making the wrong decision and making the wrong choices.

I think it's doing both of those things, and the thing I encourage my students to do is to stay on the right side of this bifurcation that to make sure that they're exercising their brain as opposed to outsourcing the thinking to it. I may have mentioned on the side that I graded a whole bunch of term papers this semester, and the average quality was better than it was last year, but that's because Claude Code didn't exist last year. Then some of the people who written papers that I didn't understand, I sent them a message saying, can you explain to me in simple English? Some of them responded with AI, which was interesting. I asked them, why did you respond with AI, and they said, Well, we trusted the response more. I said, we've got to talk, because that makes me uncomfortable. You should always be able to hold your own in a conversation and be able to use the AI to sharpen your skills so you can do that, hold your own in the conversation. The answer is, I think it cuts both ways, and that choice as to whether we are one side or the other is really a choice, but we need to get our knowledge to a certain level to be able to be on the right side of this divide.

David Gardner: A lot of it, I realize you are talking about knowledge, but I know you also, I think you say this in the book, so I assume you agree with this addition of mine, which is that if we're being bifurcated, it may not be purely knowledge. Maybe it's just intellectual curiosity, a desire to know, a desire to explore, a desire to learn.

Vasant Dhar: Very much.

David Gardner: So much of AI at this stage is about prompting. I wish that I were Vasant Dhar. I wish I knew how best or how most interestingly or generating trust out of it, the best ways to ask AI to help me out because I know you and your students are doing it better than the rest of us, but a past guest on this show, and I'd love to hear him on your show at some point, Warren Berger, author of the book, A More Beautiful Question, at one point in his book, lionizing the concept of questions, question making, sense making through questions, brainstorming, he would say, is better done as question storming. Rather than be in a business meeting and I'll sit around trying to come up with an idea he says, everybody around the table try to come up with an even better question, an even more beautiful question, and in a lot of ways, that's what we're all being challenged to do to make the best use of AI these days is coming up with a more beautiful question.

One of the things I learned about in your book, among many was, I'm going to try to get this acronym right, RLHF. Now, I know you know what this is, although I won't hold you to remembering exactly what the acronym stands for, but it is reinforcement learning human feedback. Could you briefly explain how ChatGPT uses human beings to decide for the rest of us, what would be inappropriate for AIs to express?

Vasant Dhar: There's two things I want to say, refer to something you were saying earlier that absolutely it's about asking the right question. The other thing I'll say is, are limited by the number of hours in the day to be honest. Because if I had a huge amount of time, I would talk to the AI all day and learn all things. I'm going back to it and learning about thermodynamics, things that I studied in engineering school, which I actually went into because I had a conversation with a smell expert who has a quantum theory of smell, and so I had to learn all of that stuff. But it involves going back and learning all those things. Which is fascinating. It's so much fun to go back and learn these things properly because I felt like I didn't learn them properly. It was hurried. I was taking six courses and trying to handle the exam and have a social life and everything. But we're limited by the number of hours in the day. This thing is such an amazing oracle at our fingertips. I just want to say this because to your point, curiosity is absolutely essential. If you're curious, there's nothing stopping you because you can learn virtually everything you were from a human being from the AI.

David Gardner: Isn't that a beautiful thing to reflect on?

Vasant Dhar: Absolutely.

David Gardner: I love hearing that from you. Thank you. Listening to you talk about only so many hours in the day, Vasant, it just makes me think back to your mentor, Herbert Simon, because a lot of his Nobel Prize-winning work was all about, I think was bounded rationality which is that, for most of us as human beings, we do have limited time, limited resources, and attention. We satisfies a lot of the time. A word that he coined, which is that we are making the best decision that we can, given the very limited time and circumstances we find ourselves in. Most humans, that's good enough for you and for me. I think this is the right connection to make with what you just said, Vasant because Herbert Simon was the one who really turned the world on to realizing that we're not all perfect rational human beings doing what the economists before him were saying.

Vasant Dhar: David, thank you so much for actually bringing that thread in because one of the things I hadn't explained was what Simon was actually famous for, which is that we don't have infinite attention, time, resources. We're bounded, our rationality is bounded. We take the first acceptable alternative and go with it. That seemed so obvious in retrospect, but the entire field of economics is based on complete rationality. Simon came from left field and said, the assumption underlying economics is flawed. Economists didn't like that. They said, you're right. Thank you very much, and let's just carry on. On the other hand, people in AI loved it. Because that's what intelligence is really all about. It's about learning as efficiently as possible and learning enough to solve the problem. Like I said, if I had an infinite number of hours in the day, I would be at the machine all day learning. We're limited only by time and our imagination and our curiosity.

David Gardner: Well said. My favorite conversations, especially for authors in August, end up being nested conversations where we go down a rabbit hole, and then I want to pull back to where we were. Where we were, Vasant, I'm going to quote from your book a little bit, just for background, because I didn't know this. Here we go. You wrote, “One of the less discussed aspects of LLM applications, such as ChatGPT, is that their responses are shaped heavily by humans. Using a process called Reinforcement Learning Human Feedback, R-L-H-F. Armies of humans have been employed worldwide to enforce guardrails around the LLM to ensure that it doesn't spew out things that are untrue, racist, sexist, or offensive, which violate our current social norms. The human enforcers channel the behavior of the machine to bring out the desirable parts about what it has learned and suppress the undesirable parts." That's from your book.

Now, of course, there are many directions we can go, and we're near the end of our conversations, so we won't go too many directions. But one of the things you contemplate near the end of the book is, different cultures with different values will therefore surface truths that other cultures might not, might want to hide, and vice versa. In some ways, that human factor is already implicit in the AI. Of course, the AI has scraped us and learned so much about us. Anyway, we're already there. But I was really interested to hear about what I imagine to be a tribunal or committee of people sitting around saying, I don't think ChatGPT should say that.

Vasant Dhar: This is a great point, and it actually points to the biases that humans have, that we actually impose on the machine. We sometimes forget that we're saying, no, that's not acceptable, but maybe that would have been acceptable 50 years ago, or maybe that's acceptable in China, or maybe it's acceptable somewhere else. It isn't acceptable to us. This is a bias, a cultural bias that we impose on the AI. By the way, I don't know if you know this, but DeepSeek is very defensive. You can't talk to it about Tiananmen Square. In general, DeepSeek doesn't like Taiwanese companies.

David Gardner: Great example.

Vasant Dhar: TSMC. There's those biases that are embedded in the Chinese LLMs. That's something we need to recognize. That all of these LLMs are biased in a way that we find acceptable on average, and that's what these humans do all day. They beat the AI over the head and say, no, thou shalt not say that. You will always say this. That's how we get the machine to produce responses that, on average, we will find acceptable. In a sense, that's a limitation of the AI as well, a big one.

David Gardner: It really is fascinating. Again, just as your book does, especially in the final chapter, we're just as much raising questions at this point in our conversation rather than trying to answer them. But being reminded of the inherent bias, sometimes conscious, sometimes unconscious, that means that one person's or culture's AI could be quite different or have different views of the truth from another person's or another culture's is worth reflecting on.

Before we go to buy, sell, or hold, let me ask the question I've been burning to ask most of the interview, and feel free to hold forth with all the wisdom that you have on this if, in fact, this question can be even answered. Here it is, Vasant Dhar. Let me actually suppose something first. You talked about how AI has been largely dedicated to predictions for the last few decades across many different fronts. Let's make a prediction now. Let's suppose that AI actually goes extraordinarily well. I know you're an optimist. At least you invoke the O word near the end of your book. I am also an optimist. I feel like we are outnumbered by people who live in fear of new technologies, especially such a plate tectonic shift, given artificial intelligence and what it means for the future of humanity and the world. But let's suppose it goes extraordinarily well. It's not the terminator, not catastrophe. It's success. Here's my question. In that world, Vasant, what are humans for?

Vasant Dhar: Humans are for what they've always been for. Which is to exist. I don't know; we're getting philosophical here. But this almost gets at, is there a meaning or a purpose to life? I just think that there isn't. That life exists for its own sake. That it hasn't been designed for a purpose. We create purposes, we create objectives for ourselves, and we express them. That's what it means to be human, to be original, to think, to be, to exist, to create. That's what it means to be human. I don't think that needs to change. It can change. The fear is that that can change. It may make us lazy. It may make us feel incompetent, inadequate. But I don't think that's inevitable. As an optimist, I feel like, if every kid growing up has access to ChatGPT, wouldn't you have been thrilled to have access to something like that? Tremendous power, but it also needs to be harnessed because it's so powerful that it's a double-edged sword, really, is the right metaphor to use. The trick for humans is to avoid that other edge of the sword, to build on the good one and avoid, that is the challenge facing us. But it doesn't really alter the purpose or the meaning of what it means to be human. To me, that remains unaltered.

David Gardner: A lot of conversations these days are about jobs that will be lost, jobs that are being lost, jobs that we imagine going away altogether. Often, we're not as good as human beings at imagining what the new jobs will be. But I'm very confident many interesting, in many cases, better jobs than the ones that they're replacing will show up, and we'll see what forms those take. But what human beings are uniquely valuable for? That is a question that we will dangle out in front of Rule Breaker Investing listeners in order that we might each come up with better answers, new answers in some cases. To help us understand a future that is so technology-fueled and filled. It'll be interesting to watch. Vasant, again, if I haven't already plugged Thinking with Machines, I'll do it one more time. I highly recommend this book to all of my listeners. Vasant, are you ready to play some Buy, Sell, or Hold?

Vasant Dhar: I am. Go for it.

David Gardner: Thank you. Again, these are not stocks. However, if they were, would you be buying, selling, or holding, and a sentence or two as to why? Let's start. First up, AI companions. People forming genuine emotional relationships with artificial intelligences. This thing that's happening, are you buying, selling, or holding?

Vasant Dhar: I am selling.

David Gardner: Why?

Vasant Dhar: Mental health is one of those areas where I feel that feelings really matter. When you're talking to someone about something deeply emotional, There's a expectation of a certain connection, that they understand and feel you and can respond with that in mind. AI doesn't have feelings. It can simulate feelings, but feeling is one of those things where it's not OK to be able to simulate it, like finance. Finance, it doesn't matter. Finance is not about feelings. It's about making money. It's about understanding markets. It doesn't matter how you feel about it. There's an objective reality there. But feelings are just very, very different, and they're very inherently human. To think that an AI can feel for you is very risky. I'm deeply suspicious about forming emotional bonds with AI with an expectation that they really care for you. They don't. They don't care for you. They only simulate that, and that matters when feelings are involved.

David Gardner: A strong sell from the AI pioneer. This might be an all-AI buy, sell, or hold, because I'm just fascinated from many angles in your book talks to many of them. I'm not sure you added this one into thinking with machines, though. Vasant, next up buy, sell, or hold an AI seat on the board of directors. Not merely advising management, Claude, but how about Claude as an actual AI board member?

Vasant Dhar: Buy. Strong buy.

David Gardner: Why?

Vasant Dhar: Because being on a board is about understanding a business. It's about absorbing all possible information, analyzing it, and coming up with answers. That's a real forte of the AI, very different from feelings. This is cold, hard facts, thinking through scenarios, thinking through different assumptions. AI is great for that, for helping people think through that. In my mind, it extends the rationality of the boardroom, in a sense. With humans, all humans in the boardroom, we have 20 people with bounded rationality. Along comes an AI with anything but that, and that can only be additive. Now, I wouldn't want the entire board to consist of AI [LAUGHTER], but an AI member, absolutely. Strong buy.

David Gardner: Strong buy. Next one up, AI-generated art winning major human prizes buy, sell, or hold?

Vasant Dhar: Buy. It's like music, art. Different sides of the same coin. There's creativity involved, but there's a lot of patterns as well. Like when I talk to my musician friends who are professionals, they'll say, look, it's all about patterns. It's all about stringing patterns in new ways. There's an infinite number of combinations, but the machine can explore them and actually find ones that resonate. That's a strong buy.

David Gardner: The four-day work week, if AI is making us more productive, can we cash out a little bit more leisure rather than simply more output? Buy, sell, or hold the four-day week?

Vasant Dhar: Absolute buy. Again, and I think COVID set that in motion. Because the technology was good enough. We were able to work from home several days a week. That's become a default in many workplaces. People expect you to be more productive overall when you combine coming into work, dealing with humans, versus working from home; I'd say that's a buy.

David Gardner: We didn't have time to talk about this next one, even though we spent a good hour talking to each other. There are just too many things. But healthcare. Vasant, an AI doctor as your primary care physician, buy, sell, hold?

Vasant Dhar: Buy, strong buy.

David Gardner: You talk a lot about this in your book, so I gave it short shrift in this interview, but health is just one of those areas that obviously is so important and so helpful we think for AI.

Vasant Dhar: Absolutely. I mean, health and education, to me, are strong buys. I might even see an AI professor in the future who can do a better job than I can.

David Gardner: Great one. In that case, I'm going to skip my next one, which was AI tutors replacing most classroom lectures [LAUGHTER]. Let's move on. I've got two more here. Digital twins. Vasant, you predict that major leaders may eventually have AI versions shadowing them and even conducting virtual strategy sessions with other executives' twins. I assume this is still a strong buy, digital twins.

Vasant Dhar: Absolutely, strong buy. Again, because you can bounce things off a version of yourself. You can even bias your version and say, OK, be conservative or [LAUGHTER] be liberal or whatever. You can bias it and say, now, talk to me in that mode with the same knowledge base, so, tremendous resource.

David Gardner: Vasant, do you have a digital twin?

Vasant Dhar: I do not, but I'm actually considering creating one for giving exams and tests. I see that as a really strong use case because, as it is, I'm thinking back to going back to blue books and having people write answers because taking exams with AI has become completely meaningless. You don't keep people take home exams anymore. That's absurd. If you want to test how people will be able to hold their own in conversation, for example, you need to be able to talk. When I was in engineering school, we used to have a thing called the Viva, which was, you show up, and five professors would have a conversation with you about chemical engineering, or whatever the subject was. It was terrifying but tremendously useful as a learning tool. I'm considering a digital twin that would be able to conduct exams, quizzes, things like that.

David Gardner: Sounds great to me. Last one. Here it is. The Turing test. Still an interesting benchmark? Would you say in 2030, or is this an increasingly irrelevant relic? Buy, sell, or hold. The Turing test.

Vasant Dhar: Sell. I think the Turing test is largely irrelevant at this point. I would argue that machines have actually passed the Turing test. Very often, we can't tell the difference. I think that ship has already sailed.

David Gardner: A bonus one, because I mean, this is Rule Breaker Investing. How can I not ask? Buy, sell, or hold? Human stock-picking, Vasant Dhar.

Vasant Dhar: I will still rule that as a buy for now. But as I'm about to write in my next newsletter, the DoMore Bot in my mind. To me, that's the holy grail of investing is where the machine has ingested all of our wisdom. It now has access to all the data, the news. To me, the holy grail is that in the long term, the machine should be able to do it, but that's a long ways off. I think humans have a big role to play for the foreseeable future.

David Gardner: Vasant, thank you. What a privilege to spend some time with somebody who's at a front row seat to artificial Intelligence for more than four decades and who's still looking forward, still asking questions, still thinking about not just what these machines can do, but what we humans should do with them? Thinking with Machines, by the way, just a wonderfully accessible invitation. I love how you expressed why you wrote it and who it was for earlier. I'm grateful that you've helped all of us think a little bit better alongside the machines this week. Thank you, Vasant Dhar, and Fool on.

Vasant Dhar: David, thank you so much for this conversation. I really enjoyed it. I'm super glad you enjoyed my book. Coming from someone like you, that matters a lot. I would love to continue this conversation. It's always delightful talking to you. Thank you so much.

David Gardner: Continue it. We shall. Fool on.