← Digital Disruption with Geoff Nielson
Digital Disruption with Geoff Nielson
AI Bust: Oxford Economist on Why the AI Boom Will Be Short-Lived
01:03:43
Loading summary
Transcript44 lines
- [00:02]
A
The best case is that what we're producing is going to take your job, and in worst cases, it's going to kill you. But please don't regulate us nonetheless.
- [00:13]
B
Hey, everyone. I'm super excited to be sitting down with Professor Carl Benedict Frey. He's Oxford's Associate professor of AI and work and the author of two bestselling books, the Technology Trap and How Progress Ends. As an economist and historian, Carl is intensely skeptical of the vision of the future that we're being sold by big tech, and thinks we have a lot to learn by studying other periods of rapid industrialization and technological change. Through his research, he thinks he has the recipe for why some countries have been able to use technology to get ahead and why some have been left behind. I want to know what that recipe is and if it still applies or if this time really is different. What happens next to our jobs, to our countries and to ourselves. And what do we need to do to get ready? Let's find out. Well, Carl, thanks so much for being here today. Maybe just to jump into things, I wanted to talk a little bit about your book, How Progress Ends, because I talk to a lot of people on this program, obviously, about AI, about the future of technology, and hear an awful lot of stories about this world we're heading toward, of abundance and how, you know, society is going to be incredible and, you know, nobody can wait for the, you know, this next amazing thing. And your take is a little bit more skeptical than that. And I was curious if you could walk us through, you know, some of the risks societally that come with any sort of large technological change and what you've seen in your research that led you to publish this book.
- [01:45]
A
So if we go back to the 1990s, and you would have told me that as a result of the personal computer and the Internet, which essentially gave us, the world stored knowledge in our pockets that connected some of the leading talent around the world that streamlined the research process extraordinarily. If you would have told me that all you're gonna get from that is a decade streak upsurge in productivity, mostly confined to the United States, I would probably have thought that you are crazy because the technology was and still is, so promising and useful. And I can't stress that alone. Right before the computer and the Internet, you would spend, you know, days, sometimes weeks, just waiting for material so that you could continue your research. And what we get despite that is even a decline in breakthrough innovation and research productivity. And so what that tells me is that there must be something else Going on. That is not just about the technology. And so I'm not skeptical about artificial intelligence or the potential of the technology itself, but I think as things stand, it's not on par with the computer revolution yet. The computer revolution automated a lot of downtime, sitting around waiting for useful information. AI automates cognition or the sort of process of turning that information into a useful output. But as long as you need to verify the output, you need to be up to speed with the AI. You need to learn the things that the algorithm learns, you need to understand it in order to be able to verify the output. And so I think that artificial intelligence is quite different in that regard. And in that sense, I think it's actually even less likely to boost productivity. And what we saw with the computer revolution, which again strikingly was mostly confined to the United States.
- [04:18]
B
So when you see some of these studies that say the productivity gains from AI have yet to materialize, it sounds like you're not surprised at all by that. That's in line with what you would expect given the human factor in all of this.
- [04:33]
A
Well, it's very much in line with the historical experience. Look. So it took eight decades for steam to deliver its main boost to productivity. In Britain, it took around four decades for electricity to show up in the productivity statistics. We see a similar pattern with ict. You might think that AI should be a little bit quicker because AI is, you know, it's a production technology, but it's also consumer good and you can just download it on your phone or on your desktop and you can use it straight away. So, you know, I think that AI will deliver some boost to, to productivity. I would be very surprised if it did not. But I don't think we should assume that it's going to be as significant as many people hope. And it might also be very short lived. Right. So the computing revolution did not deliver anything like the sustained growth that we were seeing from 1920 to 1970, interrupted by the war. And I think that is in part because it didn't deliver new things, new products, new industries to the same degree that we saw back then. Right. So the automobile industry was the largest industrial undertaking that the world had ever seen. It produced not just automobile industry, but a host of industries making the components that go into making a car. A range of industries making machine tools needed to make the components. Electricity produced all the electric appliances you have in your home. There was an industry standing behind those. The car gave rise to road commerce, travel, mass tourism, and so it created a lot of new kinds of things. Whereas with the computer revolution, yes, of course, it's created new jobs and industries as well, but you know, it's been much smaller scale and mostly confined to a few places like the Bay Area. And when I look at artificial intelligence today, I think of it very much as a continuation of the computer revolution. Basically every application I can think of today has something to do with automation or process improvements. And those are important. But look, if all we had done since 1800 was automation, we would have productive agriculture, we would have cheap textiles, but that would be about it, right? We wouldn't have computers, antibiotics, vaccines, radio rocket airplanes, etc. And so unless artificial intelligence develops new types of products that lead to new industries and activities, it's much more likely to be a relatively short lived productivity upset
- [07:41]
B
if you work in it. InfoTech Research Group is a name you need to know no matter what your needs are. Infotech has you covered. AI strategy, covered. Disaster recovery, covered, vendor negotiation covered. Infotech supports you with the best practice research and a team of analysts standing by ready to help you tackle your toughest challenges. Check it out at the link below and don't forget to like and subscribe. It's a really interesting historical counterpoint to talk about, you know, cars and transportation and you know, in the confines of AI. You hear a lot these days about, you know, the AI infrastructure needed and the race between, you know, within the us, some of the hyperscalers, then of course China and some of the other big players. But even how small that sounds compared to, you know, with cars, the physical infrastructure, and as you said, you know, not just factories and electricity and components, but the roads themselves and the physical environment. I'm curious, as you apply more of that historical lens, we can see a situation where now we've got these major players racing to build this infrastructure. We had something similar with cars, and again, it seemed like it was very dominated by the U.S. but historically, and you mentioned the steam engine, the industrial revolution, we've had all sorts of big players and even empires try to capitalize on these new technologies to get ahead and leapfrog their peers in terms of economic power and also just terms of power in the broader sense of the word. What lessons do those share for what you're seeing with the next kind of great technological leap forward?
- [09:35]
A
So I think a key difference with the infrastructure that was built around highways, around railroads, around the electrical grid was the technology that was built to use and leverage. That technology was relatively mature at the time, or at least the science behind it was broadly understood. I Think with artificial intelligence, it's a much riskier bet, not just because the infrastructure itself seems to have a shorter lifespan, but also because we don't quite know right now what the future of AI might be. Right now we're seeing the magnificent Seven doubling down on large language models. That's one approach to AI. But it may be that the future of AI is smaller language models, or it may be that, you know, it's world models, or it may be that it's something very different. One thing that strikes me is that humans are much more computer and data efficient than AI is today. And I think, you know, a useful analogy is with the steam engine, right? And so the early steam engines that were tremendously energy inefficient, they were basically only used to drain coal mines, and even that they didn't do particularly well. Well, and so you needed James Watt's separate condenser to make steam engines energy efficient. And that's when the sort of steam revolution takes off. I don't think we reached that separate condenser moment yet with artificial intelligence. And in the end of the day, to which degree we need this massive infrastructure build out will depend on, on that moment and what the future of the technology looks like.
- [11:55]
B
So if I can feed that back to you, it sounds like you're not necessarily skeptical of the long term trajectory of this technology and its ability to revolutionize things that we're doing, your view is more that we're still in early days and we probably need another massive invention or two to truly unlock what's capable here. And, and that if we're getting too far ahead of ourselves building out all this infrastructure before we create the equivalent of the condenser, it could all become redundant and not yield any return. Is that fair?
- [12:31]
A
I think that's a fair assessment. But I also think in addition to that, and that's the key point of the book, it's not just about the technology. It's not just about the tech, it's about the institutions and the incentives. And, and so when you get powerful technology that allows you to do more stuff, you can use that either to use those efficiency gains to dig deeper or just drill more holes. And I think one of the reasons that we seen that breakthrough innovation has declined, even though we have much more powerful tools at our disposal, is that in academia the incentive is publish or perish. And so we've seen people taking on more projects as a result of these productivity enhancing tools, and we see that people's attention is then more thinly spread across multiple projects. And the research shows that the more things you do at any given point in time, the less likely you are to make a breakthrough. And so I think those incentives matter and AI is not going to change that. And then there are many other institutional bottlenecks. Pharmaceuticals, even if AI supercharges medical discovery, which is still an open question, but seems reasonable to think, you still have to go through clinical trials and that's still tremendously expensive. And so you need to partner with a pharmaceutical firm to do that. And so it means that few players can do it. And so that might be an extreme case. But I think that in many industries there are these sort of institutional bottlenecks that makes it harder to realize the gains from the technology. And I think if you look at business dynamism or just entry of new firms, you would have thought that that should have been through the roof over the past two decades because computers, the Internet, the cloud, now AI, or they all make it so much cheaper to operate a startup and come up with new ideas. But we're seeing the opposite. Business dynamism is in decline. And so there needs must be some countervailing force that more than basically offsets those declining costs due to technology. And I think that's important to keep in mind.
- [15:26]
B
Line the, the point about business dynamism declining is, is a really interesting one. So, so in your view, what, what is that countervailing force? Is it just us reaching the limits of, you know, h, how increased efficiency and productivity tools actually help us or what's, what's leading to that decline?
- [15:46]
A
So I think there are a number of things there. I think if you look at the United States, for example, corporate lobbying has been basically doubled since the 90s. We see that that's associated with more protective regulation which makes it harder for new firms to enter. We see a revolving door between patent examiners at the U.S. patent and Trademark Office and some incumbent companies where they grant low quality patents for these firms and then go on to work for them. We see rise of some anti competitive practices like killer acquisitions whereby incumbents just know, basically buy up promising startups just to shut them down. And more broadly, I think, you know, the patent system may not be really up to the task for the digital economy. It works well in, let's say pharmaceuticals and chemicals where product life cycles are long. But you know, it doesn't make much sense to grant in, know, 20 years of patent protection for technology that's obsolete in 18 months. And then you have, you know, in all of these intersecting patents in digital technology as well, which are often poorly defined. And so it means that you only really know what they cover when you end up in court. And that means that litigation costs are very high and there are particular concern for smaller firms. And so I think this sort of taken together has created significant barriers to entry. And I think it's a big part of the story.
- [17:28]
B
I'm glad you brought that up. And it's something that in a lot of conversations, certainly that I have with people I speak with, and also that you just hear about, either in an undercurrent on the news or if you're reading Reddit or forums, is just this undercurrent of frustration with larger firms for doing this, with the political system for allowing it to happen. And, you know, I'm curious, in your mind, it's sounding like this is actually a bigger issue right now than the advancement of the technology. If we're going to have, you know, kind of a flourishing business environment for the next handful of decades, like, how, how important is it for us to course correct this if we're going to be able to build this kind of prosperous future?
- [18:21]
A
I think it's really important. And also because young firms tend to be the ones that develop new products and industries, larger firms, incumbents, they have scale, and so they're much more likely to focus on bringing down costs in what they're already doing, focusing more on automation and process improvements, whereas smaller firms that don't have scale, they're more likely to develop new products. Now, both of those are important, to be clear, right? The Model T would not have become the people's vehicles if it wasn't for the capacity of the Ford Motor Company to, you know, improve production processes to bring down costs. So they are complementary. But if you don't see entry, then you don't see as much new products, industries, jobs being created and as you're seeing old ones being rendered obsolete by automation. And I think AI is likely to exacerbate that trend unless we do something about it.
- [19:45]
B
Well, and that's exactly it. It seems like a lot of these firms are very deliberately trying to own these entire new platforms of infrastructure that the AI sits on to give themselves even that further advantage. And the, you know, you're starting to see, and there's all sorts of stories in the news, especially in the US with the government about, you know, how the government is in some ways fueling that and trying to get in on the take. So is one of the issues whenever lobbying and the proliferation of lobbying comes up is this sense that it's just an intractable problem and, you know, there's too many incentives from too many, you know, rich entities to make it go away. And, you know, you get the Bernie Sanders is of the world or, you know, some of these, you know, some of these people advocating for, you know, fighting back more against these. These enterprises that are doing that. Is it intractable? Or, you know, do you. Do you have any recommendations for what the US or any other kind of advanced country or society dealing with this can be able to do to create, you know, more business dynamism?
- [20:51]
A
So the United States had a similar challenge in the late 19th century and commonly referred to as the Gilded Age, where public private partnerships had led to widespread cronies. The way the United States responded at the time was by making the civil service more meritocratic through the Pendleton act, introducing civil service examinations, and then using that state bureaucracy to institute and create new institutions to safeguard competition, including the Sherman Antitrust Act. And what we're seeing right now, I think, is very much going in the opposite direction. There has been a tendency during the Trump administration, for example, to focus on government deficiency, which sounds good in itself, but the reason that the irs, for example, is a huge operation is that the tax code is extraordinarily complex, right? And so if you wanted to reduce bureaucracy, then you would probably want to introduce something like a flat tax, and that would simplify things, and then you could reduce bureaucracy and achieve some efficiency gains that way. If you do the opposite, you cut down on the bureaucracy without simplifying the tax code, you're just diminishing your own capacity to collect taxes. And the same logic, I think, applies across most domains. And so right now, rather than focusing primarily on simplifying rules, although there's, you know, some of that going on too, the focus on. On cutting the civil service means that the US Is actually going in the opposite direction that it was going in the late 19th century in its attempts to revitalize competition. And so I think that is unfortunate because I think there are certain things that we can learn from the past. And one of those lessons is that you actually paradoxically need strong governments, a strong civil service to enforce competition from time to time. Now, that role can certainly be overstated, and there were certainly cases where the administrations may have overstepped. But, you know, if you. Basically every case from, you know, the first Industrial revolution, where the parliament clamped down on the guilds, that unleashed competition, that made the sort of first industrial revolution possible to today, I think that logic still applies.
- [24:25]
B
It's really interesting. And you know, looking at some of the examples you dive into throughout history, there's this interesting sort of symbiotic relationship in some cases between governments and some of the large and powerful firms of their day that they have a close relationship with. And you've got cases I'm thinking of, for example the Dutch East India Company or some of what's going on with China and some of the big corporations there that are know, kind of hand in glove with the government. And then, you know, to your point about the early 20th century in America, you've got the government playing more of a, you know, trying to break up anti competitive practices, right? And you have, you know, Standard Oil that's getting broken up. And so I'm curious, when you look at the historical context, do you see, you know, more of a case for bringing in a breakup of some of the big tech players now or do you think actually that the consolidation is good if the object is for a given state or society to get ahead and stay prosperous?
- [25:40]
A
So key theme of the book is that technological progress entails both experimentation, exploration and then execution exploitation, which essentially amounts to getting things done. Scaling up what you invented and institutions that support those different faces can be quite different. And so at the frontiers of innovation, I think decentralist competition is really important, right? And so if you take the Soviet Union as an example, right, it was the most centralized economy the world had ever seen. If you were an aircraft engineer in the Soviet Union and you wanted funding for your project, you could go to the Red army. If they declined, maybe add two or three other options. If they declined, your idea would basically die with you. That's quite different from the US system of much more decentralized finance, where Bessemer Ventures famously declined to invest in Google back in 1999 and probably regret it today. But it also illustrates Google wasn't a safe bet at the time, right? AltaVista and Yahoo, they were dominating search. And so somebody needed to take the risk to invest in order to figure out whether Google would catch on or not. And so in a more decentralized system, you have more people that can take different bets. And that's really important for innovation. And so for example, the breakup of AT&T was really important for the development of the Internet and E commerce around it because it meant that when the National Science foundation released Arpanet, the predecessor of the modern Internet to the world, it wasn't just handed over to sort of the, the monopoly carrier service that would probably have bottlenecked it in their boardroom if they had seen no commercial use for it. And there was no commercial use for Arpanet at the time. And so that meant that, you know, the technology could develop more organically by its users and you know, businesses being built around that. But you know, that's quite different from catch up growth. So if you're behind the frontier and you can just make use of technology invented elsewhere to grow, then you're much more likely to have this symbiotic relationship between government and business. Or you can even have it done sort of directly through the government, as in the case of the Soviet Union. But if you take Japan for example, right. So to catch up with the Meiji Restoration, the government played a key role in establishing a range of pilot projects. They went abroad and, you know, scouted new technologies and projects, brought it back home, sold those off to emerging zaibatsu conglomerates who then, you know, merged around those technologies and effectively scaled it. And there's been a very close relationship between government and business ever since. And you know, Even World War II and then, you know, the arrival of the allied occupation authorities which sought to introduce American style antitrust in Japan, essentially, you know, failed. It was very short lived. You know, these zaibatsu conglomerates essentially just transform into keiretsu in the post war period and then do very well in catching up to the United States. To the point that in the 1980s everybody worries that Japan is about to overtake the United States. But then obviously growth in Japan peters off in the 90s, whereas America ends up leading the computer revolution. And a big part of that is that Japanese model was very anti competitive. If you were part of these conglomerates, then you would benefit from technology transfer and stable trade. But that also meant very high barriers to entry for outsiders.
- [30:15]
B
Right.
- [30:15]
A
And so Japan completely misses the switch from hardware to software and the rise of E commerce. Whereas in the United States, antitrust against IBM, which forces it to bundle hardware and software provides really critical. And then as I mentioned earlier, the breakup of AT&T. And so government and businesses can have these close relationships and that mostly can help catch up growth. If I should say, the government can play the role of the enforcer and making sure that businesses use the privileged position to make productive investments. And that has often been sort of tried to ensure that through export discipline, making sure that these firms export into global markets, which means that they need to be sufficiently productive to achieve that.
- [31:14]
B
So that's exactly what I want to hone in on, given what it sounds like is a Lot of historical success with antitrust behavior. Regardless of government appetite right now. How strong in your mind is the case for antitrust against some of these big tech firms and actually starting to break them up and kind of crack open the competitive landscape and what's the trade off there with their scale, enabling them to build out some of this infrastructure at scale.
- [31:46]
A
So I'm not hugely convinced that vigorous antitrust against Meta or Google or any of these firms would be hugely transformative, I think more likely to lead to one social media platform being replaced by another. What I do think is important is to deter some of these anti competitive practices I mentioned earlier. And I think it is noteworthy that if you look, you know, at the early days of the computer revolution, from Apple and Microsoft to Google and Amazon, essentially every major company that emerged, IPO then became a firm, continued to be a firm in its own right. If you look back at WhatsApp, WhatsApp, Instagram, YouTube, all of them were acquired and you could, you know, reverse some of those potentially. Again, I'm not sure that that's sort of the main concern. I think the main concern going forward is trying to shift back the balance and to make firms IPO rather than exiting by acquisition. Part of that may have to do with things like merger review, focusing more on valuation than turnover, for example. Part of it may have to do with making it less costly to operate as a public firm in terms of compliance cost. And so I think trying to tilt that balance should be a key priority. But, you know, trying to reverse some of the mistakes of the past, I'm not sure that, you know, that's going to get us to where we are, need to be in terms of dynamism. Got it.
- [33:58]
B
So let's, let's pivot back maybe to the technology itself and what it can do and what it can't do. And I want to come back to something you said earlier about, you know, digging deeper holes versus, you know, digging more holes and maybe we'll, we'll stay on academia for a minute. Just, you know, given how, you know, valuable an example I think it is because this is something that's been bothering me for a while now, which is if, if AI can suddenly, you know, dramatically increase efficiency in certain areas, basically reduce the cost to zero or near zero, call it an order of magnitude cheaper. You use the example of publishing papers. And you know, in academia, if you enter a world where suddenly, you know, an academic can crank out a new paper, you know, even every week, and suddenly there's just this proliferation of papers. Is that good? It feels, I mean, frankly in some ways it feels worse than the status quo. And then I worry about, well, does that mean we just need a 10th or a 50th of the number of academics and is that good? And so I'm curious in your mind, especially in academia, what are the deeper holes and what are some of the traps around potentially increasing efficiency so dramatically?
- [35:21]
A
So I think if you're an editor at any of the leading journals and you have to deal with this enormous upsurge in productivity as measured by numbers of papers being produced, then you're probably not very happy with the situation. And you're also not really in the position where you can use, use AI for the most part to sort all my change. But in the end of the day it comes down to the quality of the papers being produced. And there are numerous instances where AI has been used to spot flaws in existing works or papers that already been published. And I think we will see more of those. And that's good. I think on balance it's like, you know, with a computer and the Internet, these are good and useful tools for research that we shouldn't discount. But the question is, do you use it to improve quality or quantity? And I'm afraid we go more for quantity so far since the release of ChatGPT. But as I mentioned earlier, even, even before that, with regard to whether we, you know, need more or fewer academics, obviously, you know, I have a bias there, working at a university and you know, I often hear these things that, you know, 90% of all academics are useless, they don't produce anything of value, etc. Etc. But you know, with innovation and discovery in general, you really know what's going to turn out to be fruitful before somebody is actually carried out. And so there's always a lot of waste in the system and the successful things that come through, they more than sort of make up for it. So, you know, we can well aspire to reduce use waste, but I think that that concern is somewhat overblown. But I think you asked a more sort of broad question about the labor market implications of this. And I think right now AI is not sufficiently good to replace most people in what they do, but, but it's sufficiently good that somebody without a lot of expertise in a particular domain can produce content that in many instances is competitive. And so we have a range of experimental studies looking at this and it seems that in most settings, novices and low skilled workers benefit more in terms of productivity from these tools. And so I often make the comparison to, you know, UPS technology and taxi services. And so with the, with the arrival of GPS technology, knowing the name of every street in Toronto was no longer a particularly valuable skill. And then when Uber and Lyft arrived with the platforms matching supply and demand, basically anybody with a driver's license could get into their car and top up their incomes on the side. And that was good for them and meant more people in taxi services. But it wasn't very good for incumbent drivers who took a salary cut because of the, or a wage cut because of the competition. And, and a key difference in knowledge, work and in professional services is obviously that that work can be traded even across borders. And machine translation is making that, you know, even easier because language has historically been a key barrier to trading services. And I think that is gradually being, being eroded by machine translation. But what that means, if you take an accountant in, in Toronto and compare that to accountant in Manila or Cairo, the Toronto accountant will earn more, I suspect, on roughly an order of magnitude more. Right. And so if AI then reduces the productivity difference between the content in Toronto and the country in Manila, then obviously we'll hire more in Manila. And we're seeing that, you know, already anecdotally, you see Clifford, for example, reducing headcount in London, hiring more in India. You see firms like McDonald's that struggle to compete for AI talent in the United States. They can, you know, move operations to India and get, you know, great AI talent at lower costs. Right. And so for the workers in Canada and the U.S. and Europe, obviously this is the equivalent to automation. Right. It doesn't really matter to me whether AI takes my job or somebody in India takes my job. Right. I'm out of job in any event. And so I think that is sort of going to be the first wave more of offshoring in AI induced offshoring. And I think that's going to be quite disruptive indeed.
- [40:59]
B
That's very much in line with one of my takes that AI is going to sort of supercharge some of the offshoring we've already seen with some of the, and that a lot of the, a lot of the jobs and tasks at risk from AI are the same ones that were already, you know, earmarked for being offshoring to begin with. You started that. The answer to that question, talking about how AI actually has the most benefit to, you know, those who are newer to the workforce or lower skilled versus higher skilled, and where the conversation sort of evolved was yes, but maybe those are very different People or there are people in a very different geography than who would necessarily be hired for some of those jobs today. And so it sounded like the implication is still that if you are in the developed world and you are coming out a recent graduate or someone who's new to the workforce, then the roles that you traditionally would have been hired into are still very much at risk. Do you, do you buy that? Do you believe that? Are those the most at risk roles? Or how do you kind of see the impact on jobs more broadly?
- [42:17]
A
Yeah, no, I think that's right. I think broadly speaking, junior roles are more at risk, senior roles, which are more about leadership, management and you know, roles that require more experience art in the real world, which AI doesn't yet have, and those are less at risk. But obviously down the road, every firm needs to have a talent pipeline, right? And that's the sort of the tricky trade off. And so how much can you get in the short run in terms of efficiency gains by using AI rather than hiring new graduates, but without, without undermining the talent pipeline that you rely on in the end of the day? And I think that's a question that many firms will be grappling with over the coming years. And I suspect that they will sooner or later figure out that they need the talent pipeline and that the labor market for graduates is going to somewhat improve, at least as a result of that.
- [43:41]
B
So, so if you follow that thread through and you know, I, I imagine as I, I think you sort of implied that there may be an over rotation toward AI and then the pendulum swings back and they realize, oh, we actually need that pipeline. If you zoom out a decade or so, maybe more, what do you see as being the broader implications for the labor market from this technology? And then based on the labor market implications, what does that mean for us societally? And what, if anything, do we have to do to kind of course correct it and keep our societies and our economies vibrant?
- [44:20]
A
Obviously that depends a lot on the trajectory of, of the technology and how fast sort of the data and compute efficiency concerns around AI are solved. I think one thing that I feel fairly confident in saying is that it is not just going to depend on the technology though, it's also going to depend a lot on consumer choice. Consumers want, right? And we make these choices every day, right? Go to the supermarket, go to the self service checkout, or do I go to the human cashier, Do I do my yoga classes in front of my computer at home or do I go to a yoga studio, do I, you know, Want to sit in a giant vending machine and have my dinner or do I prefer to interact with the human waiter? And so all of these things are going to have a material impact on the future labor market. And I think there are many settings where we prefer to interact with humans, and there are probably quite a few settings where we feel that humans should ultimately have the responsibility for at least, you know, verifying AI outputs and signing off. And that will have an impact on the labor market too. And so even if we, you know, achieve superhuman performance of AI across the board, there will be jobs, but that may not be enough for many to make a living. And so that, that part I'm less certain of. I think right now and possibly for many years to come, though, AI struggles with novelty novel situations, things that haven't seen before. So most people know that AlphaGo beat the world champion and go back in 2016. 41. Few people know that humans using basic computers beat the best go programs two years ago by exposing them to positions that they would not have encountered in training and new concepts. And so what this tells you is that even in cases where AI has already achieved superhuman performance, cannot be really sure of how well it will perform when circumstances change and the state of the world changes all the time. Right? And so you may do very well using AI for inventory management, but then all of a sudden the pandemic hits and that is not working out very well any longer at all. And so I think that is a key bottleneck. That means that in a lot of jobs where the state of the world is changing around you one day to another, humans are more resilient to that and those jobs will be relatively safe from automation for some time to come.
- [47:47]
B
I buy that as well. And it certainly sounds like, you know, to use your model of kind of the innovation frontier versus the catch up of technologies that while that frontier may be developing some of this AI, the AI itself, if it hasn't seen these scenarios and its training data are going to be much less useful there than just kind of, you know, business as usual and improving efficiencies where, you know, the rules are set. There's, there's another avenue that I wanted to go down that you alluded to, which is the impact of consumer preference on all of this. And one of the trends we're seeing emerge is quite a strong anti AI backlash among consumers. It seems like especially you see it in different demographics. You see a lot of it with young people, you see some of it with older folks. I'm curious how that compares with your research on previous disruptive technologies throughout history. And what you make of that backlash, Is it productive and it's going to lead us to better outcomes? Is it just kind of this Luddite style foolishness that's going to be kind of drown in a tsunami of new technology? How do you classify it and where do you see it kind of evolving to?
- [49:12]
A
So I can't recall any time when the makers of a new technology have essentially sold a product by saying that the best case is that, you know, what we're producing, it's going to take your job, and in worst cases it's going to kill you. But, you know, please don't regulate us nonetheless. And so I think that part is basically unprecedented. But you're right that even without that, we've seen a lot of historical resistance to technologies, particularly that threaten people's jobs and skills. And a key reason that the Industrial revolution took a long time to materialize, because the steam engine was a late comer to the process. And beside that, really most of the machines and the factories were not really that complicated. Right. As the technologies could have been conceived much earlier. A key reason for that was that the guilds were quite strong and they were resisting the introduction of anything that threatened their jobs and skills. And so in Britain, you know, there was famously the lite riots. Those were unsuccessful, but on the continent and in China, those riots were actually much more successful. And in China, that delayed the industrial industrialization process by another two centuries because the guilds, the power of the local guilds. And so we've seen this repeat in history, and the 20th century is probably an exception in that regard, that we see much less resistance generally to technological change. Although, you know, we see some displacing technologies like automatic elevators or switchboard operators, automating being automated away. I think there was a general sense that people had improving outside options because there was a lot of new industries emerging and well paid jobs and in them that sort of cushioned the sort of transitional costs of, you know, losing your job. And I think with artificial intelligence, it obviously also affects workers that tend to have more political clout, a bit like the guilds, in a way. And so when robots began replacing people in the rust belt in the 80s, right, you would have a few angry op eds from those workers in either the New York Times or the Wall Street Journal. But the people being replaced in knowledge work, they are much more likely to mount more successful resistance to this. And the selling proposition of the technology companies is making that resistance a Lot easier as well.
- [52:42]
B
The historical context of the guilds is really interesting and I'm curious if you'll agree with this or if this was the point you wanted to make, that if you look at it through the lens of hindsight, it seems like where the guild's resistance was more successful actually did more long term societal and economic damage by preventing the adoption of these technologies versus the areas where there was a faster uptake that actually allowed them to kind of leapfrog in the technology and gain more economic clout. And that, you know, there's an implication, if I can connect some dots, that maybe the role of government, as you sort of alluded to, is actually just minimizing the shorter term pain caused by job loss when this automation comes into play. Do you. Is that fair or is that sort of against the sentiment you were trying to convey?
- [53:40]
A
No, I think it's fair. And so Britain during the Industrial Revolution had something called the poor laws. They were taxing themselves at 2% of GDP, which is obviously very little in modern perspective, but was a lot at the time to provide for the poor. And in places where the poor laws were more generous, you see less resistance to mechanization and you see a faster uptake of new technologies. And so I think that was already important during the air first industrial revolution. I think it's been true since as well.
- [54:13]
B
Well, and that's starting to feel pretty analogous to, you know, universal basic income, like the UBI conversation that's starting to emerge, you know, adjacent to all of this. Do you see that continuing to be, you know, a topic of conversation as, you know, we see more automation and, you know, is that, you know, a viable policy recommendation?
- [54:34]
A
I feel like that's a very American conversation. So in Europe we have the welfare state. I mean, you have it across the Atlantic too, but maybe to a lesser degree. And so the question is, okay, do you use the UBI to replace the welfare state? Well, if you do that, you're going to worsen inequality because welfare state targets people at the lower end of the income distribution. And so if you replace that with something that's universal, you're going to worsen inequality quality, because you're essentially transferring resources to the bottom, from the bottom to the top. Or if you sort of add a UBI on top of that, well, it's probably not going to be a very meaningful UBI because levels of taxation in Europe are already quite high. So I never quite understood the U in ubi. You know, Friedman, you know, the idea around the negative income tax A long time ago. I think that's, you know, a decent proposal. It means that, you know, it's essentially caps, puts a floor on how far your income can fall, and then you have, you know, incentives to top your up your income on top of that. You know, I think about AI a little bit like I think about natural resources in the sense that, you know, some places have oil. That makes those places directly. Well, there. Some places have been very good at sharing that wealth, like Norway. Some places have not been very good at sharing that wealth at all. And so I think your political institutions in the end of the day are the ones that are likely to determine what you end up doing. And so you can always, you know, come up with these ideas, whether it's UBI or welfare state, etc. But, you know, in the end of the day, sort of the question is, how do you get to Denmark or Norway? Which is sort of this classic development question. And so, you know, if AGI happens, I think countries will adjust very differently to it depending on the political institutions in place.
- [56:50]
B
So let's follow that thread about how we make sure that our societies are best equipped for this type of change. And I'll ask the general question, how should we be intervening here and should we be actively intervening on the one hand of. Let the invisible hand guide what's going on here in these economies versus active intervention. So just to kind of reframe that, to what degree is active intervention worthwhile and what types of policies would you actually recommend around AI and the labor disruptions?
- [57:26]
A
So I think generally speaking, if people have a little bit more time to look for an alternative job, they tend to find better matches. And so Danish flex security is often mentioned in this context. I tend to agree that that's a good model. You have genuine flexibility. You can hire and fire relatively easy, but you have genuine security in the sense that if you lose your job, you have, I think, something like two years where you have very generous payments and you have a chance to find something else. So if this turns out to be a purely transitional problem, then I think that's the way to go. If we end up really in a world where there is not enough work to go around, well, we can either choose to adjust as societies and recognize that before the Industrial Revolution, well, people worked, but they didn't really have jobs. And people can find meaning in family and gardening and other things too, or we can think of employment programs. If you feel that people are not able to find sufficient meaning and purpose in other things.
- [58:56]
B
So that you Know that that covers, you know, an interesting viewpoint specifically at the lens of the state or, you know, societally. Do you think, do you think for. Well, let's, let's maybe pivot over here for business leaders. If you're talking to leaders who are not, you know, in the magnificent seven, let's say how should they be thinking about this, this period of change? What should they be focused on and what are the right and wrong ways to kind of, you know, lead their organizations? And let's, you know, when I say business leaders, let's say it's, it could be public sector as well, public sector or commercial sector. What's the best way for them to lead their organizations through this kind of period of turbulence and change?
- [59:46]
A
So if we go back and look at the computer revolution for guidance, I think there's a general pattern that firms that more decentralized benefit the most. And I think the same thing applies to AI. And there have been a number of studies looking at corporate hierarchy and AI adoption. And more decentralized organizations tend to be more likely to adopt AI as well. And that makes sense because the people that are most likely to know something about the technology are the people that actually using it. That's the people at the lower levels of the organization trying to figure out new use cases and what to use the technology for. And if they don't have decision making rights, then obviously you're going to slow down adoption significantly because you need to get approval for essentially just using, using this new tool. And so I think as a general rule, more decentralized firms and organizations tend to do better during periods of quite rapid technological change. And so that I think is one lesson that applies to today as well.
- [61:15]
B
Can I push a little bit more on decentralization? How do you define that at the firm level? What does a decentralized firm look like? And maybe the opposite question, how does it differ from a centralized firm?
- [61:31]
A
So I think it differs a bit in two dimensions. And one is sort of just the number of layers in the managerial hierarchy. And then there's sort of a question of how much decision making authority do you have at the lower levels of the organization? And so how much can you just do something and implement a new technology, change your workflow, which without having to ask a superior or your superior superior for permission to do that. And so I think that's decision making autonomy is an important part of that. And then I think, you know, the reward systems matter too. Right. And so if I take time off experimenting with something new, I might be less. I will be less productive in other things unless I increase my, you know, hours to, you know, do the experimentation in my spare time, essentially. And so you want to reward that experimentation or at least not punish it. And I think that's an important part of it as well.
- [62:32]
B
Well, and has a parallel with something you said earlier about people complaining about waste in academia, that it and that if you're actually going to build something new and discover something, what waste is a byproduct of that entire process. Right. You'll never get to perfection there. And you have to have an appetite to have some level of waste as a trade off.
- [62:55]
A
I think that's absolutely right.
- [62:57]
B
Well, I really appreciate the conversation today. This has been really interesting. We've covered an awful lot of ground. So thank you so much for joining and for all your great insights.
- [63:06]
A
I much enjoyed it. Thank you for having me.
- [63:10]
B
If you work in it, Infotech Research Group is a name you need to know no matter what your needs are. Infotech has you covered. AI strategy, covered. Disaster recovery, covered vendor negotiation, covered. Infotech supports you with the best practice research and a team of analysts standing by ready to help you tackle your toughest challenges. Check it out at the link below and don't forget to like and subscribe.