What’s up with artificial intelligence? How is AI helping to drive business innovation and market competition, and how much remains science fiction? We speak with Professor Joshua Gans to demystify AI and better understand the nature and disruptive power of AI on markets and society.
In our fourth episode of the podcast, Professor Joshua Gans of the University of Toronto’s Rotman School of Management drops in to discuss the economics of artificial intelligence with Counterfactual host Charles Tingley. Professor Gans shares his expert insights to clarify the nature, power and limitations of AI when it comes to business applications and the potential to influence competitive conduct in the marketplace. In addition to canvassing the potential competition law implications of AI, discussion extends to broader societal implications of machine learning and even to the must-have gadget recommended by Professor Gans in case listeners are looking for that special holiday gift.
00:00
Welcome to Counterfactual, the podcast brought to you by the Competition Law and Foreign Investment Review Section of the Canadian Bar Association. Counterfactual takes a fresh look at issues relevant to business, competition, and related areas of regulation, and explores the real and hypothetical worlds to gain practical insights and debate policy. Hope you enjoy the show.
00:28
Hello, and welcome to Counterfactual, the podcast produced by the Competition Law and Foreign Investment Review Section of the Canadian Bar Association. My name is Charles Tingley, and, in this episode, I'll be speaking with Professor Joshua Gans of the University of Toronto's Rotman School of Management. We'll be breaking down the often discussed, but perhaps less often understood, concept of artificial intelligence, its current and future capabilities, as well as its potential to shape business, competition, markets, and competition law enforcement. Before we get started, a few words about our guest. Professor Gans is Professor of Strategic Management and the holder of the Jeffrey S. Skoll Chair of Technical Innovation and Entrepreneurship at the Rotman School of Management. Joshua is also Chief Economist of the University of Toronto's Creative Destruction Lab. In addition to his academic positions, Professor Gans is a Senior Consultant with Charles River Associates. He holds a PhD from Stanford University, and an honours degree in economics from the University of Queensland. Professor Gans’ research interests are varied, but he specializes in innovation, competition, intellectual property, licensing, utility regulation, vertical mergers, digital economics, and the economics of artificial intelligence and cryptocurrency. Among his numerous books and publications are recent titles, including the Economics of Artificial Intelligence, Innovation + Equality, Prediction Machines, and Power and Prediction: The Disruptive Economics of Artificial Intelligence. Professor Gans, hello, and welcome to Counterfactual. We're so glad you could join us. So, just maybe to start, and maybe this is a reflection of my own limitations, but I sometimes wonder, when we talk about artificial intelligence, whether we even know what we're talking about. So, I'll just start by asking, is there a commonly understood or agreed [upon] definition of artificial intelligence?
02:31
You know, it's an evolving definition. Depends on the context. The context that I'm most familiar with is the sort of decades long explosion in advances in what they call artificial intelligence. What that is, is really advances in a field of computer science called machine learning, which itself is fundamentally an advance in statistical techniques. In particular, the statistics of prediction. All artificial intelligence advances that you've been hearing about recently, every single one of them, is really an advance in our ability to predict values, take information that we do have, and turn it into information that we want.
03:24
So, you've talked about, sort of, a bit of a historical perspective – is artificial intelligence even new?
03:32
You know, one of the things that economists noticed when we started seeing these developments in artificial intelligence, such as the ones that played Go, the game Go, or the ones that were playing computer games well, and things like that, was actually they had all been developed previously by some econometrician. The difference really was on the volume of the data and the intensity of the computations going on. In other words, taking advantage of the latest developments in information technologies that made these statistical engines far more powerful.
04:12
I guess, just to round out this notion of what is artificial intelligence, do you think there's any common misconceptions in popular culture when we use that term?
04:22
Oh, popular culture is just one big fat misconception. That's really what it is. There are very few instances in popular culture [where] using artificial intelligence is close to what we're actually doing with it. All of the other ones in popular culture imagine a machine that can think. That's really what it is. And some are very good at thinking and acting. Some are a little more clumsy. And it's basically a different form of life that we're sort of dealing with – a different species, if you will. That's the popular culture notion. And it really, you know…there's nothing in what we're doing currently that is reflective of that. Everything else we've got is really the more boring end of statistics.
05:11
All right, so then let's maybe talk about the practicality. I mean, how have businesses deployed AI in the marketplace? I mean, I wonder if you're able to give us a few examples and maybe sort of categorize or organize, you know, the most common business uses of AI?
05:30
Yes, certainly. As I said before, it's an advance in prediction. So, it wouldn't surprise you that most of the initial uses of AI were actually to use these new techniques in predicting things that we’re already predicting. So, if you're a financial institution, and you have to have some way of predicting whether there's credit card fraud or whether there's money laundering, or something like that, [you] would now use what we call artificial intelligence to assist in doing that. In fact, in Canada, there was a firm called Verafin from St. John's in Newfoundland that was acquired a few years ago by Nasdaq for several billion dollars. And that firm had previously been involved in the any ways [sic] of automating fraud detection, and things like that, and now had been recently, of course, using artificial intelligence. And that's why they were attracted to that market. That firm remains in St. John's. It can do it all from there. So, those are, you know, the current applications that, you know, a lot of businesses are taking opportunities in. There's a whole lot of artificial intelligence being used by, for [lack of] a better term, Big Tech. You know, we have it in our phones. It's artificial intelligence that can recognize our fingerprint or our face in order to unlock our phones. It can recognize, nowadays, which apps I use at different times of the day to have them surface and show me their existence, and be ready for me, say, in the morning. Again, that's artificial intelligence at work. And even selecting our musical playlists, and things like that. So, we see that sort of consumer process now being powered by artificial intelligence. It's not that they didn't try to do these sorts of predictions before. It's now that they just use these new techniques. What we haven't seen yet is artificial intelligence coming in and doing something so new that it transforms an industry. The closest one where there have been attempts to do this is, of course, in self-driving vehicles. Self-driving vehicles are designed to transform how we get around and how much effort we have to put in, in doing that. But I think, despite high degrees of optimism that we would have full self-driving capabilities a few years ago, that seems to have staggered off. I think, in reality, there's a lot of things at play. There's the quality of the machines themselves and their ability to navigate, that's probably quite high, higher than people think. But, at the same time, you know, regulation is not going to move very quickly on allowing a driverless vehicle on our roads. And what the criteria should be is still up for grabs. So, I think that is slowing down all of this. Also, what is people's desperation for that –
to be relieved of the driving task? You know, if I have to sit on a commute anyway, the driving isn't the biggest worry I have. It's the traffic. Self-driving cars don't solve the traffic. You know, what I would like to use self-driving cars for is where I can't currently drive. For instance, I'd like to have the car drop me off at work and then go find somewhere to park or do something else for the day and come back and pick me up. You know, until we are able to have a car without anyone in it, we're not going to be able to get that situation.
09:40
So, even where those sort of regulatory or, you know, more conceptual issues don't apply and you're just doing what you used to do, but perhaps presumably much faster and much better in terms of prediction, where do you see the power realistically of AI to disrupt markets?
10:00
Well, I think, you know, there are some jobs that will start to get transformed by AI. I think if you're a graphic designer or anything to do with, you know, putting in…developing pictures for the purposes of advertising and stuff like that, I think, we're not that far away from saying, “Well, I don't have to hire a model or camera [operator]…studio or anything else, I can just knock up whatever fashion photos I want right here on my computer, and no one's going to really care”. So, I think we're going to see this is going to start to change things in that regard. But really, you know, despite a lot of interest in artificial intelligence – some you know, 50 or 60% of businesses exploring whether AI could be used – we've only seen about 10% employed in any capacity at all. And the reason is because we have built up a lot of things to deal with the fact that we can't predict stuff. You know, it's just part of basic risk management. So, if you're a retail store and you have trouble predicting demand, what do you do? Well, you don't like to run out of stuff. So, you hold inventories for that purpose. So, if I wanted an AI that was going to predict demand, I would have to be assured that it could, I could, supply that demand. Otherwise, who cares about knowing whether I can actually…which products people want, unless I can supply it. And, at some level, that production has to take place before I really know whether consumers are going to want something, in order to be meaningful. And that's just a bigger problem than just prediction of demand. And, so, I think, where we will, in the future, point to AI as transforming things is still very much an open book because it's really going to be the case that the AI has to itself be very, very good And so good as to cause people to say, “I want to build something completely new around this – this will overturn this industry”. I mean, just to take an example, not from AI, but from our recent past. You know, no one when they saw Steve Jobs present and unveil the iPhone back in 2007 thought to themselves, “Well, that's curtains for the taxi industry”. Right? It wasn't like the first thought you – one – might have had. But, you know, in reality, in a big way, it was. It allowed for applications that will allow you to broadcast your location and hail a ride. It also allowed the drivers themselves to get the skills of a taxi driver, virtually overnight, through navigation apps, which is really what the skills of a taxi driver were – how quickly to get from A to B. I mean, if you think to yourself, would you have taken an Uber or Lyft if it had been somebody without any particular notes, without any experience, or with[out] your current knowledge of how to get to places, and not be frustrated quickly and run back to taxis? So, in that regard, a technology that was ostensibly called a phone – and we still call it a phone – was able to disrupt a major and global set of markets in regard to personal transportation.
13:47
Well, that's really interesting. And, of course, even with the disruptive influence of, I guess, those geolocation apps that allow people who aren't from a particular place to actually know how to get around, I always find that when I get in one of those and the old app comes up, I sort of go, “Oh, no”. But, nevertheless, it's obviously much better than not knowing at all how to get around. Certainly in our field – in the competition field – there's a lot of discussion about, you know, AI, and I guess what fuels it for machine learning. And perhaps that's data. But, you know, what is the relationship between AI and data in your view?
14:28
Well, I think, I mean, there's obviously an extremely important one. There’s no AI that can run without data of some degree. Where that data comes from is the critical question. And I think largely, during… when this was just getting known about 2017-2018, businesses were sort of looking and saying, “Hey, artificial intelligence is coming. What sort of advantages do we have? What's going to give us a leg up in this market or give us some assurance that we'll be able to play?” And they, you know, immediately cracked open the filing cabinets and the storage disk drives and said, “Look at all this data we've been collecting for years that we haven't been using. AI needs data; we have data. Great!” And, so, I think there was a lot of optimism regarding existing data sets and how they might be able to be used by artificial intelligence…these new techniques…to be valuable. Are we going to be able to make it valuable? And people talked about data being the new oil. The problem with that is that data has to be fit for purpose. You know, some of the earliest examples of this were in images collected, used…collected in the past and then used to diagnose whether somebody has a malignant tumor or a problem, or some other problem. And they fed those pictures and [sic] appropriately labeled into the AI. And they turned out it…”Oh, wow, it has…it's the equal of humans, right from the start”. Well, it turned out that actually, the images, you know, had a bit of extra information in them. When a person, a doctor, thought that the X-ray that they we're looking at, or taking a picture of, was an issue, they would swap out a ruler, and try and, you know, measure what the size of it is. And some of these, many of these images were captured on that ruler. And, so, what the AI was basically realizing is that if there's ruler in the picture, it's probably a serious problem from when there isn't. It might not have been quite a ruler, but it was something along those lines. And, the point being is that, you know, people were already sorting those pictures…people were already distorting them. So, to build an [sic] artificial intelligence, you had to have very clean data…it had to be clean. And, so, how you do that is a whole other matter. And, moreover, and this wasn't a surprise to we economists, you know, to the extent that it was just being able to apply statistics, we knew there were a whole lot of challenges with that, you know. We know that it takes a PhD economist, sometimes with several years [of] experience, to be able to derive the right inferences from a given data set to answer certain questions. There was no reason to suppose some engineer with off the shelf items, just running data through it, was going to count…were going to come up with actionable items without their own biases being introduced, without there being, you know, mistakes being made, incorrect inferences, and all those sorts of things. So, what that basically meant is that the data in people's coffers was worth a lot less than they thought. And from a competition perspective, that particular fact means that, you know, if you're thinking about mergers and other things on that line, or a source of market power, this didn't seem to be it. However, once you have…understand AI and its data requirements, the ability to see into an environment and collect data, and collect the data in a way that's fit for purpose, that ability becomes the source of potential market power. And that ability is often correlated with things we're very familiar with, such as market share, or total employment, or total capital investment, and all those sorts of things. So, in that regard, AI represents, you know, no different situation than we would have [to] assess normal competition policy with, except that now, you know, here's another reason why it matters to have a higher market share –
where you can train your AI eyes on a larger data set. And you'll get an advantage over time potentially because of that. So, it's not the stock of data you have, but the ability to generate new data and continue to do it into the future. And sort of owning the customer relationship, as they call it, is one way to do that.
19:43
Now, just quick, and this may be a quick question, but, you know, I hadn't really thought about it, but it occurred to me that we talked about, you know, who has the data, who generates the data, but presumably they need people to run the AI and to, you know, harness the technology. And, so, we don't sometimes think about, you know, who are the actual AI companies, if such a thing exists? And is there any ecosystem there that's of interest? Or is it a highly fragmented area where people focus on particular things they can solve? Or how does it work?
20:22
Well, we don't quite know yet. This is still being played out. But, you know, there is talent at the heart of artificial intelligence. But the talent is very diverse, depending on what is being done. So, you know, we've been fortunate in that the Big Tech companies have largely been in the business of trying to encourage people to use artificial intelligence. And, so, they've made the software and even the compute…you know, the cloud computing facilities and other things like that relatively cheaply available to people to use to do sort of off-the-shelf AI. Now, I've already said previously, off-the-shelf AI sometimes can work wonders. You know, if you are doing a school project, or something like that, you'll be able to use it quite well. In other situations, it's not going to be enough, because you're going to have to have the talent, the statistical skills to really make use of that artificial intelligence. On the other side, there are now these very large models coming out there…sort of large language learning models and things like that you might have heard [of]. One’s called GPT-3. Meta released one of these just recently. Those same models are also being able to produce art, or things that people argue about whether they're art or not, which I guess is the same thing. And, so, those things require a huge amount of computational energy to use. So, even the firms who are trying to encourage people to use it have had to have people sign away for some of the usage of those things. It's still not terribly expensive, but it's there. And, you know, we don't know whether those models and those very capital intensive ways of producing artificial intelligence are going to be the key dominant ones. And, if so, you know…yes, we'll…like every other time with Big Tech, there will be a big AI tech. Will they be the current players? Will there be someone new? Still hard to tell. But, you can sort of see those processes coming. But, you know, it is amazing how early the days are in that regard. You know, there's been very little to excite a competition economist on what current…people are currently doing with artificial intelligence. So, we're still in the realm of speculation.
23:02
Yeah, I mean, speaking of early days, you know, back in 2018, the Competition Bureau released sort of a discussion paper on big data and innovation. And, at that time, they said that they had not seen evidence that cartel agreements are reached purely through interactions between different AI technologies, absent any human, you know, direct involvement. But, they did recognize that technology, of course, and business practices continue to evolve. So, it's been about five or so years since then. And, I guess, I'll throw it out there, you know, can AI collude on its own…can it facilitate collusion or maybe kind of undermine collusion?
23:41
So, you know, it won't surprise you that myself as a long-standing competition economist who, you know, six years ago suddenly found out about artificial intelligence, and even in its current form as a statistical advance…where that…“Oh, this is an interesting problem”. If these AIs can be trained to play games, well, maybe, they can be trained to play economic games. And what's the best economic game we [can] think of – is how to firms actually achieve collusion without talking to each other? And, you know, it automatically is an intriguing legal question, right? That if you can…what the people were doing to get someone to win Go. They said, you know, train an AI and give them [sic] the objective of doing what it takes to win a game of Go. That's actually sort of hard to sort of, you know, structure and do. You know what's easy? Is to take an AI and say, “Maximize profits, maximize profits”. And I can now say to an AI, “And don't just maximize today's profits. Maximize profits going off into the future”. And, so, one could imagine a situation where Firm A and Firm B both employ AIs, and the only thing they've told them to do is maximize profits. They haven't told them anything else. And, you know, and then they give them power over setting prices. And, the question is, where are they going to end up? Now, from [an] economic prediction point of view, if they were all fully, you know, super rational agents, they could easily end up with a collusive outcome. But, they're not quite super rational. They're a bit reactive to that. In other words, they don't have a model in their head of what the other player is doing. They have themselves [sic] data when they did stuff of what was going on. And, so, at the time, it sort of seemed a bit hypothetical to consider. We might have this sort of getting around the law collusion because we've told our AI there is nothing wrong. We may have had the intent that they're going to make lots of money, but, you know. And, we certainly haven't told them not to collude. But, you know, do you have to do that? You tell me. That's an interesting question of whether you have to actually explicitly tell your machines, “Please do not collude”, whether that's a thing. But now, what happened was, of course, that many economists got interested in that, and they started building these things. You know, why not? You could build an AI in a simple market and see what happens. And, you know, funnily enough, you know, sometimes you got competition. But sometimes, with very simple models, you could get some collusive outcomes as well. And some things were less surprising. The more you told AIs to worry about the future than the present, the more likely they were to find a collusive outcome. And none of this is new, by the way. There was back…30-40 years ago…there was a computer tournament given where computer programs will [sic] be programmed to play the prisoner's dilemma. The prisoner's dilemma is a classic compete versus cooperate type situation. It's actually…we have the same in…price competition is a prisoner's dilemma – the prisoner's being the firm's and the dilemma being they don't want to…they would prefer not to compete with each other. But the…and they played…you know, they set out these little algorithms. And they basically said, “Well, if you win, if you get some higher payoffs, you get to propagate more and, if you lose, you don't”. So, it was like a natural selection, survival of the fittest type situation. And, at that time, an algorithm that did the tit-for-tat strategy. That is, I'm going to play nice and try and cooperate unless somebody has wronged me, in which case, they're dead to me. I'm going to not cooperate with them at all. And so that did quite well, in this environment. In the AI context, they were coming up with similar strategies, similar things sort of evolved from that, and would get some, you know, even if it's not collusive outcomes, some higher than competitive price outcomes. And there's all sorts of wrinkles to that. You know, one wrinkle was, you know, what if the AIs were bidding in an auction, would it matter whether you had a first price auction where, you, the highest bidder paid or a second price auction where the highest bidder paid the second highest price? Turns out, AI's play those games very, very differently. They find it much easier to collude with a first-price auction than a second-price auction. So, that's interesting. And then another thing and given the title of this podcast, Counterfactual, is quite interesting. If you help the AIs along with some extra knowledge, for instance, you tell them that a demand curve is supposed to slop downwards – that is expect that if you get a higher price, you'll get a lower amount of demand in ordinary situations – or you program them to occasionally do random stuff to see what happens in the counterfactual, they do better at finding ways to collude. And they're not explicitly colluding. Remember, these are computer programs. They don't know what collusion is. They're just playing a way and anticipating the reactions of others and finding that spot where they are comfortable not lowering their prices and being competitive. So, they understand that things didn't work out when that occurred.
29:48
I mean, does this research tend to involve the use of the same AI for each of the, you know, theoretical market participants? [Yes, they do.] Because, obviously, in the real world, that's maybe not, I don't know, a less likely situation.
30:07
Who knows what it would be. The actual issue with these experiments is they’re run on very quick time. They're run without people actually, you know…when we said higher prices, then customers react in some way. Well, here, we will just go just assuming some functional relationship. In reality, there's just much more uncertainty than that. And so, you know, we don't know whether, you know, if you were a large multi-product online retailer pricing in competition with another large multi-product online retailer, whether those same AIs would be able to reach…find some sort of outcome. But, you know, at the very least, you've been able to, we've been able to, nudge AIs in the lab, let's talk about it that way, towards those outcomes, if you so choose. And it stands to reason that some of these firms will be thinking, I wonder if there's a way in which we can improve our profitability by using AIs. And, so, there might be experiments there. But again, you know, some of those things in the past have been quite dangerous. You know, there's situations where people have used algorithms to deal with pricing, and all of a sudden, you end up with books, priced at several million dollars on…by independent booksellers on Amazon, you know. And they're clearly not responding to sales. Well, somebody bought the book for several million dollars. I guess that means something. But they're posting their price. So, you know, that's something wrong. Or we see in computer stock trading, we see these flash crashes and stuff like that. And what that is, is some computer algorithm, getting a feedback loop that's all wrong and going off in the wrong direction. And someone's paying the price for that. So, really, you know, we're still a ways off seeing that. I, you know, I'm hopeful, hopeful as a person who's a competition, testifying expert, that one day, we will see one of these firms actually manage to do it, and we'll have a case about it. Because I think it'll be fascinating. I realize the word hopeful isn't what I should be using there. But I can't deny my actual feeling.
32:37
Intellectual curiosity always gets the better of us. So, I mean, putting collusion possibly to one side for a minute. I mean, I guess where else do you think AI may contribute to meaningful competition problems? You know, we hear a lot about the impact on network effects, perhaps, examples of self-preferencing, in vertical platform environments, things like that. But where do you see a lot of potential activity or thinking around this?
33:07
I think, I think actually, you mentioned self-preferencing. I think that's an area where it could come up. You know, if you're an engineer employing an AI to bid, you know, in what should be a neutral exchange, but you…the AI, sort of in their objective function realizes that the benefits if the trades occur on platform rather than across platforms are better for the person who programmed the AI, and now the AI itself, they will naturally discover some forms of self-preferencing. You know…that sort of thing, that sort of thing will come up, if you just had a salesperson and you said, you know, I want you to think about us and not the other people, they sort of tend in that direction. But, so, I think we're going to have situations where even if there wasn't an articulated PowerPoint, saying, “Here's how you should exploit your market power”, the AI will pick up the market power and act on it, if given the right incentives, or will send information back to people, or act on, that will end up having those circumstances. The analog or the way you already see this occurring is not in the competition realm, but in, say, using AI for hiring practices to sort through CVs or do anything else to help with hiring. We know that those AIs are trained usually on data sets that involved humans. And, to the extent that they were, they often can inherit the biases of those humans. The one thing that's useful there is we can have an AI measure this a lot more easily. I don't know…and so those algorithms can be fixed not to have those biases, so they could actually improve on people in that way. With respect to these…to ones that might be deployed in different parts of business, that may not be as readily apparent. I think, I think we're still a ways off, being able to note that. But really, you know, when it comes down to it, what an AI might do is not that much different to what a highly incentivized engineer might do. If you give…if you're in a tech industry, and you give your engineers control of the code, and you also say your bonus will depend on firm profits. You know, regardless of whether they know about antitrust law or not, they're going to just try and maximize profits, which may lead to things that are, you know, without a keen eye on it will look like, oh, this is arising because you had market power. So, I think, I think those situations already occur. And I think, will they have an AI at the heart of it at some point? Yeah, yeah, they will. There'll be some non-neutrality that keeps in. The AI works out, it would be better if we just cut off these people entirely. They're never giving us any good, these suppliers or whatever, and you'll suddenly have a foreclosure case there.
36:31
Now, this may be straying a little bit beyond our remit. But, you know, you mentioned some examples about, you know, sorting the CVs, for instance. But are there, you know, do you see AI as contributing to other, you know, significant non-competition issues? Or, if we can call them that, you know, whether this relates to privacy, deceptive marketing, or even, you know, national security. Our listeners are very interested in, obviously, national security issues as well.
37:03
Well, no, well, my presumption would be that the most sophisticated uses of artificial intelligence are already occurring in the national security realm. It's just that, for good reason, we don't hear about them. But, you know, that's where the incentive is. And then there are applications galore that could occur in those environments. Will it impact on privacy? Yes and no. You know, privacy is [sic] always been a complicated beast. Are there algorithms out there now being trained and developed on your data? Yes. If they didn't have your data, would it be…make a difference? Probably not. So, you know, your data is being used, but it's not really necessarily consequential. The way in which this starts to become more consequential is through personalization. If I can feed the algorithm that has already been trained and [is] already running, more data about you, they can work out better what you should buy, what ads to put in front of you, all those sorts of things. Is that, you know…if they do it with your permission, there's of course not a violation of privacy. If they do it without your permission, or they're doing it for somebody else who wants a personalized prediction about you, then it gets complicated, right? So, those sorts of situations are…we will eventually get much better at distinguishing those as we sort of apply the stuff or as…privacy law is evolving law all the time. It goes hand in hand with technology. So, I'm sure we will be doing that activity there as well. The final place that…where AI is going to prove interesting is intellectual property, per se. There's been a case in Australia where somebody applied for a patent on an invention, a discovery. And they listed that the AI, the software that was used, as an inventor. And the initial court in Australia found that that was okay. That you could list this as an inventor. I don't know if that's going to stand. I found it ludicrous because I was saying well, that's like saying, “I'm you know…I discover something using statistical software or maybe Excel does some calculations and I'm going list Excel as an inventor on this thing”. You know, like, I think…I know why the person involved filing the patent did this because it [sic] was trying to promote how good their software is in inventing things. But, you know, can we assign a patent to a statistical package? It doesn't sound as obvious. But the court was…may not have had that knowledge of what the AI really was being put before them.
40:21
Right. I mean, I suppose that could contribute to competition issues as well.
40:26
Yeah. No, absolutely. I mean, if people, if people are going to say, “Well, we've deployed in AI”, and people say, “Well, if the AI had the intent to do blah, blah, blah”. And, of course, people might describe it that way. Sometimes we anthropomor[phize]…too early…late in the afternoon to use multi-syllable words. You know, if we humanize that type thing, you know, we can be led astray. But I should tell you right now, there is no AI out there whose objectives are not actually programmed in by a human. And, so, from, you know…if that is a factual question that determines intent, there's something…you know, that has come from somewhere. AIs may learn to collude and other things like that, but you have to give them the…they have to have a North Star. You can't put one in without a North Star. You've got to type it into the code, even if it's some sort of default. You're doing it. So, it's very important that we realize that it's not the AI doing something. It's often as a result of the very conscious decisions of people.
41:41
So, we haven't read yet reached the sort of “Revenge of the Robots” stage of the Sci Fi world that we live in yet?
41:47
No, no, we haven't. I mean, that's the, that's the best stuff to talk about. I mean, it's so much more interesting than everything I get to write about. That sort of thing. You know, the robots being mistreated, and then coming back to haunt us. I mean, I think I suspect everybody in the back of their mind is far more polite with Siri then would surely matter. Because, at some level, they're sort of wondering if someone's going to take note. But, you know, that's fantasy, of course.
42:21
Right. Right. Well, maybe this is a time where we might switch gears, because we like to include in our episodes a segment that we call Overtime. And Overtime is where we take ourselves outside of regulation play and explore some additional lesser known dimensions to our guests, including their personal interests and pursuits in compliance with [the] National Hockey League shootout format. In our Overtime segment, we take three shots at getting to know our guests a bit better. So, here goes. Professor Gans, you hail from Australia. I've had the occasion to visit Australia quite a number of times. And it's lovely. The Toronto waterfront is not exactly Bondi Beach. What do you miss most about living in Australia?
43:11
I actually miss the parts of the slower pace of life. You know, in North America, less so Canada than in the US, of course, you know, obviously, getting together with people often requires an agenda of what you're doing. That wasn't the case in Australia. You know, I wouldn't…we'd never have issues getting colleagues to come in, you know, and have lunch together every day and things like that. But, here, things are just a bit busier and more of a rat race. So, I do miss that. And I do miss that sort of a more laid-back nature.
43:57
Having lived in New Zealand for a number of years, I can see where you're coming from there. So, question number two, with your finger apparently on the pulse of business innovation, what would be your favorite gadget or innovation that affects your everyday life that you might recommend to others?
44:19
Okay, you're going get a sales pitch because this is always what happens when I get asked this sort of question. The gadget that everybody should look up is called a Thermomix. Exactly how it sounds. Comes from Germany. And what it is, is a, you know, a mix of a heating pad, a blender, [and] a whole lot of stuff hooked in with a computer. And it's basically as if Apple had designed a, you know, almost universal thing in the kitchen that could help you cook and make you into a top tier chef. Well, not a top tier chef, but at least someone who apparently knows how to cook. And not just cook a very limited range, but thousands of recipes. So, we use our Thermomix, five, at least five nights a week to cook dinner. We have…we've parceled it out the children for many, many years. Because they…because the problem of getting kids to cook is it's terrible – what they might do. But, with the Thermomix, they were able to cook, you know, adult-quality meals. And, so, we did that for a number of years. So, you know, it costs something like $1,500 plus in Canada. But I can tell you, it is well worth it. It can cook anything. It can cook bread…can bake bread, it can make stews, cake. I make a mean cheese cake now.
45:54
Wow, is this sort of like a portable oven type situation? How big is it?
45:57
It's very hard to describe. It's like a pot on a thing. And it's just…beautifully…it's like only three buttons on it. It's beautifully easy to use. It's got a scale incorporated. You can put…it tells you to put it in this amount of spice here. The mixer does…sautés, slow cooks, whatever. You know, it makes ice cream, if you want it…I can't describe the beauty of this product. It is just amazing. It's a product that actually was…it turned out. I mean, I didn’t know this at the time. It has in recent decades taken Australia by storm. It's the biggest [demand] for this product, but it's been around for decades. But they've kept up and I don't understand why everybody doesn't have one of these.
46:41
Okay, that's a fantastic answer. And I'm not sure we'll be able to outdo that one on question number three.
46:47
I know what's going to happen now. I'm going to go to conferences…law conferences in Canada, people are going come up to me and say, “I saw that Thermomix, and I bought one, and it's changed my life”. And you know.
46:58
Well, of course, just in compliance with testimonials and advertising, I don't know if you have any affiliation with Thermomix?
47:07
No, but I should, I should, I’ve sold a lot of these things. I have sold quite a lot. I mean, they should, should be…
47:14
You should be getting a commission, for sure. [Exactly.] Dear. Well, so, my last question is sort of similar, but maybe lower tech. But is there anything that you're reading or maybe watching in your spare time that you might recommend to our listeners?
47:33
Oh, that's a very good question. I'm always watching something. The show I keep telling people, you know, with a blank slate of what they should be watching is a show on Apple TV+ called “For All Mankind”. And it is basically an alternative history of the world and the space race, if the Soviet Union had beaten the US to the moon. And it's just fascinating how much could change quite plausibly with that, with that difference. And so I find that fascinating. The politics are fascinating. There are, there are issues and other things that come up, that you know are coming because stuff happens. We know that history. So, it's a delight to watch that show.
48:30
Oh, fantastic recommendation. Will be sure to look that up. So, I guess, coming back to AI. And I'm just thinking now kind of more macro or a societal level. You did mention a few times in your prior remarks about sort of how workers are maybe being affected, or certain types of functions are maybe going in or out of style as a result of this, or maybe more or less valued, but how do you view AI as maybe changing the nature of work?
49:08
I think, like a lot of things, you know, our work changes very much when we take tasks that we previously did, and we automate them to some degree. You know, we're living that right…doing whatever we're doing right now is an example of that. And, so, AI is another thing on that path. AI has the potential to take tasks that we've done and automate them. I mean, even at the moment, some of these AI applications are reading, sucking up all of the scientific literature and allowing you to ask questions of that literature without actually having to read the primary source. Basically taking away a few steps in the research process and in the process of research there. And even in, you know…let me go to the to the legal profession. That's something where these things are being used quite a bit. There's a company here that went through the Creative Destruction Lab and is founded by a couple of University of Toronto law professors, including a former student of mine from Australia who's part of it. It's called Blue J Legal. And they had the idea of taking the tax code and the tax decisions, and using artificial intelligence to answer questions of interest to taxation lawyers and attorneys, and, it turns out, judges, you know, because it's a very complicated area. And, you know, it was one way…it wasn't sure, you know, could you do this? But they seem to have been quite successful at it. I can imagine many other, you know…tax codes are very…people…you know, so hard to keep everything consistent. It's the whole point of that complex thing. So, it turned out that was a very amenable part to artificial intelligence doing that job. But there are other situations. There's another firm that we had that would take a legal document and [sic] you give it some instructions, and it would retract things, at least do the first pass, or getting 90% of the things that ought to be retracted in a document retracted, and then you could go in after that, and tweak around it. But you can imagine that's just taken off some time from you as well. And, so, you know, things that you've handed out to a paralegal or something like that may start to be capable of automation, which will change, you know, the nature of how we work. You know, back 30 years ago, you wouldn't be typing your own documents or statements, but, of course, now we have tools that have now handed that to you to do. But AI will do the same sort of thing. And an entrepreneur, especially in the legal…is one area, are really looking for those opportunities.
52:07
And I guess you talked about streamlining, but does AI create jobs? I mean, other than, obviously, the people who are generating the AI, but…
52:17
I think yes, it will. I think it will create jobs. I think in the sense that, you know…just to go back to our earlier example regarding ride sharing and Uber and Lyft. Those were technologies that unequivocally created jobs. There are many, many more people driving other people around than there were previously. You know, sometimes five or 10 times more. That's job creation going on. One can imagine other areas in which new tools will similarly empower new jobs to be…new employment opportunities. Sometimes those things may look very different from current jobs. That's harder for me to speculate on, though.
53:04
Yeah. And, again, sticking with societal shifts, you know, there's been a lot of discussion about echo chambers politically, and, in fact, I just read that the UK is looking to require social media companies to disclose their algorithms because of concerns about, you know, the sort of the atomization, or the isolation, of views and the inability to, you know, cross the hall and find consensus, but I'm not sure where you see things going in that direction.
53:40
I mean, I think it's…I mean, I understand the desire and the push. Can this actually be done? In a lot of cases like…can a social media company actually tell you what their algorithm is? Describe it in some way or show it to you – that prioritizes one bit of content above another? You know, I don't even know if there exists something. In other words, they might not know themselves. I know with TikTok, which is an AI first version of this…and we've seen how “successful” (got little quotation marks), it has been because it's developing algorithms quickly for every individual based on just, you know, microseconds of how they stare at different content, what their content is, etc., and throwing things up for them. Is there an algorithm that can be exposed there? You know, the best we can hope for is that we…so when they…let me give you a slightly technical bit. When an algorithm…algorithms are improving, they have to know how they're getting closer to a goal. So, at the heart of any algorithm is a decision on what has got a loss function. That is, what should the AI look to be squeezing to get closer to zero? They've got like an error. And what is the measure that's going into that error? Somebody has to put that in. TikTok and everything else that measure. What that is may actually be the most revealing thing about an algorithm – what that measure is. You know, so, for TikTok, it may be, you are doing better if you're getting people to stay longer at each individual thing that gets shown to them. They sit through the whole lot, and then they stay in a longer session. That might have been already coded in there. Well, that will be useful information. So, we may be able to do some expertise and look at that. I haven't seen the [sic] case come up with that. That will be fascinating. But that would be where…that's about as far as you can get. What have you told the thing to do?
55:57
I guess, coming back to, I guess, the competition sphere a little bit. But, again, in sort of a longer range view. I mean, we're constantly hearing about the, you know, the fangs, Big Tech, in economic and political discourse, and certainly in the competition enforcement space. And I guess, do you see a world in which the fangs, just to use a shorthand, are replaced in the not too distant future? You know, in economic cycle terms, like, where do you think we are in the process of disruption and renewal even for these big, big names?
56:33
You know, it's very hard to tell. You know, interesting, left out of some of those is Microsoft, which has been with us for decades now and has not been disrupted and is as valuable as ever. But it's not listed as a competition issue anymore. So, you know, I think there is a potential for some of these firms to get disrupted. But, you know, there's a confluence there. Some of these firms are doing things that many argue [are], and may well be, violations of competition authority, to slow down that process. And, so, you know, they may succeed in that. And, so, we wouldn't necessarily see a turnover. But, you know, something tells me that there will be those sorts of things. Look, we've seen entry into the car industry in the last 15 years that has now taken over and become all the way to the top, at least in terms of cars valued and getting further in terms of cars produced. That's an industry that for like a century had hardly any entry in it. And, all of a sudden, technology driven changes led to that. Cars we're far more less likely to be disrupted, then, you know, say, a search engine or even a social media company. I think there's…so there's chances that that will occur. I think the big AI companies…We won't even call [them] big AI companies. We might call them bigger systems. They might turn out to be a health system or set of educational programs or something like that, that ends up being that. Who knows? Still open.
58:24
Yes. No, no doubt. I wanted to hit on something else. And this is, you know…you're obviously involved in the…is it…the Destruction Lab…[Creative Destruction Lab. We’re not all about destruction.]… Creative Destruction, of course. You're not just going out to destroy the world, but to do it creatively. But, you know, there's been a lot of talk in competition circles about so called, you know, nascent competitors, budding competitors, you know, something that might come out of an innovation lab. And, you know, what happens when incumbents or established players, frankly, whether or not directly competitive with these nascent competitors, you know, acquire those competitors at a relatively early stage, and should competition agencies be held to a relatively exacting standard of predicting what the outcomes will be absent an acquisition versus with the acquisition? And are, you know…is competition being snuffed out or, frankly, is competition actually being accelerated by virtue of getting access to capital and finding applications that may be quite helpful, and get to market more quickly? I just wonder what your sort of overall thoughts are on the concern about acquisitions of, you know, small competitors that are…may have innovations?
59:58
You know, I tend to be on the side of, you know, it is very, very hard to predict a problem, you know, based on a forecast of technology. That is, the potential uncertainty and regulatory uncertainty that could occur. You know, startups, you know, are operating in a space where they have to compete intensively for capital. They for many years are scrounging along at the edge of bankruptcy, even the best of them, until they sort of find their footing. And the problem with, you know…if we said, “Tomorrow,”…We said, you know, “Any firms valued over – I don't know what you want to call it, let's call it over – $1 trillion, can't acquire any other firms”…that itself would have a chilling effect on startup investment, because, you know, the startups have goals, they've got aspirations, they've got the, you know, like, “If things work out swimmingly, we'll be an independent firm, IPO, competing against the big guys. If things work out terribly, I'm going to be bankrupted”. In the middle, there are other options. And, you know, would you want to fund a firm facing uncertainty that you cut off the middle option, and it's either all to the moon or bankruptcy? And I suspect not. I think, I think, you know, the vast majority of firms at some level have on their list of options the ability be acquired. Now, that does not mean that at the time of the acquisition, when that appears, we can't look closely at the rationales for it, etc. For instance, you know, looking for discussions, evidence, or, you know, evidence from the marketplace that the reason for this acquisition is to take out a nascent competitor. That should always be something we could do. And especially if that's the intent. We should try and uncover that as the intent. And I feel that when we sort of, you know…it's interesting…when people look back, for instance, of the Facebook acquisition of Instagram, there's two things that they forget to consider in that environment. Instagram is big now. But back when it was acquired, it was 30 million users, and it was only on one platform, the iPhone. Now, was it going to expand to Android? Yes. Was it going to expand to more users? Yes. Did it have growing pains in doing that? Absolutely. Might not have made it. So, that's issue number one. Is that…in any stretch of the imagination, you know, it was hard to see how it was directly competing with Facebook, and it wasn't 100% clear it was going to continue to compete with Facebook. The second part, though, was to look at what Facebook was doing. What was Facebook doing? Or just before Instagram was acquired, people forget that Facebook launched an Instagram clone called Facebook Photos. And I…that doesn't seem to come up very much anymore. But I remember it because I remember watching that occur. And, so, if…once you take that into account, that Facebook seemed to be, you know, copying this model and competing head-to-head, well, that's a different situation, isn't it? Now, where was that in the law enforcement activities back in 2012? That's what I always had a question about.
1:03:50
Very interesting. Now, look, there are so many other issues we'd love to discuss and we could go on for hours. But we've taken up enough of your time, I suspect. Professor Gans, it's been a real pleasure to speak with you. I'd like to thank you for taking the time to share your knowledge and insights with the Counterfactual podcast. We'll certainly look forward to hearing news of your ongoing research projects and hopefully catching up again soon.
1:04:14
Great, thank you so much.
1:04:16
Thank you.
1:04:17
Thank you for listening. Counterfactual is produced and distributed by the Competition Law and Foreign Investment Review Section of the Canadian Bar Association. The opinions expressed by the participants in this podcast are their own and do not necessarily represent those of their employer or other organizations. If you enjoyed this podcast or would like to join the Canadian Bar Association, please visit www.cba.org/sections/competition-law.