Personal Wealth Management / The Well-Read Investor
Episode 3: Joshua Gans
Award-winning economist, author, and professor, Joshua Gans and Investor Mike Hanson examine the economics of Artificial Intelligence.
In Episode 3 we continue our mini-theme on Artificial Intelligence with award-winning economist and professor Joshua Gans, speaking about his book Prediction Machines: the simple economics of Artificial Intelligence, co-written with professors, Ajay Agrawal and Avi Goldfarb. This is an expansive career that features several books, numerous publications and awards. Aside from all the books and accolades, we think you'll find professor Gans, not only a great mind, but a gregarious one as well. This interview was conducted just before the COVID-19 pandemic gripped the world. On that note, you may want to check out professor Gans newest, short, but thoughtful and very timely book, Economics In the age of COVID-19 published by the MIT press, available on digital and audio. For more information on Josh or his books, visit https://www.joshuagans.com. He’s also on Twitter @joshgans.
Full Episode Transcript
Michael Hanson (00:09):
Hi everyone! Today is July 15th, 2020. And welcome to another episode of The Well-Read Investor, the podcast that profits your mind and your money. Today, we pick back up with our mini-theme on artificial intelligence. Last episode, we had a wonderful and far reaching talk with professor Melanie Mitchell on the nuts and bolts of AI. And today we're delighted to have professor Joshua Gans on to speak about the economics of AI with his book Prediction Machines, the Simple Economics of Artificial Intelligence. Co-written with professors, Ajay Agrawal and Avi Goldfarb. Among much else, Joshua is a professor of strategic management and holder of the Jeffery S. Skoll chair of technical innovation and entrepreneurship at the Rotman school of management at the University of Toronto. He's also the chief economist at Toronto's creative destruction lab, where he teaches entrepreneurial strategy. This is an expansive career that features several books, numerous publications, and awards.
Michael Hanson (01:06):
And in fact, back in 2007, Joshua started his career at The University of Melbourne, Australia, close to where he grew up in Sydney. And I think you're going to pick up on that accent throughout the interview. Aside from all the books and accolades, I think you'll find professor Gans, not only a great mind, but a gregarious one as well. Something we could use a lot more of these days. This interview was conducted by phone, as Professor Gans was in Toronto at the time of the taping. And in fact, the first set of our podcast interviews were all conducted just before the COVID-19 pandemic gripped the world. And on that note, you may want to check out professor Gans’ newest, short, but thoughtful and very timely book, Economics In the age of COVID-19, published by the MIT Press, it's available on digital and audio. And I not only enjoyed it, but I couldn't help but be impressed by his ability to synthesize so much information so quickly and get it out to the public. So, without further ado, our conversation with professor Joshua Gans, enjoy.
Joshua, welcome to the podcast. You know, one question we always ask our guests is, why should investors be familiar with your work?
Joshua Gans (02:18):
In particular, we wanted to be able to think about and do what we did best as economists and business school professors. And think about how to explain and structure decisions with regard to, you know, is artificial intelligence relevant for your business. We spent a lot of time talking to economic historians about technological change, and whether something's a big radical technology or not, which helped to crystallize our thinking in terms of, what do you look for with something called artificial intelligence that can get some economic meaning. How do you translate what is quite technical into things that are relatable from an economic sense? And so that's why we eventually wrote the book.
Michael Hanson (03:00):
So tell us a little bit about yourself, you know, how did all of this begin?
Joshua Gans (03:03):
Well, we became interested in artificial intelligence around 2014 when we saw through our program, the creative destruction lab here at The University of Toronto, many startups coming, applying this new form of technology, machine learning, which is a catch-all for the latest advances in artificial intelligence. And, you know, when you see that happening, literally on your doorstep, you get interested as to whether you're seeing something significant or not, so we became convinced that this was going to be a big thing.
Michael Hanson (03:38):
How does the collaborative process work when you're working with other prominent economists, other economists who do quite a lot of writing, how do you come together that way? How do you have a common idea and really how do you work together?
Joshua Gans (03:50):
Well, I think we're all colleagues. Uh, we, you know, in the same building, we were all involved in the program for the startups firms. So these things come naturally through discussions. I think at various times we saw different angles on the phenomenon. And so it was through that interaction that we ended up writing the book together. And it's an interesting process. We each have our strengths and weaknesses and I guess we ended up complimenting each other.
Michael Hanson (04:19):
Yeah, I would say so. And you know, one of the real, I think just stark and interesting points of view of the book is that, when you look at AI machine learning as an economic issue, you sort of describe predictions ultimately as a kind of commodity or at least some point they'll be in the future kind of a commodity. And that really one of the things I learned from it was that, we just do a lot more predicting day to day than we realize we do, that life is just full of those predictions.
Joshua Gans (04:44):
Yes. You know, just to step back a second, artificial intelligence itself has a whole lot of loaded meanings, which is, part of the confusion, you know, it gets built in popular culture, Terminator movies - what have you. And so that gives us sort of distorted view on what we really have right now. I see our role as economists is to somewhat de-sexify these sorts of things, and what that meant in this case was, you know, what is this machine learning really doing? It hasn’t created a robot or created a new intelligent life form or anything like that. What it does is one thing, it improves our ability, as essentially statisticians, to predict stuff. So that's all it did. And so, when you realized that, you start saying, well, where do we use prediction? And in some places it's fairly obvious. You can predict the weather, whether you take an umbrella to whether you should launch an invasion of Normandy, those are reasons to have those predictions.
Joshua Gans (05:45):
Michael Hanson (06:27):
So, I've seen you in several venues and one of them, you were speaking about Nick Bostrom's book Superintelligence and you sort of described yourself as kind of in the middle on AI, particularly as you've described it in terms of its dangers and its promise. Where are you on that today?
Joshua Gans (06:42):
I think I'm still about the same. I would be, relative to a few years ago where people were a little bit more excited about how things were improving. I would be a little more pessimistic, you know, in my forecast of when we're going to get the real intelligence, not just prediction, but something artificially intelligent. At the same time, I think everything that I've seen in, you know, I guess about five years of that time has been reinforcing the idea that artificial intelligence works best when it's matched with humans. So that's kind of where these things are going. And just to give you an example of that, is back in 2015, people were forecasting by 2019, we would have self-driving cars on the road, absent any legal issues about whether that was going to happen or not - we haven't seen that. And it's a difficult task because essentially 99% of the time, you can't have a self-driving car. Well, the 1% or whatever it might be, you need somebody there when those things occur, if you don't have a human doing something, the consequences are so costly that you might as well have a human driving around. But things have slowed up in terms of some of the projections that were made.
Michael Hanson (07:52):
That is quite interesting. And you know, in our field of endeavor, which is investments, such things have been employed, attempted to be employed. But my personal view is your insight is absolutely correct, that you still need the overlay of human intelligence or some kind of decision making facility there, at a minimum to be in concert with some of the models that get created. And yet, a lot of folks in this industry will claim that you can just set a model and make it a prediction machine. What do you think are the limitations of these things? I mean, can these things ultimately really predict majorly, let's call them chaotic systems?
Joshua Gans (08:27):
Essentially our ability to predict with anything, let alone AI, is grounded in our past experience. And even if that experience is extremely rich, there are situations, and there are places where new things occur, but when new things occur out predictive models, AI, or otherwise break down. And we have to re-learn them. Which domains are conquerable versus not, and how far can you go? That is still very much an open question. And I think we're still learning on it. I mean, I think we're relatively more optimistic. But being able to forecast financial markets, I'm pretty sure that's not on the table. Because of the excellent reason, that as soon as we develop something that is forecastable, people change their behavior and the model goes out the window. That’s the first thing. But the second thing in terms of the adoption and performance of AI, is in knowing what to do with it.
Joshua Gans (09:20):
So, one of the things we learned from the history of major technological change, electricity, computers, et cetera, is that it takes time, you know, the productivity didn't come from electricity until about 40 years after it was first deployed for use to power machines and things like that, because factories had to be redesigned completely around the fact that they now had cheap power. And once that redesign happened, then things could get exploited. And so I think we're going to see the same thing happen. And one of the goals that we had writing the book was to get people thinking about the fact that, it’s not just some magical thing, you can't just buy some AI and you're going to have dividends. You have to work out how it's going to fit in.
Michael Hanson (10:06):
So, you mentioned this notion of productivity, and I know that quite a lot of your writing and quite a lot of the book actually does some discussion about how to companies deal with this. And there's a lot of fear and there's a lot of hope for it. And one of the things you mentioned was that in fact, you know, humans have to interact and I've often been fascinated by the game of chess because Gary Kasparov in particular, his view is very close to the one you articulated, which is that even in a system like, which is a fairly closed system, like a chess board, the interaction between the human and what version of AI that they use for chess is in fact often the most potent thing. It's not just the machine itself, but there's been so many famous predictions and you're talking about the delays in technology and how long it takes to implement something about productivity and when the end of work will come and so forth, you seem more optimistic about that. What are you saying though, to companies right now, as they ask you about what they ought to do with these things?
Joshua Gans (11:00):
If I have a broad message. It's that if you were sitting there saying, I could adopt AI and on the benefits side of the ledger, it's something calculated like, here's the amount of labor costs I can save by doing this - some projected amount. My advice currently is to throw away that kind of thinking. Invariably, your ability, just like with a self-driving car, the ability to kick out a human, you got to get things right on a lot of dimensions to be able to do that. Because more often than not, when we try to remove a human, you realize that they were doing more than we thought they were. So, you know, people might talk about self-driving cars, you can get rid of a human who is driving a school bus. And that may be true, but that's a very different question because now you say to yourself, ah, so you would be happy having your kids get together, get picked up by a self-driving bus with no adults on board.
Joshua Gans (11:48):
Parents are just saying, hold on a second, right? Ah like, no, no, no. That's different, because the school bus driver is not only driving the bus. They're protecting the kids from various things, including each other. In other words, you should be thinking about, well, now that that's the case, what can I do with that person? They don't have to worry about driving. Maybe they can start the class lessons on the bus. And I think that also applies to things like truck drivers - truck drivers, this big category, 3 million in the US, one of the biggest employment category’s in most economies. So, people say, Ooh, we can use self-driving things, we're not going to have to have any of these truck drivers. And then you start to think of that, do you really want to send a load all the way across Canada, East to West, with no one there? What if something happened? What if somebody tries to high-jack, you know, steal whatever's in the load? And you start to realize that the driver is doing more than just driving, but the idea that you're going to fire that person, not as obvious now. So that's my advice for businesses, is don't think about it that way. Don't think about human replacements. It's a bit of a pipe dream. Think about human augmentation first.
Michael Hanson (13:00):
I'd like to go a little more broad, and talk a little bit about your career and some of your other influences. I know you work at the center for creative destruction. How important to you are some of the economic influences as Joseph Schumpeter? How do you see things like creative destruction today? Both good and bad.
Joshua Gans (13:16):
So, the term creative destruction was coined by an economist, Joseph Schumpeter, who noticed in the early 19 hundreds, what was driving the economy was innovation and new things, new ideas. First of all, innovations are inherently creative. You get a new, better mouse trap. At the same time, when you get a new mousetrap, what happens to the people who were making the old mousetrap? Any innovation that comes into play, is usually destroying something. But it's also interesting, sometimes you see an act of creation that ends up destroying something that you wouldn't have forecast. So for instance, we got the iPhone launched in 2007, and I don't think anyone who was looking at the iPhone said, well, that's it for taxi. Taxis are gone because of this, but in fact, that is what it's happened. You see where the destruction comes in. So sometimes it's a long way removed.
Joshua Gans (14:05):
I think that's actually the notion that we don't know for any given innovation where the best creation is going to be let alone where the destruction is going to be is part of that. As economists, we like to remind people of that fact, and we also like to in our creative destruction lab, emphasize that when someone comes in with a scientific breakthrough, the path to market means adjudicating now, a whole set of market and other experiments. And it's going to take some time to work out how that occurs. Then moreover, the scientist are going to need some help along the way. We can’t kind of expect the scientist coming out with a new breakthrough in AI to also, to really understand the nuances of business development around that. So, you have to put all those things together, and I feel our job is to remind people of what else is required in getting innovations to work.
Michael Hanson (15:00):
And so, you've alluded to this a few minutes prior, but you think that artificial intelligence or perhaps we'll call it machine learning for this, is fairly early in its stages of that process?
Joshua Gans (15:11):
Yes, yes, absolutely. You know, if you imagine that fusion of a technology is like a long an elongated S-curve, you know, at the early stage of the technology and sort of like, gets a little bit of traction and then at some point it takes off everywhere, and then eventually hit some sort of upper limits. And we're at the upper limit of adoption of the iPhone. And that took a process of a decade or so. But with AI, it's not just that we're trying to push people using this stuff. We're still trying to get it technically better and still was still in that middle, bottom part. We could be getting a lot of improvement, but I still think we're still a ways off knowing what to do with it, knowing what the potentials are, et cetera, et cetera.
Michael Hanson (15:55):
Now, many people think that rate of change is accelerating, but you know, you've sort of alluded to there's friction here. There are frictions, and do you see an accelerating rate of change?
Joshua Gans (16:06):
It’s simultaneously possible to believe something is accelerating and also to believe there are frictions that are holding back from accelerated even faster. So, I think that's basically where we are. So, I would say yes, if it's accelerating, very exciting developments have occurred, there are fields that have seen big jumps in, sort of, what they thought, could be done by computers. But what we don't know yet, is where the biggest bang for the buck is going to be. We may very easily use AI to work out what's in pictures, and that's useful if you're interested in searching pictures for particular things. That's great. But, what we don’t know, is now that I can do that, now what? Is there something surprising that might come of that? And we’ve seen surprise happen before. What computers do is one thing, they do arithmetic, and they do it very well. And so initially when we got computers, people used them to calculate things that they were already calculating, such as artillery tables, statistic things, etc. Then eventually the cost of doing those calculations became so low that people realized, hey, I can use this stuff to encode and distribute music. That's just arithmetic at some level, essentially, that's what's going on. And so you get these entirely new things. And I think what we haven't seen yet with artificial intelligence, are those new things. And that's when you'll know that things are starting to get real.
Michael Hanson (17:28):
One thing I've noticed about your body of work across many books now, is that you just have a real facility for communicating these difficult economic concepts to a broader public. How in your career have you approached that and thought through doing that?
Joshua Gans (17:43):
Thank you for saying that. I think it's something that developed through practice. I guess I was fortunate to have received my academic tenure and be sort of a little bit more relaxed at a time when being able to write a blog started to become a thing. I guess I just started making that part of my day to day life. And so, writing and doing this sort of thing, is a bit like a muscle, if you practice long enough, you start to get better at it. And so, I found that blogging was that for me, I did blogging for a number of years before I started writing books on the subject. I think now actually blogs have sort of disappeared by the wayside, but it was really that early practice of day to day writing two or three blog posts about random topics, trying to explain it to myself and the world.
Michael Hanson (18:30):
It's funny. Yeah. Trying to explain something to yourself, it's usually one of the best ways to do it, but what have been some of your key influences and what are some of your favorite books on economics?
Joshua Gans (18:40):
Oh. So, I've got different ones depending on mood and other things like that. In terms of inspiration for people who write about translating economics for the real world, I'm a big fan of the work of Tim Halford, who started with a book called The Undercover Economist. And he has a particular style about him that I really appreciate, the ability to sort of translate that information. In terms of economics books that I've read that have influenced me quite a bit, there’s one that’s by Ken Arrow, who was a Nobel Prize winner in economics and also my thesis advisor called The Limits of Organization, which was a series of lectures that showed why it was that organizations had trouble staying at the frontier of efficiency all the time. So, you have this sort of choice between I can be in the frontier now, or I can be a bit away from the frontier, but be more resilient. What I liked about it was just how clearly it was written and laid out.
Michael Hanson (19:33):
Ken Arrow, to my understanding was, was a real pioneer in what you might call some of the complexity sciences as well, fusing some of economics and complexity sciences. And in fact, we had just recently had a discussion with Melanie Mitchell at the Santa Fe Institute who also spoke about Arrow and his influence as well. That's really very interesting. To finish things out here. What is it that you wish everyone knew or understood better about prediction machines, artificial intelligence in general? What's the one parting shot you would say I wish everyone understood?
Joshua Gans (20:03):
I think at the moment I would like people to understand both how boring it is, and how great statistics is. Simultaneously, it is a mere advancement in statistic but that’s great. That’s what I want people to understand.
Michael Hanson (20:22):
Well, my guest today has been professor Joshua Gans. Thank you again so much. Really, that was a pleasure to do.
Joshua Gans (20:28):
Thanks. It’s been fun.
Michael Hanson (20:42):
Well, there, you have it. Our conversation with professor Joshua Gans. For me, the lesson, and this is so often the case in investing, is that unintended consequences are the norm, especially when you're talking about things like new technology. I personally loved professor Gans. His idea that, you know, when the iPhone arrived, nobody instantly said, well, that's it for taxis. I think that's a brilliant way to consider that. And it's also very consistent with Melanie Mitchell's view when we had her on the program that yes, AI is powerful. It can be powerful, but in very specific domains and it's no replacement for human judgment and context, not by a long shot. At any rate, thanks again to Josh. And thank you again for listening! Wherever you may be hearing The Well-read Investor, please comment, like and subscribe, or visit our website: www.wellreadinvestor.com for more information, and look for additional book reviews, not featured on the podcast, coming soon. Meet us back here in two weeks on July 29th, when we speak with Dr. Brian Arthur about his book, Complexity and the Economy. We will embark on a very broad discussion about economics, complexity, the nature of technology of which Dr. Arthur is an expert, and we’ll even touch on the idea of free will and human economic action. It's going to get weird. So, until then, may all of your reading profit, your mind and your money and take care.
Investing in securities involves the risk of loss. Past performance is no guarantee of future returns. The foregoing is for general informational purposes and should not be regarded as personal investment advice, nothing herein is intended to be a recommendation or a forecast of market conditions. Rather it is intended to illustrate a point. Current and future market conditions may differ significantly from those illustrated here. Not all past forecast were, nor future forecast may be as accurate as those predicted herein.
See Our Investment Guides
The world of investing can seem like a giant maze. Fisher Investments has developed several informational and educational guides tackling a variety of investing topics.