THE AI RECKONING: The Conversation That Will Define the Next Decade

Featuring Prof. Vasant Dhar, Prof. Robert "Bob" Gordon, Jim Bianco, Prof. Campbell Harvey & Edward Chancellor

Most industries are underestimating how fast AI is transforming everything—and financial markets might be leading the charge. Vasant Dhar, a pioneer in AI research, reveals how recent breakthroughs are finally fulfilling the ambitious visions of the field’s founders—and what it means for your money, your job, and the future. In this episode, Dhar shares wild stories from the early days of AI—like how a 1994 experiment with Wall Street traders unlocked predictive patterns still used today. You'll discover how machines learned to find trading signals before researchers even knew why they worked, and how modern AI is now capable of reasoning about different economic regimes, including turbulent shifts like rising interest rates and inflation. He breaks down the evolution of AI from rule-based expert systems to deep learning, and how today’s models are dissolving the boundaries between actual expertise and common sense—something once thought impossible. We explore specific breakthroughs: the transformation of AI from a tool for pattern recognition to one that can simulate human-like reasoning—like evaluating complex scenarios such as trade wars or regulatory changes in seconds. Dhar explains how large language models now hoover entire data sets, including economic regimes and market cycles, to anticipate shifts in unpredictable environments. Dhar warns about the limits of AI during regime shifts, and the danger of deploying models trained on outdated data when the world changes faster than our algorithms can adapt. This is an episode about whether machines will truly understand economic complexity—or simply pretend they do—and what that means for human decision-making. Perfect for entrepreneurs, investors, policymakers, and anyone navigating the era of AI-driven markets. If you're intrigued by how AI is reshaping finance and strategy—yet wary of its limits—this conversation offers crucial insights to stay ahead in a rapidly evolving landscape. Tune in to discover how machine learning’s past, present, and future could redefine your role in the new economy—and whether we’re nearing a genuine breakthrough or just another illusion of understanding.

We close the episode with incredible insights from the likes of Jim Bianco, Dr David Kelly, Prof. Campbell Harvey, Prof. Robert Gordon and Hayek book prize winner Edward Chancellor

Transcript

SPEAKER_07 0:00

You know, what changed with ChatGPT is, you know, in my mind, it became what's called a general purpose technology that now everyone can identify with. You know, it just opened the floodgates. And to me, the fundamental breakthrough, if you were to ask me like what's the one thing that's different about modern AI from you know the sixty years before that, I'd say that it's the sort of dissolution of this boundary between expertise and common sense. If machines finally began to, you know, sort of work with common sense, simulate common sense.

SPEAKER_00 0:28

Welcome to the Most Important Thing podcast, where we speak with the best and brightest in finance, economics, investment and geopolitics to bring you the truth about what's driving the global economy and financial markets. Every week, our chief market skeptic, Dr. Alan O'Sullivan, and former government minister Ivan Yates sit down with a global expert to provide invaluable lessons from a lifetime of markets experience. Economic expansion or recession, market highs or underlying vulnerabilities, AI dystopia or just the next bubble, escalating geopolitical tensions, or simply the new global order. These are deep conversations searching for truth and practical investing strategies to improve your chances in this uncertain market environment. To optimize your learning experience, visit our learning hub at the mitpodcast.com, where you will find the latest guides, investment ebooks, and a community that focuses solely on expert insights.

SPEAKER_06 1:15

Enjoy the show. John McCarthy is very well known to those that are familiar with artificial intelligence and the origins of artificial intelligence in the United States. But Irish people mightn't be that familiar with John McCarthy. His father, John Patrick McCarty, immigrated from a small village outside Clarkon, County Kerry called Croman. So there was a very famous meeting in 1955 at Dartmouth College, I think it was, where famously Dr. John McCarthy coined the phrase artificial intelligence. So on today's episode, it's going to be focused on AI, artificial intelligence. And I've got a great lineup. So the main part of the interview is going to be with Professor Vassant Darr. Professor Darr released a book recently called Thinking with Machines. I'll put links to that in the show notes. And it's a really accessible book for anybody that's trying to get a primer, perhaps, on artificial intelligence, large language models, uh, really accessible across a number of sectors and industries. And as he said himself in the interview, his main aim writing the book was to make it accessible. So we cover a lot in the first part of my interview with Vassant Darr. He's a professor in NYU Stern in New York. And really fascinating interview. Then after the main interview, I got some snippets of some of some other interesting people. Professor Robert Gordon, Bob Gordon. A lot of people will know him. 2016, he was voted in the top 50 economists in the world. He wrote a very famous book, The Rise and Fall of US Growth. And if you want something interesting, go on to YouTube and look at his TED Talk. He has a TED Talk where he talks about the decline or the peak US growth because of four structural headwinds. Really fascinating talk. We have got Jim Bianco, uh one of the best macro guys in the world, uh, one of the best fixed income investors, really sharp, contrarian thinker. Jim gives his views on AI, a bit more of a positive spin on artificial intelligence. I've included David Kelly, uh JP Morgan. Again, he has some good insights uh on AI, which I thought would be useful. We have the winner of the 2023 Hayek Prize, book prize, Edward Chancellor, for his book, The Price of Time. And again, that's an interview coming up, and Edward talks about AI in the context of independent thought insight. And we finish up looking at Professor Campbell Harvey. Campbell Harvey will be well known for one of the most prolific uh publishers of academic research, but he's also an industry guy involved in research affiliates, and well known, perhaps best known for identifying the inversion of the yield curve as a recession warning. So again, a really star-studded uh list of guests and contributors, and I hope you enjoyed it show this week. And one last shout out to Professor John McCarthy, whose father came from the humble surroundings of a small village in uh County Kerry and went on to make a huge impact in the world of artificial intelligence. Okay, Vassant, delighted, uh especially for my Irish viewers, because I've there's a bit of an Irish connection uh that we're going to talk about shortly. But first of all, I want to show this book here, okay? This is uh Thinking with Machines. It's Professor Vassant Dar's new book, and I've just finished it, and it is a fantastic read. And probably the best compliment I could give you, Vassant, is that it's very accessible. Uh obviously, you've been in the area for decades. I mean, you started uh before I was born. I don't want to age you, but I I think 1979 is when you were looking at in the internet uh machine. Well, it reads like a novel, it reads like a biography as well, because it goes through your life story, which is fascinating. I hope we get into that. But the Irish connection for me is Dr. John McCarthy. Because Dr. John McCarthy, to many, is known as the godfather of AI. He's f came from a small village, his father came from a small village in in Kerry, County Kerry, Ireland, called Cremon. Uh he was born in Boston in 1927. Um, but there was a famous meeting in Dartmouth College. Maybe we can start with that, uh, Vasan. Yeah, yeah, absolutely.

SPEAKER_07 5:49

But by the way, thank you, Alan. I appreciate that you like the book. Uh, you know, I've written it for everyone. You know, it's meant for students, parents, grandma, my colleagues, policymakers, everyone. So there's something in it for everyone. And yes, I've tried to make it accessible, and people have told me that I've succeeded. So I'm quite pleased with the product. So thank you about that. Yeah, so let's start with that workshop because you know, one of my mentors, Herb Simon, was also at that workshop. And you know, McCarthy is, you know, known for coining the term artificial intelligence, you know, um, prior to that workshop. So he came to that workshop suggesting that they called it AI. You know, he encountered some resistance, you know, some people had other labels in mind. But I'm glad that he won out because I I think that describes the field. Um and it was a new field, and it was artificial intelligence, and their goals were very lofty, by the way.

SPEAKER_06 6:41

Yeah, and I I was chuckling when I was reading the the preface to your book where Scott Galloway said, you know, a lot of people think um AI is around for the last five minutes, but you've been looking at this uh since the late 70s. Um can you give us a brief kind of background in terms of how you uh fell into the field and how it sparked your interest initially?

SPEAKER_07 7:06

As is often the case, you know, in human affairs, you know, that some of the most important events in our life are purely serendipitous. You know, I mean I was 23, wondering what to do with my life, was writing a program on punched cards for one of my professors, you know, implementing some decision-making process called the analytic hierarchy process. And the student comes to me and says, Hey, you know, Vasant, there's this guy who has been working on AI for 10 years in medical diagnosis. I want to get him to offer a course in AI. And I said, What is it? And he said, Well, it's about computers to be smart. And I said, Well, that sounds interesting. So we went up to the medical school with two other PhD students, and while we were waiting for him, uh, you know, we saw that there was a, you know, he was on the phone, there was a big screen in the middle of the room, and this gentleman um called Jack Myers, who was talking to the computer. When I say talking, he was actually literally talking, and his assistant was typing because no one knew how to type those days. And he was puffing a cigar, and they he was discussing a case with this system called internist. And, you know, it was a real case, and the system asked him a bunch of questions. He answered some of them, he couldn't answer some of them. And at one point in the interaction, it asked him another question, and he said, Why are you asking me this question? And the response completely blew my mind. You know, internist said, because the evidence you've given me so far is consistent with the following hypotheses, and this question will help me discriminate between the top two, you know. And I was just standing there and saying, holy smoke, how is a machine doing this? You know, I'd you know, I'd used computers for engineering. To me, they were calculating machines, not reasoning and thinking machines. Um so that just completely blew my mind. And I was like, well, this is what I want to do with the rest of my life. Um and Harry Popel, who was a professor who'd built the system called Internist over 10 years, you know, we went into his office, had a great discussion. He told me, you know, told us that he'd be delighted to offer the course, which he did. And then a few months later he introduced me to his advisor, Herbert Simon, you know, who was a Nobel law laureate in economics and one of the fathers of AI. So I just fell in with the right people, you know, just serendipitously and and just let life take me along.

SPEAKER_06 9:13

What I really like about the book is the start of it, you go through the evolution of AI from way back in the 50s, 60s, 70s in. And what what was surprising to me was how lofty and how ambitious the goals were at the start. Uh and m is it fair to say that we we just didn't have the computing power, the data uh to be able to make those goals a reality? Maybe talk about the earliest days in terms of its evolution.

SPEAKER_07 9:40

Yeah, so in in retrospect, you know, things are always very clear, but you know, when you're in the moment, it's anything but clear. You know, so the goals were lofty. The the the vocabulary of AI was you know consisted of terms like thinking, reasoning, planning, understanding. You know, those those uh and people actually believed that programs could do all of these things. Um the tools were in retrospect somewhat limited, but we didn't know any better. Um and so you know, we pursued these objectives, and that sort of that paradigm of expert systems where you extracted knowledge from experts uh and put it in the computer ran into a wall, right? Because uh for for several reasons. One is that you know we know more than we can articulate. But the second is that expertise and and common sense don't have well-defined boundaries. You know, physicians use common sense all the time in reasoning and thinking. And yet uh we really needed to define boundaries around expertise and have systems that could perform well in these well-defined domains, like medicine or engineering or you know, tax planning, or you know, things where you could sort of put a boundary around the expertise. So that ran into a wall for several reasons. Uh and then you know, machine learning sort of came to the rescue in a sense because data has started becoming available, and people said, well, why don't we just learn all of this knowledge from examples, right? So you can look at medical health records, you can look at all that stuff, and in theory, you can learn the sort of underlying structure, under learn the underlying knowledge. And that's when I went to Wall Street, you know, early 90s with that same objective of prediction, right? So the emphasis of the field shifted to prediction from data. Use data, build models, predict. And that's kind of what AI was for the next 30 years. You know, that's the singular focus on prediction. We sort of lost sight of planning, thinking, reasoning, understanding, all that stuff. Like it's too hard, you know, let's focus on prediction because this is the way we can make progress. Um and you know, and we did. There were there were lots there were lots of great successes with standard machine learning, and then subsequently deep learning, which was all about um sort of perceiving the environment directly, you know, through sight, sound, you know, language, and now smell, uh, you know, where we said, you know, instead of translating data for the computer, let's just work with the raw data itself and have it do, you know, things like feature construction and all of that, you know, which was sort of a you know pesky problem in standard machine learning to take data, construct features, you know, and that was a somewhat creative and open-ended sort of exercise. And so it got rid of that largely, you know, that deep learning got rid of that. And that was a big step forward, you know, where machines could actually see. So if you were discussing a medical diagnosis case, you didn't need to translate the x-ray image and say dark spots on the right lung, you know, you could actually literally feed the image, right? So that was a big step forward in enabling machines to perceive directly. But it was still AI was still an application. You know, what changed with ChatGPT is, you know, in my mind, it became what's called a general purpose technology that now everyone can identify with. You know, it just opened the floodgates. And to me, the fundamental breakthrough, if you were to ask me, like, what's the one thing that's different about modern AI from the 60 years before that, I'd say that it's the sort of dissolution of this boundary between expertise and common sense. Machines finally began to sort of work with common sense, simulate common sense. And by the way, you know, I was uh you know, I will I observed several projects when I was at MCC in Austin, Texas, you know, at this uh uh lab, there was a whole group devoted to making machines understand common sense, and it just failed. You know, you just couldn't do that. Um and what modern AI did, again, serendipitously, because Google wanted to do sentence completion in Gmail, you know, that required solving a much harder problem. So doing sentence completion involved learning about the world in general, and so that's what AI did. It learned about the world in general, and now you know Chat GPT doesn't know whether it's talking to you um, you know, about something deep or whether it's just something you know involving common sense. It just doesn't know and it doesn't care. And that is quite amazing, you know, you when you think about it. Um and and that's sort of really opened the floodgates to um you know all kinds of possibilities now with AI.

SPEAKER_06 13:59

Yeah, I had an aha moment when I was reading the book, and they're the best moments because my background is, I suppose, wealth management, investing, finance, economics, and I'm dipping my toes as I have to now, and we're forced to into this into this world. But I think it's really interesting the way you start with expert analysis and what the the starting block was we were f you were feeding the machine, okay, and there was a bound there was a bound on that. There was a a limit to that because of it wasn't picking up the nuance or the characteristic nature, okay? But then it became less ambitious and then the machine learning came in where you were actually looking at raw data and the data was giving the the answers, which which made sense because there was uh improvements in computing and technology and so forth. But for me then it penny dropped when I realized that AI today is addressing or has the ability to answer the questions that it sought to answer at the very start. Now, not fully because we have a debate about understanding, but uh is that accurate to say that that we've come kind of full circle as such in terms of addressing the shortfalls?

SPEAKER_07 15:11

Yeah, in in a sense we really have, you know, and you know, one of the chapters in my book is about this uh machine that I called the Damotharan bot, which is uh designed to think like my colleague Aswat Damotharan about valuation. Now, this was to me, yeah, and and by the way, I had done a uh you know, ten years ago I did a small segment with Scott Galloway about, you know, should you trust your money to a robot? And at the end of it, he ended by, all right, Vassan, so what you're saying is trading flaws will disappear, but private equity and venture capital is safe. And I'm, yeah, that's pretty much it. You know, that's a very human kind of thing. There'sn't enough data to train a machine, right? But now I wonder, right, whether that is off-limits to AI. Because with this Demotheran bot, you know, what we've demonstrated, you know, I've done this with one of my colleagues, Jesus Sedok. What we demonstrated is that it's it's actually possible to build a machine that you know does that sort of thinking like uh an expert, like Demotherin, but it was it was it's impossible to reduce his expertise to a bunch of rules, right? That's you know, it's it's impossibly difficult to do that. On the other hand, LLMs have now provided us a substrate, something to build on where we can actually build that functionality. And now what's interesting about him, about the Motheran is that he's published hundreds, maybe even thousands of reports uh on valuation of companies that we can use as training data and built in, you know, built into this sort of multi-agent system. But to come back to your question, yeah, I think that we are now uh addressing those sort of lofty goals of AI, you know, like you know, we wanted machines to think, to reason, to understand, to plan. And now that we have uh a machine that you know has this what I call general intelligence, right? We've we've now what I see is we're in this paradigm of general intelligence where the machine knows something about everything. And I distinguish this from AGI, which I think is a bit of a red herring and a bit of a distraction, personally. Um yes, it would be good if the machine were able to simulate human capability in all these ways, but I don't think that's necessary or even desirable, you know, possibly. Um so we where we are now is this paradigm of general intelligence where the machine knows something about everything, and that something is getting deeper and deeper, you know, by the day. Um which, of course, raises concerns that I'm sure we'll we'll talk about. But yeah, we've come back to those lofty goals now with a new set of tools, uh with a much better chance of addressing them, you know. And one might argue that you know the machine doesn't really understand anything, but that's sort of a little bit besides the point. You know, I I think if it does a really good job of simulating understanding and simulating common sense, then it's practical for a large number of problems, maybe not some, and and you know, we'll we'll talk about those. I mean, when it when it gets to things like mental health and AI companions and stuff, I I get concerned because there I think just simulating is is not adequate. You know, you you you can do a tremendous amount of harm. So um, but yeah, to to to sort of wrap up the the the answer to your question, yeah, we're we're back to those lofty goals now with a much better set of tools and a much better chance of success.

SPEAKER_06 18:22

I definitely want to come back to the uh D bot, that the Motoran bot. I'm a huge fan of Aspot. The motor and by the way, he's been hugely generous with uh the way he provides a lot of the material and his valuation work on YouTube and and other things. Bringing you back to the nineties again, Vassant. I mean, I found it intriguing the experiment you did with the traders. So it's serendipity again, you meet uh Kevin Parker in '94, and I mean, one of the very earliest, if not the earliest, the use of AI for systematic uh trading. Can you speak to your engagement with the traders? And it was in it was it was funny listening to the pushback uh that you were getting while I was reading the book as well. A lot of pushback.

SPEAKER_07 19:04

You know, I mean, a lot of people, you know, because Wall Street sort of the quant at that time was dominated by physicists and econometricians who kind of have their way of looking at the world. And they saw me as bringing this, you know, voodoo kind of set of methods. But, you know, I had had great success with this, by the way. You know, before I went to Wall Street in '94, I'd done some work with Nielsen and had great success finding patterns in consumer data, like really interesting stuff. So I knew that I was on to something, you know, that these methods were capable of finding patterns that humans couldn't. And my intuition was that financial markets were a great test bed to test this out. You know, that markets are complex, they're not easy to understand, you know, the simple rules that people uh say work don't work, and I realized that very quickly in on Wall Street that there were lots of assertions that didn't stand up to scrutiny when you actually tested them against data. So that convinced me that maybe there's something here where we can actually make progress. And Kevin Parker hired me, you know, to um work with his prop trading group, but also to work more generally across the firm and making sense of customer data, which I did, you know, to, you know, with great success. But but I was always I was really fascinated by the prediction of financial markets. And so um I realized that prop trading groups at the time and still wanted to know everything you know, and they didn't want to tell you anything they know. And so I proposed an experiment, which is give me all your trades over the last five years, and I'll tell you if you could have done better without knowing anything about your strategy. So they gave me their trades, and you know, I was working on this genetic rule learning algorithm at that time with which I'd had great success on a number of problems, you know, uh with Nielsen, with actually scheduling problems. So it was like a very robust kind of pattern discovery algorithm. Um I, you know, I took the trades, I uh amended sort of market conditions to the trades, like what was happening in the market at the time, you know, things like volatility and stochastics and all these kinds of indicators that one constructs. And I turned the crank and out came a bunch of you know rules. And I'll I'll never forget my very first meeting there with the prop trading group, where at the end of the meeting on Fridays, you know, Kevin says, So, Basant, have you found anything? And I said, Yeah, you know, I've found something, but I have no idea what it means. And he said, like what? And I said, Well, when the 30-day volatility is in the lowest quartile, your trades tend to be three times as profitable as the average. And the silence around the room, and then they start cursing each other. You know, there's expletives flying around, you know, you know, how long have I been telling you to look at volatility? And here's this guy who knows nothing, and he's telling us that it matters, you know. And, you know, I just watched this for a while and I said, Can anyone tell me what's going on? And they said, No, not really, but we know that whenever volatility spikes, we lose a lot of money. Um, and so it's interesting that you're telling us this. Now, the interesting thing is that I only found the reasons for the pattern several months later. You know, I j I dug through the Journal of Finance, you know, and all all kinds of things, and I found the reason. And the reason was like brain dead simple, you know. Which is that they were trading a lot of treasury bonds and they were mostly long collecting the carry. And whenever volatility spiked, you know, as we now know, volatility is sort of like the seasons. When it spiked, chances are that it could stay elevated. And so it was not a good time to be long. And so that's all I'd really found. It was a pattern like the nip, you know, the hip bones connected to the knee bone. But what was interesting is that I'd parameterized it by saying the lowest quartile. Because what happened is, you know, whenever volatility spiked, they put this overlay on their strategy, which actually improved it and improved it by cutting down the holding periods of the strategy from three weeks, let's say, to ten days. Then I conjectured that maybe most of the P ⁇ L was coming from the early parts of the trade as opposed to the later parts of the trade. And sure enough it was, right, to my delight. And so my sort of summary advice to that group was that you're taking risk unintelligently. You know, that your signals are actually quite good, but then they get stale and you stay to, you know, you you you sort of overstay your welcome at the party, and you're not being compensated for taking that risk with extra, you know, with a longer holding periods. So that was like a revelation, right? And and and and so and and told me something general, which was that patterns emerge before reasons for them become apparent. And I've observed this this phenomenon over and over and over again. You know, that is whenever I've sort of done any sort of sub substantive machine learning projects, you know, you see the patterns and then you conjecture reasons for them, and then the reasons do become apparent. Now, the important thing is to not fool yourself into believing the the wrong reasons, um, you know, because that's sort of a bit of an open-ended exercise. But ultimately, there are reasons for why you discover patterns.

SPEAKER_06 23:43

I thought it was fascinating because you you really identified vol clustering, so volatility. So volatility when it spikes, it uh it tends to cluster. So high volatility follows high volatility, low volatility follows low volatility. So it tends to cluster, tends to behave like it sticks together. And another thing was this, without getting too technical, you know, the unconditional and the conditional. So I I talked to students about, you know, if you're talking about the unconditional expectation, that's essentially just the history, the data and the past. You're not you're not conditioning your expectation on the variables that exist today. That's high inflation, high volatility, high interest rates. So what you were doing, which was really insightful, was you were identifying the conditional expectation and also identifying the behavior of the underlying variables, which really was insightful.

SPEAKER_07 24:34

Yeah, I mean that that's what the genetic algorithm was doing, right? It was essentially manipulating a population of SQL queries, right, for those people who understand databases and where each query was a complete trading strategy. So that that's what it was manipulating internally, and it was then handing it out to the backtester that would run the backtest on the strategy and give back the results, you know, the information ratios, the drawdowns, etc. And the genetic algorithm would then re-rank its uh you know, queries and say, okay, these look interesting, and you know, over time bubble up sort of the interesting conditional um patterns, right? And and the 30-day volatility, for example, was an interesting condition that distinguished between um you know high PL and low P ⁇ L. So that that was a uh a discriminator, and that's what it it had found. And I realized then that the machine really had become sort of a theory generator. You know, it had become sort of an automated um generator and tester of hypotheses. And that was like a, to me, like just like light bulbs went off when I realized that. And that was more than 30 years ago, you know, that you know, we we now had them this capability where hypotheses didn't need to come from the human, they could come from the machine, and it could, you know, it could find interesting ones. Um but the hard part is the sense making of the patterns, and that sense making was to me an inherently human exercise, right? That the computer couldn't actually tell you why this pattern was working. It could just tell you that it worked, you know, and and you know, you then had to find the find the reasons, like I did. You know, I mean I looked at the impacts of volatility on the yield curve and all that kind of stuff, and I realized why the pattern worked. Now, someone asked me the other day, well, you know, that was then. What about now? You know, could ChatGPT also find the reasons, underlying reasons for the pattern? And the answer is, yeah, in theory, the the machine has now gotten good enough that it could actually find the pattern and then it could go look for reasons for it as well. You know? Um it it's still a bit of a stretch, but in theory, I could see that becoming possible as well, right? Which is this sort of sense-making exercise could also be done more and more by the machine, which used to be human and quite laborious, you know, 30 years ago.

SPEAKER_06 26:51

This is it's a good case study that just for listeners and viewers, and correct me if I'm wrong here now, uh in terms of my interpretation, but what the traders were doing was going long bonds essentially. In in an environment where volatility spikes, yields will rise and the bond price will fall. Yields and pro bond prices move invertly. So if they were long, they were hoping for capital appreciation in the bonds, but when when volatility spiked, the yields went up. So that's the that's is that fair enough? That's the kind of logic around where the came from.

SPEAKER_07 27:21

Yeah, exactly. And yeah, but and by the way, I had no idea what their strategy actually was. You know, I learned later that they were following something called the Elliott wave theory, you know, but I didn't know that. And the Elliott Wave theory basically says that the market, uh at least you know, m several markets go in in cycles, you know, they they go up, down, up, down, up, you know, and the the middle up is like the longest phase, you know, the the bull phase. Now, it's just a theory. Um, but you know, people at the time, some people believed it, and that's the strategy that they had implemented. So I had no idea why they were going long or occasionally short. Uh I just knew that they were going long a lot, um, and and it made sense in an era of sort of falling interest rates, you know, because interest rates had been falling and so bond prices had been going up. So it was a no-brainer to, you know, to be going long bonds. And by the way, this is something that Michael Lewis talks about in one of his book, I think maybe Lyle's Poker, you know, that there was a you know, this was a feast for Wall Street that the Fed gave them, you know, by just sort of being in this era of sort of lowering interest rates. So that's what this group was doing, you know, basically going long bonds, and my overlay basically told them periods when it was not wise to to be long bonds because volatility went up.

SPEAKER_06 28:34

So uh if we get to the the next question then in terms of that approach that the traders were using, and if I kind of hinting at something, a c a a quote that I've used a lot in this series is Jeff Jeffrey Gunlock's quote about, you know, how everything we understand about financial markets has been informed by an economic regime that no longer exists, and that is the structural decline in interest rates and the structural decline in in inflation, let's say, on since 1980 to 2020. So all that data, the the large language models are covering all these those relationships. But what happens when you get a structural break where we're moving into a higher for longer interest rate regime and a return to normalized inflation? How accurate will the LM be then in that regard?

SPEAKER_07 29:24

You know, I I I feel like someone asked me this question at that uh the workshop in Ireland, and maybe it was you. That was me. Yeah, yeah. And and I I remember that question clearly. It was a great question. And um I don't remember exactly how I answered it, so I hope there's some kind of you know, I hope there's some consistency here. Um But you know, you're absolutely right. If you if you train a machine on that regime, then that's what it's trained for, and its predictions will work well as long as you're in that kind of regime. Now, what you're talking about is uh, you know, let's say a possible regime shift, you know, where you know uh of higher inflation and maybe even interest rates rising. Um and if you were to be building a traditional, you know, uh uh machine learning program, you would probably then sample more data from those kinds of periods, you know, like maybe 1994 to 97 or the early 2000s, you know, where we did have those kinds of regimes. That would be one way to do it. But to me, the sort of magic of you know, I shouldn't use the word magic too much, but this capability of general intelligence is that it now has the ability to reason about these different regimes as well, you know. So it's it's it's learnt enough to know that markets have regimes and that this might be a you know a changing regime. And so it you know, in theory, again, you know, practice is different, but in theory it should be able to sort of think about the new environment, you know, in a different way. Uh and s and and recognize that many m that much of its data from the past is not likely to apply to the future. So that's how I would think about it. Like that's how I would think about regime shifts in the current uh environment, right? So if I were building a bot, uh uh you know, like actually the Domoran bot, the D Moduran bot is a good example, right? That you can actually, you know, one of the interesting things about the bot is that you can give it a thesis. So you can say, okay, and and by the way, I did this. Uh I I ran it on NVIDIA pre-earnings, post-earnings, and I asked the Motheran for his input on it, which was really interesting. So um, you know, at at a at an event the other day, and I and I give live demos of it. I said, okay, evaluate Apple, and I said, and I gave it an investment thesis, which was assume that they need to come up with a new product, maybe in the augmented reality, virtual reality space, but they need some innovation and value the company with that sort of in mind. And it did. It focused on its product lines and and you know uh broke down revenue and you know applied the Motherin's discounted cash flow model and came up with the valuation. I then told it, I said, well, assume that Trump escalates tariffs, right, and that we get into an extended trade war with China. Like how would that affect Apple's valuation? And you know, to my um uh delight, uh it started focusing, it it said, okay, you know, the total volume of business that Apple brings from China is about 150 billion, you know, with 60% tariffs, you know, 90 billion of this goes up in smoke. Um, you know, and there's going to be margin compression. So instead of focusing on product lines, it then started focusing on margin compression and who's gonna be able to absorb this margin, right? Will it be the producer, will it be the consumer? You know, you know, where how how how will these margins get absorbed? So it was a completely different type of a reasoning and thinking, you know, based on the investment thesis that I'd given it. And that's what I find sort of really m exciting about modern AI is that uh it's it's not just like training a machine on past data and then hoping for the best, right? And hoping that you don't get too much of a distribution shift and that the future doesn't look like the past. You know, when you actually have a machine that can think and reason like humans can about these situations, right? How would you think about this situation? Well, you'd you'd know that this is a different regime and you should think about it differently. You know, you'd go to first principles, right? And so the machine can do that as well now. And that's what I find interesting. Uh, you know, what excites me about the moment of thought is that it can do exactly the kind of thing you're talking about, where you give it an investment thesis and it has the knowledge, it has sufficient knowledge to be able to reason about valuation of a company given that thesis and given the current economic environment, which it snarfs up by looking at the news.

SPEAKER_06 34:01

So we will have more from Professor Vassantar next week. Now let's go to those brief conversations specifically on artificial intelligence with first Jim Bianco, Professor Robert Gordon, Dr. David Kelly, we have Hayek Book Prize winner Edward Chancellor, and finally Professor Campbell Harvey. Joy. Staying relevant and adding value in this AI age.

SPEAKER_02 34:26

They're gonna probably do a better job of it than us old fossils, because they're gonna understand AI a lot better than the rest of us. And they're gonna understand the new tools and the new techniques that are gonna be used faster than their managers and the like. But let me just say one quick thing about AI. I'm friends with um the an economics professor at Northwestern University named Bob Gordon. And Bob Gordon has written a number of books about the Oh, you did? Okay, good. He's written a number of books about the impact of technology on jobs. And bottom line is it's a net creator of jobs. Technology is a net creator. So I am of the opinion that AI is going to create jobs. It is not going to eliminate jobs. Now, here's the rub. I can point to the jobs it's gonna eliminate. Uh back office jobs in the uh financial services industry, the processing of trades and loans and the like. Um, maybe maybe drivers, you know, with automation coming with uh with drivers. I can't tell you the description and the companies that are gonna be created. But AI is gonna change business models, it's gonna change cross structures, and it's gonna make things that are currently not viable as business plans viable in the future. It will create jobs. Now they won't come and go at the same speed. I think that to I have uh I have three kids in their 20s right now, and I've and two of them work in the financial services industry, and I've said to them, it's a great time to be in it. There's gonna be a lot of change, there's gonna be a lot of innovation. You're gonna be on the forefront of that change in innovation. You're gonna be able to ride that surfboard as that change comes. The older guys are gonna be afraid that they're just gonna get swamped by that wave because either they don't understand it or are unwilling to understand it, or maybe, you know, um unwilling because the, you know, that that it's alien to the way that they've been accustomed to doing things. So I think it's a great time to be in this job. Uh I'll give you one quick example. When I started Bianco Research in 1998 or so with my partners at Arbor, I looked back at what how I did my job. I was mailing my reports, I was using a word processor, I was faxing them out. I had all of my spreadsheets that I was updating by hand in Excel. I don't even remember how to do any of that stuff anymore. I can't tell you the last time I looked at a fax machine, I haven't mailed anything in a year and a half. I rode that surfboard, I rode that change into the internet and into data and into the way that we do certain things. And you'll be able to do that too. And I think that it's an exciting period that's coming right now. Um, you know, so I get it, you might want to go for a standard, traditional job, and that standard or traditional job might be getting eliminated because of technology. But don't worry, it will be replaced by some other job that can now be done because of technology. The last quick thought I will give you um Bob Gordon had as an example in one of his books, VisiCalc invented the spreadsheet in 1979. And they said, oh, that's gonna be the end of the accounting clerk job. And it was. Hundreds of thousands of people lost their jobs that was under the title of accounting clerk. But now that we had these numbers in electronic format, you could do things with them. So we invented a new job in the 80s called Financial Analyst that would project and use these numbers. And we actually created more financial analysts than we lost in accounting clerks because these data became more valuable in an electronic form instead of written on a sheet of paper. And so there was a great demand to hire people to do something, to learn something about our business or our product by looking at this data, and we were anal we were hiring financial analysts left and right throughout the 1980s using spreadsheets. I think something similar like that's gonna happen with AI.

SPEAKER_06 38:44

This notion of perpetual growth, even that notion that we're always gonna continue to grow in the context of 300,000 years where nothing happened until there was the Industrial Revolution, is that deeply flawed, this notion of perpetual growth in and itself?

SPEAKER_01 39:03

Oh, it goes beyond that. I mean, the the current Economist magazine talks about infinitely exploding growth, growth going up not to from 1 to 2% to 20 and 30 percent a year. Uh that's a total fantasy. Um and that brings up uh artificial intelligence, which we might as well talk about now. Uh the idea that computer-driven artificial intelligence is gonna take away all the human jobs and mean that human-like uh computers are gonna do everything and destroy all the jobs and give us somehow infinite output with nobody working is just a total myth. Artificial intelligence is nothing unless you're not sitting in front of a computer using it. But only about 30% of the employees, and I'll talk again about the United States, are have their jobs involved with staring at a computer. So the other 70% are not affected by artificial intelligence at all. They're engaged either with human-to-human jobs or human-to-material jobs like construction, like mining, like manufacturing. The the future uh like nurses in a hospital. I've spent some time in hospitals, and I can tell you there's almost nothing that the people in that hospital are doing that can be replaced by a robot except for transporting patients back and forth for tests. I mean, a a robot could push me down the hall on a gurney. I I accept that. But almost everything that they do when they come in to your hospital room requires a human person with enough touching and feeling uh to help you with all the different aspects of your personal life. So that so if you take all those different kinds of jobs and involving person to person or person to materials, uh and the the future doesn't depend on artificial intelligence at all for those jobs. It depends on the speed with which technology invents robots that are really human-like in their capabilities. And you know that we're way far away from that. Just walk through your daily life and look at everybody who has a job and say, how likely is it that in the next 10 or 20 years that job is going to be replaced by a robot? So I think that brings down the temperature on the the infinite growth aspect of artificial intelligence. The the real issue is what percent of tasks are gonna be eliminated by uh replaced by artificial intelligence? How much is that gonna hollow out white-collar work? How much of that's gonna aggravate what I described earlier as the education crisis of college graduates coming out and not being able to uh to find jobs? As the number of people working and being replaced by artificial intelligence uh grows, there will be a boost to productivity growth in that part of the economy that is centered around white-collar middle management uh kind of work. I imagine that to take a recent example, uh artificial intelligence could well take over pricing algorithms for airlines or airline scheduling and eliminate jobs of people who are now involved in in doing that. That's an industry I know a lot about. It's not gonna take away two pilots in the cockpit. It's not gonna take away flight attendants going up and down the aisles. Can you imagine a robot going up and down the aisles of an airplane trying to serve you? It just is isn't uh conceivable with today's technology of robots. So uh you asked about whether growth is gonna be continual or fade away. I think it's a contest. I think it's a contest between the obvious now apparent effect of AI on white-collar work, competing against all these headwinds, the growing burden of old people who are not working at all and not really affected by AI, except to the extent that a if AI can create um better diagnosis of Alzheimer's or uh invent new drugs for dementia, then that can have a big effect on elder the elderly population. But but basically we're we're talking about a growing share of elderly people who are not working, not affected by AI. And so that's the headwind aspect of growth, pushing against this growing ability of AI models to take over tasks that have traditionally been done by humans, like computer programming, software coding. Epic design, designing book covers, making movies.

SPEAKER_06 44:04

From what I'm taking from what you're saying is, and it makes sense that, you know, the manual work, uh farming, mining, you know, uh other other other work like that is pretty safe nursing. But you would be kind of concerned if you if you have a grandson or granddaughter and you were uh they were going into accountancy or law school or looking to become a maybe a wealth manager like myself. Um we hear the line uh AI won't replace humans, but humans with AI will replace humans without. You you obviously have some concerns about white-collar uh jobs, but how quickly is this gonna happen?

SPEAKER_01 44:45

Ten years ago, we did not predict that the ability of these large language models would explode so quickly as it has in the last two or three years. Uh so um my gut feeling is that we've had this explosion in the last two or three years and it's gonna level off because these large language models have been built on scraping the internet of all this information. We're running out of information to scrape. You know, so we're gonna hit diminishing returns in the ability of these models to incrementally they're gonna go from massive improvements in the ability to mimic humans. They now can write little essays on any question you ask them. But they can already write a book. Uh so w what's the next step of their ability to uh to do this kind of conversion of existing knowledge into answers to questions? It's a great convenience, all of a sudden, to be able to type into Google search some question. And instead of being sent to three or four different websites which might or might not have the answer, immediately within two seconds you get a paragraph answer through AI. That's a great service, but it brings up the question I have: what's the underlying economics of AI? Who's gonna pay for all these data centers, all this electricity, all this incredible investment that's going on, that's allowing these millions of fancy computer chips to answer all these questions? Who's actually paying the bill? You know, we're getting this stuff for free from Google. Who's gonna create the revenue to repay Microsoft and Meta and all these people for uh investing $500 billion in data centers? The story waits to be told. One of the predictions about AI is that more and more money will go to the rich and to the owners of the AI companies, and less and less to the middle class that are being uh pushed out of jobs. Uh so the only real solution to the demographic crisis is to raise retirement ages. And that's sort of easy to say for white-collar work. I mean, look, I'm 85 and I'm just uh retiring now, not at age 65. The French are fighting to keep their retirement age at 62, which is ridiculous. People are really healthy in their 60s. People are pretty healthy these days in their 70s. Um the the problem with that is that life expectancy is not uniform across types of people. People in the top 10% of the income distribution live about 10 or 15 years longer than people in the bottom 10 percent. People with manual jobs, really tough physical jobs. I think of the guy who takes care of our yard here, he's looks like he's getting on to age 70, and it's getting harder and harder for him to just do the bending and physical work. So that that retirement age issue is really a hot one because you've got a sort of obvious case for for forcing people to retire at yet older ages before they get their retirement benefits.

SPEAKER_06 48:08

And and finally, that a little bit of advice for young students, uh, for graduates that are looking to that are living in this AI world now. What kind of advice would you give them in terms of looking at a 30-40-year career? Is it to get in, is it to just if you can't beat them, join them, or is it is there something else that you would advise uh young graduates, young students?

SPEAKER_01 48:30

Uh I think we're we're entering a world in which we have a growing surplus of college graduates looking to be interns, to be middle management, to be tech workers. And we have a growing shortage of plumbers and roofers and construction workers. And I just think we have too many people going to college. That's not very helpful advice for the people who are in college and worried about getting uh jobs out of college. Uh, but I think we have an imbalance in society. It's I think it's inevitable as we go forward with the effects of AI that we're gonna have a growing share of these person-to-person and person-to-object jobs that are gonna be hard for robots to replace, and we're gonna have a shrinking share of the population doing white-collar work. And that's not what college graduates want to hear. So my message is really for people coming out of high school, do you really think it's gonna be helpful to go to college, especially if you can't get into one of the top colleges?

SPEAKER_05 49:38

Yeah, that's very honest advice and useful. Uh what AI is very good at doing, it's very good at cribbing, it's very good at copying what's out there and presenting it. You know, lots of students are using it to write their blasted essays. I think it puts the premium you know, we were talking earlier about thinking outside the herd, think trying to think differently. It puts a premium on not doing and not thinking in exactly the same way as everything else.

SPEAKER_06 50:08

I had a meeting with a client very recently, and it was very interesting. Uh, this person had a role, previous role in IBM, and I was getting some insights into the scale of the calling of back office uh employees uh automation uh because of AI. I mean, how impactful is this going to be, in your view, uh over the next couple of years? Is AI going to going to be this uh huge hockey stick uh exponential growth, or or is it is it going to meander up slowly? Uh what's your take on that?

SPEAKER_03 50:47

Well, I think the exponential character of it is probably correct, which is really interesting. I mean, I remember I remember my my most recent encounter with exponential curves was watching the spread of COVID. And once you watched it for a few weeks and did the math, you realize, oops, we we have a problem. But that's true with AI. And the reason it's true with AI is because AI is getting smarter. Every new addition is smarter. Every new addition means that there are more workers that this thing could replace. Every time that happens, it encourages more capital spending. It also encourages more innovations in complementary technology such as robotics that could use AI. And so what I think is going to happen is, you know, AI is going to impact the environment over the next five years, but it's going to impact the environment far more over the following five years, and far more than that over the five years after that. There are negatives, potential negatives, because of what it does, its potential for misuse and the implications of democracy and so forth, and I suppose, you know, the misuse of it and things like computer viruses. But overall, I expect that a combination of AI and robotics will each year make more and more workers obsolete. And that's quite dangerous if you have a recession. Because if you have a re- what happens is if you have an expansion, companies say, Oh, you know, maybe we'll keep Joe working. I mean, he's a nice guy. We you know, we don't want to cause any misery here, and we can we can get by okay. And then when the recession comes, look, we're gonna have to make tough decisions and and uh out Joe goes, and and and all the people that could be replaced by AI suddenly find themselves looking for a job. Um, so I think it could it could lead to a significant reduction in workforce come the next recession, as people say, well, you know, RSA could do that job better with with uh with a machine. So I think over time it's gonna make more and more workers obsolete. Economists always say, and I I still kind of believe that technology tends to create new jobs uh as the technology comes in. But this this is going to be an interesting challenge because its impact is not going to be meandering. I think it's gonna be exponential. Every year, the impact it has on um labor demand is gonna be greater. And so there'll be a greater need to find other things for people to do usefully, but AI can't do faster and cheaper than them. Uh it will uh it'll be interesting to live through this.

SPEAKER_06 53:08

In respect of your most important thing then, Professor Harvey, I know it's very difficult for you, but looking out into the next decade, if you had to narrow something pinpoint to town, what what would it be? It's obvious.

SPEAKER_04 53:21

Uh AI. Which is which is greatly underestimated uh today. So um it it it is a bonanza for management, uh for investors. It will it will make uh things so much cheaper. It will have a very negative effect on global trade because there's just no reason to outsource production to a low-wage country when you've got a robot for one dollar an hour that'll work 24-7. So production will be onshored. This is uh you know amazing for for even what I do. You mentioned my YouTube videos. I've already converted them to 4K, and now I'm in the process of quote unquote dubbing them into a hundred different languages. And and dub, what does that mean? It means that it is me. It looks it is me, but my mouth is speaking Arabic to any lip reader with my voice. So so this is uh this is something that will be very important in making markets much more efficient. Um indeed, it is an unusual time where we've got we've got uh four major disruptions happening simultaneously. So think of AI. The one that doesn't get as much attention is quantum computing, but it's big. Uh one that's poorly understood is decentralized technologies, is poorly understood because people just uh focus on what Trump is doing and the price of Bitcoin. There's so much more uh beyond that. And then the number four is biomedical research, where combining with some of these other technologies, we will accelerate the rate of breakthrough uh in a very, very substantial way. So this is a highly unusual time. If you look historically, it's usually like one thing. So you look at what happened with the introduction of the internet, that was like one thing. Um or the introduction of computing, one thing. This is four going on at the same time. So there are just enormous opportunities for investors getting in early. And one thing I've advocated strongly for is for the U.S. to change the laws on kind of qualified investors. Most investors, most retail investors don't have access to non-public equity, and you need to be rich. That's fundamentally unfair. And it kind of gets back to what I was uh what you mentioned in terms of financial democracy. Decentralized technologies have the possibility of tokenizing private equity. And again, most people don't realize that the value of private equity today, and I'm talking about non-public equity, so equity that has never been public, as well as equity that was public and taken private. The value of that is about the same value as the public equity. So investors need access to that. So investors, especially young investors, should have a diversified portfolio that includes plenty of allocation to startups in this space. And again, it is important to have a diversified portfolio. Now we can do fractional shares and things like that because as we know, uh these investments, most of them will go to a value of zero. But again, consistent with what I've been saying earlier, you need to uh not just think about maximizing the expected return your portfolio for a level of volatility, but also the skew. And especially for young people, you want that big skew. And uh that's why it's important to have investments in these emerging uh technologies.