Interview with NYU Professor Vasant Dhar: Thinking With Machines

0:04My fear is that we are slipping into a Huxleyan kind of world, perhaps even without our realization, right? that we are gradually disempowering ourselves in many areas of our life. The machine has become a gatekeeper of human activity in many ways.

0:25That was NYU professor Vasant Dar, author of the new book, Thinking with Machines, The Brave New World of AI. I'm Motley Fool producer Mac Greer. Now, Motley Fool analyst Asit Sharma recently talked with Professor Dar about that brave new world. Greetings, Fools. I'm Asit Sharma, Senior Analyst and Lead Advisor at The Motley Fool. And my guest today is Vasant Dar, Robert A. Miller Professor of Business at NYU's Stern School of Business. Professor Dar is a pioneer in the field of artificial intelligence. In fact, he received his PhD from the University of Pittsburgh with a specialization in artificial intelligence in 1984.

1:08Among his many achievements, Professor Dar is noted for bringing machine learning to Morgan Stanley's proprietary trading groups in the 1990s. You may have listened to the professor's popular Brave New World podcast, and he's out with a new book entitled Thinking with Machines, The Brave New World of AI, which is the topic of today's discussion. Vasanth Dar, welcome to The Motley Fool. Thank you, Asit. Delighted to be a fool. Awesome. Well, I wanted to start with your early childhood, which you recount in the introduction to thinking with machines. You were born in the 1950s in Kashmir, India, and you note that you rode to school in a horse-drawn cart.

1:47You also moved around quite a bit in India, and by the time you were nine, your father was posted to Ethiopia on assignment as India's military attache to Africa. So I wondered, Professor, can you tell us a little bit about these formative experiences and how they helped shape the person and scholar you became? You know, all amazing experiences growing up, including, you know, what you mentioned in Ethiopia, you know, my, and I described this in my book in a humorous kind of incident, you know, my mother put me in the wrong grade. She put me in seventh grade instead of fourth grade by mistake and only realized, you know, her error six months later when it was too late to do anything about it.

2:24So here I was hanging around in class with 15, 16-year-olds, and I was like nine. So that was a hell of an experience growing up. Then I went off to boarding school in India after that, which was also another. So my trajectory was third grade, seventh grade, eighth grade, and then six, seven, eight. You cannot make this up. That's what had happened. But it made me resilient, I guess, in some way. And it was a really unusual kind of upbringing. I'm happy for it. So fast forward to Pittsburgh, Pennsylvania, at a time where you were attending school and intersecting with a very exciting world, the very nascent world of artificial intelligence.

3:06You met an AI pioneer in Herbert Simon who had received the Nobel Prize in Economics for his work in revealing the limits of human rationality and decision making. Now, Professor, I remember still in the early 90s, late 80s, early 90s, taking a college class in microeconomics in which rationality was still the governing principle or said to be the governing principle by which most humans make their economic decisions. That's right. But Professor Simon had a different idea. He called it bounded rationality. I wondered if you could explain that to us. Well, essentially what he said was that humans have limited cognitive resources, that we are not able to, you know, enumerate all possible alternatives and evaluate them.

3:54That's just like too taxing. You know, we'd never get through the day if we did that and that our attention is limited. And then we tend to focus on the most plausible things to pursue, you know, and we do this through heuristics that are learned through experience. And so heuristics actually sort of focus our attention, you know, to the right parts of the problem. And when we find an acceptable choice, we take it, you know, and we move on. Right. So that was his theory, which was called bounded rationality. But I have to say that economists sort of said, yeah, that's that's true. But let's just move on.

4:24So for the most part, you know, they still sort of, you know, because it doesn't lead to very good theories. Right. I mean, it sort of messes up sort of nice mathematical. It's messy. It's messy. And economists don't like that. So, you know, it was just, yes, it's true, but thank you very much. Whereas his ideas really sort of took root in artificial intelligence, you know, which was really all about, at that time, all about, like, how do you represent knowledge and how do you traverse it intelligently? And that was called heuristic search at the time. And so heuristics became big in AI and they were the sort of primary paradigm at that time of expert systems where we tried to build these impressive applications in areas like medicine, where you would extract knowledge from experts and use the heuristics that they had acquired through experience to actually do medical diagnosis.

5:16And that was my first real experience to AI, just watching this system called internist interact with an expert and elicit information and arrive at the correct differential diagnosis. I mean, I was just watching this and it just blew my mind. And that's when I decided this is what I'd like to do with my life. You posit that oftentimes success in the markets or in other probabilistic endeavors is made up of small edges that compound, compounding small edges. And you bring up the commencement address of tennis great Roger Federer last year to the graduating class of Dartmouth. Can you start with what interested you in that commencement address and explain the concept of compounding edges to us, please?

6:04The statistic that Federer said that really sort of stayed with me, and it's so similar to financial markets. So I view financial markets and sports as being sort of two sides of the same coin. He said, over the course of 1 ,526 matches, I won 80 % of them. What percentage of points do you think I won? And he paused and he says, 54%, barely better than even. In financial markets, you do 54%. you should be managing the world's money, right? As long as you're winners and losers of equal size, right? But what Federer was really saying is that, you know, it's that little edge that just compounds over the course of the match, right?

6:46If the match was just one point long, then Federer would win 54 % of his matches, right? But the fact that it sort of goes on over time means that he's got time to regroup, even though he loses a point, right? It's that little edge that just sort of keeps multiplying over time. And so the longer the match, the more matches he'll win, of course, as long as he doesn't get exhausted, right? So stamina also matters. Boris Becker, by the way, won almost 80 % of his matches with only like less than a little over 52 % winning points because he had a tendency to win the really important ones like tiebreakers.

7:18But that's the point is that you don't need to be perfect. You don't even need to be really good. You need to be just slightly better than the average or some benchmark in order to be successful. And that applies to almost everything in life. As long as you're slightly better, that edge will just continue to compound and that you'll get better and better in your outcomes. Hey, I'm the Sparfuchs from sparsim.de. You're a little bit of a smartphone, but your current phone-to-handy-tarif is more like a toxic relationship? Then I have something for you.

8:12Do you think that some of the same principles you've applied to systematic investing on a short-term basis, where you're looking for a higher probability trade with a shorter duration apply on the other side to long-term investors like myself? They do. And for the reasons that you pointed out, right, that you need numbers. And in fact, in 2015, I went to my colleague Aswath Damodaran because I sort of believed that machine learning and quant methods really applied to short-term trading, where you could identify an edge, where there were lots of numbers involved. But it was hard to apply to long-term investing with holding periods of many months or even years, because you just couldn't get enough sample size, you couldn't get enough training data.

9:03But I was really intrigued by my colleague, Aswad Damodaran, who's considered Mr. Valuation on Wall Street. And so I went to him in 2015 and we had this conversation about whether it would be possible to create a bot of him. And I'd had a similar conversation with my colleague, Scott Galloway at the time, should you trust your money to a robot? I'd just written this article, should you trust your money to a robot? And I made the case that you should when it comes to high frequency trading and short term, but when it came to long-term investing, that it was impossible to train a machine like you could with shorter duration stuff.

9:40I remember Scott, at the end of that conversation, saying, okay, so what you're saying is that trading flows will disappear, but venture capital and private equity is safe. I said, yep, that's pretty much it. My conversation with Damodran was similar, that it would be too hard to actually try and replicate him. What's interesting is post-Chat GPT, we sort of revisited that question, you know, and so I went back to the mother and I said, you know, do you think we could actually build a bot of you now, given this new technology? And he said, sure, let's, let's give it a shot. You know, you've got all my training data.

10:15And so that's what I've been, you know, involved in for the last couple of years. You know, we've built this bot that's designed to think like him. And, you know, my initial thinking was that we could apply that systematically as well, that we could just apply Damodaran to the S &P 500. It's impossible for him to do it because he can't evaluate 500 companies in a day or even in a week. It's just like too much work. But my thinking was, if we can build a machine like him, why can't we just apply it to the entire index and then use it systematically? It's an interesting idea. It may actually work.

10:59But I've actually become intrigued with a different type of application of the bot, which is something that allows people to think and reason about companies in a deeper kind of way, to run scenarios and say, you know, what if Trump escalates tariffs? Like what will valuation of Apple or NVIDIA, whatever, look like? Or what if this is tariff was a head fake and we go back to, you know, sort of the era of low trade barriers? This kind of stuff is very laborious for people to do and very time consuming. I find it sort of interesting that we can apply AI now systematically to long term investing as well.

11:36What was the thing that surprised you most about the demotor and bot? So basically, you had access to all the training materials, public, famously public materials. And you also had access to Professor DeModeren's very elaborate write-ups, his blog post, which they themselves, if you marry up the public spreadsheets that he has for investors, they're an object lesson in how you draw together numbers and narrative. what surprised you most in this latest phase post-ChatGPT where you took more modern tools let's say or more contemporary tools and recreated the idea maybe the biggest success you had or the biggest pitfall that you didn't expect?

12:20You know when I started this two years ago with one of my colleagues Jaume Sidok who's a LLM person we had no idea whether this would work. It was a wild idea you know, to build a bot like him, you know, and we tried what most people might try, which is, you know, give all his valuations to an LLM, fine tune it, and then have it, you know, think about a new case. It just didn't work. It didn't sound like him. There was nothing deep about it. There was nothing profound about it. So we just sort of went back to the drawing board and said, let's just identify all components of his thinking. You know, fundamentally, he's got this quantitative model that he calls the Ginzu.

13:00That's like... It's a spreadsheet, right? It's a spreadsheet. I've used it. Yeah. It's incredibly complex. It has all kinds of switches and context and all that kind of stuff. But at the end of the day, it's a quantitative model. Inputs gives you evaluation, you do a sensitivity and there you have it, right? But the question is like, how do you marry a story to the numbers, right? Like what's the story that is consistent with the numbers? And the story, involves sort of stepping back from the particular company. So I'll give you a great example. So when he evaluated NVIDIA in 2023, the first question he asked, I call this a framing question, was, is AI an incremental or a disruptive technology?

13:40Now, why would you ask a question like that? Well, you ask a question like that because the markets in those two scenarios tend to be very different. If it's incremental, it's pretty well-defined. You can put a boundary around it. If it's disruptive, it's much more uncertain, right? You need to think about what that really means. Disrupting what? Every industry? Is AI like electricity? Is it like the internet, right? So it makes you think about the problem in a really broad kind of way, right? And then his subsequent question, when disruptions happen, what's the distribution of winners and losers?

14:14And he shows that you get a few winners and lots of losers, lots of wannabes. And he says, okay, I think NVIDIA is going to be a winner. So they're going to have a dominant position. And then he goes about sort of thinking about it, like what are their margins going to be like? Well, and he says, well, what are the margins of people in the semiconductor industry? Well, that's a good place to start. And the work of Phil Tetlock, by the way, also applies here. He has this work on super forecasters, what makes them good. And what makes them good is that they anchor themselves in the right part of the problem, as opposed to a biased part of the problem.

14:48They tend to be sort of relatively unbiased. And so I realized that Aswad Damodaran was what I call a super forecaster, right? He just has those properties of what Tetlock calls, you know, super forecasters, the ability to really ask the right kinds of questions, you know, and insatiable curiosity of anchoring himself. Hi, I'm Neil. And I'm Ken. And we are from the Triviality Podcast, a pub trivia-style game show where a lack of seriousness meets a little bit of knowledge. Join us each week for an hour-long game of general knowledge trivia featuring special guests from around the world, plus tons of extra themed episodes.

15:23If you want to improve your trivia game, or you just want to scream at us in your car when we get easy questions wrong, then we're the show for you. Find triviality on all your favorite podcast apps. But you know that, because you're already listening to a podcast. If you had a scale today, where would we be wading more towards that we will govern AI or AI will govern us and why? You know, my fear is that we are slipping into a Huxleyan kind of world, perhaps even without our realization, right, that we are gradually disempowering ourselves in many areas of our life. The machine has become a gatekeeper of human activity in many ways.

16:04You apply for a job, you're screened by the AI, you might even be interviewed by the AI increasingly these days. It's not a warm, fuzzy feeling when the machine has become a gatekeeper to human activity. So my fear is that we might just slip into this without the machine sort of having evil intentions or being programmed to do harm, that we just sort of slip into this without our explicit realization. That's really my concern. Which stakeholders do you think would be important to ensuring that we don't slip into such a future? I mean, obvious answers would be, okay, governments, perhaps we need regulations, academics, also big tech maybe, but I don't know, what about people who use the machines as well?

16:54Who are the stakeholders that should put a voice forward in this decision? Well, they more than anyone else, like everybody, right? And that's why I wrote the book for everyone. I meant for this book to be accessible to everyone because this applies to all of us. And I tell my students this as well, that it's easy to use this technology as a crutch. It is so tempting to use it as a crutch. but that'll in the long run will be debilitating, right? You don't want to go down that road where you've got a question and you just throw it to chat GPT and say, what do you think? Because that's the surest way of going into cognitive decline.

17:34You know, and I can feel it, by the way, when I use maps, I don't think I navigate as well spatially as I used to. You know, I think I've lost that facility by relying more and more on maps, you know, and I'm aware of that. And I now try and navigate myself manually sometimes just to sort of keep that spatial mental muscle alive. And that applies to all areas of our lives. And so individuals, more than anything else, really need to ask themselves how they're consuming this technology. I mean, as it is, my colleague Jonathan Haidt says that some of these social media platforms have caused tremendous harm to teenagers.

18:06We ain't seen nothing yet in terms of the potential harms that AI could cause if we just let it go unfettered. And it's a tough area because as someone said, I mean, I think I was reading a piece by Ezra Klein this morning where he said, you know, who are we to tell people what to consume, right? I mean, and Sam Altman said, we don't want to be the moral police of the world. You know, we'll open ChatGP2 to adult content. All true, you know, all fair, but that's why it imposes the burden really on the consumer. And so among all these people, the burden really is on the consumer to be aware of how you're consuming AI.

18:45And as I say in my book, you can consume it to become superhuman, right? It can really serve to amplify your skills if you use it in the right way. But if you become dependent on it, it'll lead to cognitive decline. And that's no good. And that's one of the points I try to make in my book is how to think about that, how to think about being on the right side of what I see as this sort of impending bifurcation of humanity. I think one of the clearest examples of all this is a choice you make that you describe in the book. Some people ask you, why don't you use your chat to you to write the book?

19:16And you say, well, right now the machines don't write as well as us for now. Okay, I get that. But I think also underneath that is the desire to express yourself in your own unique style to make the points that you want to make and to have your expression, which is beautiful, by the way. It's a great expression and well-written book to be the statement that you put into this work, not to rely on the crutch just because it would be easy to input some bullet points and perhaps spit out the product and you go talk about it. It's a world of ideas that you are putting forward. So I really appreciated that part of your book, which was, hey, there's a reason that I'm writing this myself.

20:03Exactly. So, by the way, thank you for that compliment. I really appreciate that. But, you know, in addition to the fact that I think I write better than ChatGPT and I want to express myself in my own style, it's also so much more fun, right, to do that. And at the end of the day, what's life about if not for having fun? I mean, life is about having fun and this should be fun. And I had so much fun writing it. And there's so much of a sense of accomplishment and satisfaction from producing something good by yourself. And that's what we should strive for. Professor Vasantar, this has been an extremely illuminating conversation.

20:40And above all things, it's been a lot of fun. I really appreciate your time today. And I hope that you will come back for another conversation at some point in the future. I'd be delighted. Thanks so much for this, Asit. I really enjoyed it. Great questions. And I love the conversation. Thank you again. Thanks.

21:00As always, people on the program may have interest in the stocks they talk about. And The Motley Fool may have formal recommendations for again. So don't buy or sell stocks based solely on what you hear. All personal finance content follows Motley Fool editorial standards and is not approved by advertisers. Advertisements are sponsored content and provided for informational purposes only. To see our full advertising disclosure, please check out our show notes. For the Motley Fool Money team, I'm Matt Greer. Thanks for listening, and we will see you tomorrow.