Transcript:
Rick Brink: Hello, everyone and welcome to this edition of Beyond Consensus. My name is Rick Brink, Market Strategist. I'm joined by Inigo Fraser Jenkins, Co-Head of Institutional Solutions. This is a blast for me to go one-on-one with Inigo. And the topic is a topic that you can find everywhere right now, the idea of AI. And there's really two main ways that AI is being approached in the investment world. One is something that actually I just did on our Disruptor Series—shameless plug by the way—and this is the idea of how AI gets adopted. The S-curves all the way through physical AI, the needs for energy, what it means as an investor and what gets produced. That's one side.
But the other side is centuries old, literally in the research, and this is the impact of technology of any type on labor over time. That's been explored ad nauseum, Inigo and I, and many others, will harken back to economic possibilities for our grandchildren in a 1930 paper by Keynes that Inigo references in here and has spoken about before. It goes back even further than that. Historically, the idea has been, there's sort of two takes on this. One is that technology leads to this quasi-utopian society where we don't need to work anymore. Or we work half weeks and everything is great and suddenly, the meaning of life. That's one side, which hasn't really worked out. And the second one is that AI, or sorry, that technology in general is the killer of jobs, which also hasn't really worked out because you do kill some jobs, but historically you creep at least as much and generally more. This second part is what Inigo got after in this most recent paper. This is what we're going to be talking about today. You can read the paper. We have the paper available to you 20 ways to Sunday, and I encourage you to read it. There are lots of data points, there are papers referenced, there's more nuance and detail, but there's no point for us to just get on and read the paper to you. The way we're going to break this down is into three sections.
First, we're just going to talk about the philosophy. When you go to tackle this issue, you have a lot of decisions to make. We're going to start there—how Inigo went about it, what he decided to focus on, how he tried to measure. Part two is going to be the findings; what came out of it. Part three, which I think Inigo and I are both thinking is going to be the most fun, we'll see what you think about it, we'll call it future speak. What if AI, because now it's a thinking form of technology, actually does become a killer of jobs? What does that mean and what are some of the questions, almost existential questions, that we're going to have to get inside of?
I'll just start with an obvious question here or a pretty straightforward question, Inigo, which is exactly what I teased, how did you approach it? There is limitless research on this. As you sift it through, how did you come up with the way you wanted to attack it?
Inigo Fraser Jenkins: Yeah, and thanks for the setup. The big questions that we're asking here is one, is the size of the overall productivity gain from AI, and the second is this interaction of both with demand for labor. These two things are obviously intimately linked as kind of questions. The first thing we have to note is we need to be really humble in our approaches to this because past attempts to forecast both the total size of the increase in productivity from new technologies, and past attempts to forecast how much that is a job destroyer versus a job creator have been really tough. And people have come up with huge revisions, for example, back in the tech boom of 1999 and 2000 had to undo them afterwards; so going into this with a kind of humble view. On this kind of question of the interaction of AI and labor, I guess the starting point for us is some of the academic work that's been done on what the exposure is of different tasks to AI—and there's been some quite good work done—of the individual tasks level across the economy, how exposed is this task to the new generation of LLMs?
Then from that, we can aggregate them up to I guess how many tasks or different types form different kind of professions and different sectors in the economy, and then make some assumptions really on two fronts. Firstly, make some assumptions in terms of what the size of a productivity gain could be, given some assumptions about adoption rates we have for a range of sectors and add that up across the contribution of the sectors to a profitability in the economy. The second route we can go down and say, well, given the exposure of this task and this profession to AI, if we make some assumptions about what proportion of any productivity gain comes from displacing labor or enhancing labor and then sum that up, but by what the share of employment is in that sector. And that gives us some sort of starting point, if you like, for a trade off or at least a way of thinking at least about what a potential productivity gain can be versus how much of that comes from labor destruction versus labor enhancement.
That's one angle of it. And then the other angle is this question of, what is the total productivity gain of AI? Certainly, we go into client meetings and the whole time we're asked, is this going to raise GDP by a huge amount? Is it a game changer? Is this going to affect the way people think about overall levels of net debt? Is it going to change the way people think about growth rates? The problem with that, as I said, is that past attempts to forecast productivity have been fraught with difficulties. Our first approach to that is to actually not even try and forecast that productivity. Instead ask the question, what are the other things happening at the same time that might depress those growth rates? Because AI, although it's huge and a dominant question that all of us are dealing with the whole time at the moment, there are other things happening at the same time. Most of those other things are downward forces on growth rates. And then ask the question, is it plausible that a productivity gain from AI could offset those downward forces of growth, either comparing that to history or comparing it to attempts to forecast productivity growth. That's the methodology and the setup that we've used in the paper,
Rick Brink: The two major themes of this other stuff. The way that I think of it is AI doesn't exist in a vacuum; it's being introduced into a world that already existed. And as you point out in the paper, and one thing that we've both talked quite a bit about, is demographics. A major element here is a much more forecastable, slower-moving animal that we have a much higher degree of confidence around. So that's a big one. And then, we sort of start to forget about climate a little bit, but there's clearly a fair amount of research. Now there are the bands around expectations are also quite wide, but demographics is a large one. This brings me just to tease for everybody the right side of the colon, as I always refer to some of our titles, which is, might it be a good thing if AI is job destroying?
Because if we do have demographic challenges and shrinking labor forces, isn't this just exactly the right time for the robots to show up on the scene and bail us all out? A last thing I'll toss out into the audience is that this isn't homogenous, right? There's a global way to break this down that you've done, but then there's a critical regional question as well. With that in mind—and by the way, let me just say so that it isn't lost here because the comment that I think you made was really important—it is incredibly difficult to forecast productivity. The range of ways that it's even been attempted is enormous. The error terms when it has been done in the past are awful. When I read it as a reader, I thought it was a pretty elegant approach to turn around and say, “well, instead of trying to forecast a level of productivity, let's look at these other forces and see what kind of productivity we'd have to solve for to stay where we're at, to improve, et cetera, et cetera.” And then we can get to the idea of what does it do to labor displacement, et cetera, et cetera. Why don't we talk about how you did it and then what you found?
Inigo Fraser Jenkins: Yeah, okay. I like the idea of pairing AI and demographics as topics. I think they're actually intimately related in a number of ways. First, as you said, these are two things which are huge forces on the drivers of macro growth and potentially inflation over the next 5, 10, 15 years. These two things are happening at the same time. But, the linkage goes even deeper than that, I think, because AI does raise these kinds of questions about, what is the future of labor demand in aggregate and how much labor is there? And that's obviously […] term by demographics. Also, the other aspect of demographics is the shift in the representation, different age cohorts in the population and the idea that if AI is changing not just the quantum of labor needed, but the type of tasks that are valuable, are those the tasks that are suitable for a population that's made up of a different age cohort? I think there are a whole bunch of levels on which these two topics are linked.
Again, the second thing I'll say is this intriguing question has been asked in a few client meetings as well, which people start to ponder the idea of, okay, maybe we have a smaller working- age population in the future and maybe we have a smaller need of labor. Is this the super convenient, as you said, is this a good time for the robots to show up? Push beyond that and say, one of the huge advantages that the US has, as we normally think about it, is a growing working-age population unlike other regions. But maybe that becomes a problem if they're too many excess workers and maybe Europe could benefit enormously if it hasn't got any excess workers in a world where fewer workers are needed.
I have to say I'm pretty skeptical of those kinds of arguments because it would be an enormous coincidence if these two completely separate things happen to balance each other out, both temporally and geographically. I don't think that argument holds, but then maybe we come back to the regional elements of this a bit later. At least pairing demographics and AI I think is an important part of, as we described, it's a more humble approach to try and forecast what AI can do. Because when we think about the 10-, 15-year forward prognosis of growth rate, one of the huge shifts that's apparent, is at very least the growth rate in the working-age population is going to be lower than it has been over the last 40 years or so. Now, one advantage that the US has is that the absolute level of that growth rate in working-age populations is probably slightly positive, whereas for Europe and Japan it's probably negative. But across all of the developed world, that growth rate in working- age population is going to be a smaller number. And we suspect that the combination of that plus any degradation to growth rates that comes from adverse climate outcomes, in the US at least, points towards something like a lessening of real growth rates per annum 10 years hence, by about one percentage point per annum. And for that to be a slightly worse number outside the US because the demographic outcomes are worse there.
The question is, if there's going to be a lessening of real growth rates over the next 10 years by 1% per annum,. AI has to make up for that in terms of its productivity growth. The core question is, is it plausible that any uplift in productivity growth from AI could be of the order of one percentage point per annum? Because that's a big ask. People should be aware that if you're talking about productivity growth rates less than that, well that's helpful, I'll take it, but it's not going to, in aggregate, lift the overall level of growth in the economy. And then there really are two ways to answer that. One way is to appeal to history. I think both you and I like long-run charts. In the note, we referenced this chart that starts in the 13th century and shows that for 600 years prior to the invention of the steam engine, the growth rate in GDP per capita in the UK was zero. And the steam engine comes along in the early 19th century and lifts that, lifts it by about 0.8% per annum. The historical analogy is that if AI is as good as the steam engine, then that implies that you can lift growth rates in a sustained way by a sort of magnitude.
Also intriguingly, I think it acts as a bit of a speed limit. If people come along and tell me there's going to be a growth rate from AI that's greater than 1% per annum, you're saying it's going to be better than the steam engine in a sustained way. Obviously, it's possible and people should be kind of aware that that is a big claim. And then the other way to approach it is to go through approaches that are similar to this sort of cottage industry of analysis and academia that's emerged in recent years and go back to that approach I mentioned earlier on in terms of methodology, which is try to think about what the exposure is of different professions to AI, think about what the adoption rate can look like. And on average you come up with the numbers of the order of 1% per annum in terms of what that productivity growth could look like. So again, it's plausible that you can think about offsetting that downward force on growth, but with a caveat that we have to be humble about these forecasts, there are huge error bars around it, and that's still making up for lost growth rather than giving us an extra boost of growth, but that we didn't have before.
Rick Brink: We're talking right now, your comments now are about in general, sort of labor, but I want to put a little more meat on the bone because one of the things that you did is you got into the idea—you sort of ran past this in one of your comments—but the difference between a job and a task. And being able to use the concept of a task. You go to your job and there are lots of things that you do during the day. Even right now as the rapid adoption of AI has taken place, you see AI from a form of productivity. People are saying, look, I can do this faster. I can ask it to write an email for me. It can summarize something; it can create a flow for, say, a discussion on AI that you and I might have for example, if I talk to it long enough. But one of the ways to break it down, that you broke it down is to then say, instead of the job and you're following in footsteps here, I want to make sure credits given where credit is due, but the idea is you're breaking the job down into tasks and using that to try to get a sense of jobs exposed to the level of exposure. Maybe talk about that.
Inigo Fraser Jenkins: Yeah, exactly. I think when you think about what a productivity gain looks like from the application of any new technology, it can come from a few different sources. It can come from displacing a unit of labor and automating a task that wasn't previously automated. Or it can come about through enhancing that unit of labor by giving that person better tools so they can focus on the higher achieving parts of their overall job, for example. Or it creates a new job that no one thought of before and that person leaves the previous employment and takes a new job where they are even more productive. You have this combination of some of this productivity shift comes about from displacing labor and some of it comes about from enhancing labor. As you referenced at the beginning, there's been a debate for at least 200 years about how much any new technology brings about an increase in productivity from one of these sources versus the other. The optimist would point out that in the past we had growing working -age populations and so we managed to create more jobs than we destroyed each time to keep the level of structural unemployment constant essentially in the long run, at least it hasn't risen over time. I guess that the first observation that we would make of why things look slightly different from that now is if you look at those tasks that are exposed to AI, and aggregate that up to sectors in the economy, it's quite apparent that the sectors that have the greatest exposure to AI are the least unionized sectors in the workforce today. This looks very different from the automation rounds of heavy industry and the car industry, et cetera, of the last 40 or 50 years, and at least the setup looks like a starting position whereby a greater power probably is accruing to capital rather than labor if the starting position is such a low level of unionization and labor bargaining power essentially. That's the one element of it.
Then we want to be a bit more formal, as I mentioned in terms of methodology, try and trade off this idea of how much productivity we can expect and how much that is requiring a destruction of labor in the near term. Now, these are very much numbers which are work in progress. We don't really have any good data yet in aggregate in terms of what the impact of AI adoption is on labor disruption. I mean arguably maybe in the latest earning season, we started to get a sense of that. But I think we need to see at the very least a few more quarters before we get any sort of sensible data. We made some assumptions terms of what adoption rates look like. Is the adoption of AI going to be as fast as the internet or much faster than that? I guess it looks like it be much faster than that. And then make some assumptions in terms of, is that extra unit of productivity coming entirely from labor displacement, entirely from enhancement or something in between? The bottom line from that approach is if one is demanding something of the order of 1% per annum productivity improvements, given a range of assumptions, it looks like you have to start to see something of the order of one percentage point extra losses of employment per year. Now, in the context of a recession, that's not a huge number, but certainly that's cumulative. That starts to be very socially disruptive.
Rick Brink: Alright, thanks very much, Inigo. I want to stay on this thread. You talked about adoption rates a moment ago. One of the things that is important to get here is this interplay between adoption rates and productivity displacement rates in a role. So let's talk about those three going together and as we start getting into takeaways, the other thing is that there's a range that you come away with. So you have a midpoint, but there's a low and a high. So let's talk about those three and when you put them together, what that produces along the spectrum.
Inigo Fraser Jenkins: Yeah, absolutely. There's a huge range of productivity forecasts being made around the role of AI and, how can it be that people can come up with such different numbers.? I think a lot of that comes back to this point you made of the interaction of assumptions of adoption rates, assumptions of, if there is the use of AI in a certain task, how much does that raise productivity by? and then the other element is, what is the exposure of a given task to AI? A lot of the difference between people's views on productivity gain of AI can come down to some combination of these and different assumptions about it. It's pretty clear, I think, that the adoption rates seem to be fast just in terms of the number of users of chat GPT for example, and just how that's grown very quickly. That bit doesn't seem too controversial, but that element in terms of what the exposure of an overall job is to AI and what the productivity gain is, and what the mix of the productivity gain is from displacement versus enhancement, is all open to discussion. That, I think, explains at least the range of forecasts people make. In our work we are really trying to think about what in a fast adoption scenario that means for this sort of interplay of job destruction versus enhancement on the one hand, and how much productivity you gain, but that you get on the other hand. These are very much numbers that we are using as sort of guesstimates and guides at the moment because we don't have good data. Our intention is very much to come back after a few quarters of data and then come back again after another few quarters of data and actually populate this as real data. The finding really is that to see a productivity gain of more than 1% per annum, even under fast adoption rates, it looks like you have to see a pretty significant amount of labor disruption in the near term.
Rick Brink: But the cross section of that distribution of where that shows up though is different; it's not everybody's job equally.
Inigo Fraser Jenkins: No, exactly. That could bring about this huge question of, who benefits? We have referenced quite a bit of the work of Daron Acemoglu in this where he's made the point that when past shifts in technology have come—the steam engine or all the inventions since then—there's often nothing inherent in the technology itself that determines what the cross-sectional distribution of the benefits looks like. Instead, it seems more likely that is determined by the social and political context that exists at the beginning of that time. For example, in the context of the steam engine, yes, the aggregate benefit was large and over time there wasn't a structural increase in unemployment if you look at it in a very coarse-grained way, if I can put it that way. But under the hood, there was clearly a huge disruption of certain sectors and the evidence that there was a multi-generational stagnation of real incomes for many people; a huge shift in that distribution.
You referenced […] Keynes at the beginning of the whole thing. The question of, why hasn't there been an ability to shift over working time versus leisure time over the period of huge shifts in automation? A big chunk of that, not all of it, but a big chunk of that comes down to how the benefits were spread out across society and the idea that they haven't been spread equally— actually far from it. That is some of the reason why there hasn't been a mapping of productivity gains onto seeing all cohorts benefit equally. I guess when we apply that onto the current state of AI, it raises some interesting questions. Because it looks like that combination of the work, I referenced before where we're looking at the least unionized sectors, the ones that have the highest exposure, plus the observation that corporates seem to be firmly in the driving seat of deciding what AI is developed, what AI is released, but that implies that the setup is, that at least in the near term, it looks like the more of this AI is going to be optimized for automation rather than necessarily enhancement, and perhaps something of a shift wider in terms of inequality in the near term.
Rick Brink: I've got two things that I want to toss out. One is because we have this section three coming up where we presume that AI is a little bit different and we're going to do our random musing along that low, medium and high range, talk about what high looks like. Those are not fun numbers.
Inigo Fraser Jenkins: No, they’re not. The good side of it is the aggregate productivity gain is north of 1% per annum. It’s not at 1.5% per annum, but that’s enough to offset the downward force on growth from a smaller working- age population and perhaps even offset the downward force on growth from climate. You see a lifting growth rate, a lift in EPS—that’s presumably going to be good for aggregate stock market returns, et cetera, et cetera. On the flip side of that, you are seeing something between 1% and 1.5% extra destruction of jobs per annum. Now, again, as I said before, in a business- cycle context, that might not be off the chart, but if there’s any sense that, at least in the medium term, that’s cumulative, then that becomes socially ugly pretty quickly.
Rick Brink: Right. Exactly. You start compounding that, you get into the teens, and that’s not nothing. So for that reason, the other side of this is the politics. So Acemoglu , again, all respect to the MIT crew on tech and labor over many, many years, but the verbiage that he uses, when he says it, I hear politics and policy and also the use of non-unionized labor, the thing that might be least protected in terms of being able to displace it. Those are some of the things that I extract from the verbiage. From a policy perspective, this is something that we want to sort of talk about, but from a jobs perspective, a labor perspective, there’s a lot of question marks that have to be answered over time. You mentioned adding stuff into the data as we go because we are seeing headlines, we’re starting to get a little bit of data.
A little bit of data is going to go a long way as we make our way through here. But to your point, I think the biggest takeaway here is, if it is everything that people speak to it being, and if it does indeed have legs and it does indeed usurp the steam engine, then we have some serious discussions around the nature of jobs and work and everything else, which leads us into our musings. But the other part of this is that it is not uniform by job, but it is also not uniform by country and region.
Inigo Fraser Jenkins: Indeed.
Rick Brink: The idea here is, when we talk about adoption rates and productivity and displacement, we’re sort of talking at a large scale and you’re referencing the United States—let’s take a little tour of the globe.
Inigo Fraser Jenkins: Yeah. I think there is a case that there are regional differences, which are further exacerbated by this. The first regional difference has got nothing to do with AI, is the reason why we’ve already written research on the idea that we are defending the exceptionalism of US equities, at least, and that comes down to a superior growth rate in the working- age population in the US compared to China and Japan and Europe, and also the sort of already existing greater level of profitability of US firms, for example. Now, you add AI into the mix, and I think there is a good argument to be made that US firms are probably better suited in the near term to make greater benefits from AI. And some of that simply comes down to the structure of the labor force in the US and the greater service component of the labor force, say, compared to China, for example.
Another part of it comes from the evidence— in the past, US firms have been able to extract benefits from advances in IT more quickly than firms in other countries. And the third element is the blunt fact that US firms could fire people faster and more easily than firms in many other regions, and that presumably is at least one route via the increase in productivity in the near term. So, however skeptical or optimistic you are about the aggregate productivity benefit from AI, I think it does add to the idea that from an equity perspective, at least, there is further defense of US exceptionalism and, for us at least, a strategic overweight on US equity still.
Rick Brink: I have a wild card question for you, which is, as you went through this process—because I always have these, when I do my stuff, there’s always something or things—what sort of surprised you? Or what were some of the, we always say the word “aha” or “aha moments,” but what were some of the things that kind of jumped out at you as you went through this that maybe you didn’t expect?
Inigo Fraser Jenkins: Some of these issues around these difficulties of trying to come up with a model for the difference between a labor enhancement versus labor destruction and just the lack of any consensus at all on that. And I see the same thing, really, since writing this and going around speaking to clients around the world, that I really don’t see any strong consensus around either what the productivity gain is overall or where this comes from. I mean, a lot of people end up probably on average looking quite glum about this in client meetings in terms of the labor impact, but there’s such an enormous difference of views. I mean, I find that actually really interesting.
Rick Brink: Yeah, it’s almost like AI has to run the gauntlet. First, we have questions about spending, we have questions about adoption. The S-curve has got to become pervasive, otherwise we’re building a bridge to nowhere, data centers to power nothing, whatever, all that stuff— it has to make that, it has to become pervasive, it has to become used across industries. Then it’s going to have an impact on the way we do our jobs, then it’s going to potentially have an impact on the number of jobs. It might enhance this. In financial services for example, maybe we just all get a lot better at analyzing the data, and we still have an arms race to try to grab that last incremental piece of information, but it doesn’t displace anyone. But, if the AI can do it without any of us, then we are absolutely displacing us.
And you make a reference to this idea that we might get to a place where AI allows us to solve deeper questions. We just don’t understand the answer. There’s a whole gauntlet that we’re going to run, but ultimately as data comes in, we’re going to get a sense of this. I’m going to try to gently segue into our third part, then, is our future speak. In terms of what you’re going to be working on that follows this, and you mentioned obviously wanting to put data in, but what’s in the hopper to follow on to this work, and what are you going to be keeping an eye on?
Inigo Fraser Jenkins: Yeah, so there are a few different angles here. There’s one angle is to try and get more hard data on how AI is actually adopted and what the evidence is of productivity gains in certain sectors and what that means for labor. And especially what the optimist would be looking for would be any signs of increasing demand for labor, which obviously has happened with previous technology advances in the past. That’s gone one area. The second area is the inevitable issue around AI capex spend and the link between whatever the productivity gain is and how much that can justify the huge spending on capex that is taking place. That’s the second clear avenue. But then there’s the other areas too, which get a bit further away from the core of finance and economics research, which are broader social and political questions.
And this idea of, as you potentially see a big impact on the labor market, what might the response to that be?
Obviously, we talked already about the politics of this. Is there a need to respond to this via regulation or a political change in some way? Does this become an electoral issue in certain countries around the world over time if there is an impact on labor markets? And also, I think that there are interesting questions around to what extent any displacement labor could be addressed through regulation or economic incentives versus the bits that really are hard to address. And the idea that in the modern economy and with modern expectations that the idea that work gives meaning to people, I think, is really important. If there is something that comes along that either changes working practices in a way that is negative or takes labor away in a sustained way, what does that mean socially, if that is something that’s so core for people’s lives?
Rick Brink: Which takes us directly into random musings, and I appreciate that. That was a nice easy move into this. There may be a touch of data in what we’re about to say, but mostly it’s going to be random musings. We’ve talked about the steam engine on this—Inigo put it in the paper. You can argue that the champion of technological productivity in terms of the magnitude and duration of the steam engine, which is why it’s something you really have to go up against if you’re going to try to be in the productivity hall of fame. But there’s recently, foreign affairs—there was a reference to some new work, but an old essay that I had missed, another pair that you have, Brynjolfsson actually in one of the papers here, I think this referenced, Brynjolfsson and McAfee, and this was a 2015 essay that I had missed, and they bring up another historical powerhouse as it relates to being able to produce.
There’s a finer point when we say that humans historically when technology has come along have not been displaced. There’ve been jobs destroyed, but at least as many, and in some studies more, jobs created as a result. But if I change the definition to living beings, then I can bring up the example of the horse. And at one time there was a massive amount of horses in use in the world, particularly in the United States with some interesting data around it. And then when the steam engine came along, they were displaced, but as the essay suggests, they didn’t really complain about it. There was no political move to save the horses’ jobs, all of this stuff. I find it, I don’t know if I find it a nice homage or insulting that we actually reference the output of things like steam engines in the form of horsepower.
We do have a historical example of living beings displaced, and it would’ve been pretty catastrophic if horses had to bring home and pay for their shelter and everything else. We do have some history there. Let’s say that humans now go that way. And you talked about the meaning of work, and this is really interesting because Owen, Keynes—a lot of these guys have talked about the idea that because we don’t have to work that now we can have this different meaning in life and we can explore and we can do all of this stuff and apparently become masters of art and music and philosophy and all of that stuff. We haven’t gotten to that place. And yet, in most studies, when they ask people to describe themselves or to define themselves, work is always at the top of the list—what I do.
There are real questions around what it means if there isn’t this thing that I spend approximately one third of my adult life doing. With that as a setup—alright, here’s the question you and I run into all the time, if AI replaces lots and lots of jobs, then who’s going to actually buy the stuff that AI Produces? AI automates an entire factory, there are no people working in the factory. AI is not going to buy what’s coming off the line. Let’s talk about— how do we even think about consumption in that world? Random musing question number one.
Inigo Fraser Jenkins: Yeah, that’s a great question. And there’s a thing that was brought up by Henry Ford a lot when he was trying to think about the affordability of his cars, et cetera, et cetera. This kind of question has come up many, many times. Yeah, I think when we get into conversations with investors around this, pretty quickly, someone in the room will mention a universal basic income as the answer because if one does choose to take that as your starting position, that labor is destroyed in a way that hasn’t happened with previous technologies—and, of course, we don’t know if that’s the case yet—but if that was the thing that happened, then is the response that the government has to start giving people money, which of course is not so far -fetched because it’s what we saw in COVID essentially. We have been here before, and it has been contested.
The governments can hand out money to people.
Of course, that immediately raises a few big questions, but number one is how is this funded? As we’ve referenced in separate podcasts and separate research that both of us have done, we are at a level of public debt to GDP, not just in the US but for the G7 overall, that previously has only been seen in times of existential conflict such as World War II and the Napoleonic Wars. And we’re back there at the same level of debt already, but we haven’t necessarily invested that level of funds in a productive technology, at least for the evidence hasn’t happened so far. We’ve done that more to keep consumers’ standard of living happy. So, there’s a big question mark of how on earth you’re going to fund that if the starting level of debt is so high. Presumably, the answer has to be, well, if some corporates are making large amounts of money from this, then there has to be some tax mechanism for reclaiming that. But, yeah, it’s a very different model of consumption. Certainly that’s not clearly on the cards in any kind of big country at the moment, but it’s an interesting question of how quickly any fears about labor disruption might at least get on the list of political concerns and then raise some questions in terms of what a response might look like. I have to say that I think the governments around the world are probably on the back foot on this question, and there’s a lot of catching up to do.
Rick Brink: Again, it always seems that we don’t really care about something until we decide to care about something, and it’s usually in the 14th hour when it comes around and we can look away and look away, and then finally we decide to pay attention to it. UBI is—I’ll throw my random musing in on UBI because I often joke about T-shirts that I should get when I’m doing things like this. and I should have one that says, “all roads lead to UBI.” Because anytime you have an AI discussion and you start talking about jobs being taken away, it invariably leads to, what do you do if you have mass displacement? You have to provide some level of income. It’s not a choice; it’s a forced displacement. So now we come up with UBI as you point out, how are we going to pay for it?
We’re going to tax folks. And the question of, who do we tax comes in, and you and I have already talked about this offline, so I will point out that corporations become the most likely because they, in effect, are sort of quasi- funding their own consumption, if you will, through the increased taxes. And it’s kind of hard to tax the folks on UBI, but then you might tax the high earners, that might go up again. So, you start musing about where can the tax dollars come from? And then the second musing I have here is that, practically speaking—and again this is just chit chat for a glass of wine—but it’s the idea of, if you are to provide UBI and it is for a significant portion of the population, but there is going to be a large portion of the population that continues to work—so, think of a surgeon for example, and they have to keep showing up. They’re on call. I’m thinking of a friend of mine that is the head of a surgical unit and the way his schedule sort of works. The question is, what would the level of compensation need to be for UBI so someone could have a life and live and meet all needs? What would that level be, and how much would there have to be a gap, let’s say, for someone that’s continuing to work 40 to 50 hours a week? What would be the incentive or disincentive for someone on the left side of the fence to say, you know what? I think I would like to just go to the right side of the fence and stay over here. I think my other random musing, if it comes to this, is how do you solve for the level of UBI, how you fund UBI and what keeps a fully engaged workforce?
Now again, if we define ourselves by the tasks that we take on and the jobs that we do, maybe that in and of its own right leads to some level of whether it’s because you’re trying to serve the greater good, whether it’s because that’s what you want to do with your day, it’s a passion, whatever it is, I think that’s one of the other bigger ones. And then, let’s come back around to policy and AI. Legislating jobs, union versus non-union labor laws, makeup of labor in a country. Now we’ve seen little tidbits of that— every time I think of this, I think of the West Coast port workers a little while back when they had labor negotiations coming up and they were trying to embed some level of protection from automation. What do you think? Just random musing now. Now we’re becoming politicians. Random musing beyond what you already said.
Inigo Fraser Jenkins: I think that this has to follow evidence where job dislocation is happening and how fast it is, and which sectors its happening in. And whether any job dislocation that happens is materially different from past rounds, as I said, in terms of whether this affects perhaps higher- earning professions than perhaps it did in the past or maybe not. But the makeup of that starts to look very important. I guess that the other element of this, which reflects on what you were talking about earlier really, is that it seems plausible that through this transition—so through the capital deepening that’s taking place with the spending on capex and through the starting position of corporates being in the driving seat—it looks like capital versus labor share increases. Now, I had this sort of forecast a long time ago that the profit share of GDP in the US would have to fall at some point—it would have to mean revert because surely there was some social limits to that—but I’ve given up on that view because there’s been no evidence of any kind of limit being reached, and I think AI can push that even further. But at some point there are presumably arsenal limits around that. So that’s one thing I’d say. The second is, I do hear this pushback on UBI, aside from the fiscal pushback, which is how on earth does one pay for it?. Just a more philosophical pushback was that the idea is so depressing. It’s just kind of giving up on the idea of labor and the meaning that’s attached to it. Is it plausible that a society could go down a route? We obviously don’t really have the modern parallels of that.
And the final thing I’d say is that I think a lot of the focus around UBI comes from people who are frightened about job dislocation.
Of course, there is another reading, which most famously was set out by Keynes, as you referenced few times before, which is he made the claim back in 1930 that by 2030, i.e., by now, that we’d all be working 15- hour weeks because human labor wouldn’t be as needed. Actually, he was prefigured by a whole century in the work of Robert Owen, in the wake of the steam engine and its advance, he’s making a claim that surely human labor won’t be needed because of scientific advances. And he was thinking about how wonderful that is in terms of the ability to cut both adult and child labor, because obviously both of those were issues back then. And you’ve seen this echoed much more recently with work on the more socialist/Marxist end of the spectrum, particularly in Europe where there’s a left accelerationist movement is making this bold claim that we should end the need for work and demand more automation as a route to a happy society. Now, of course, you haven’t seen any of this take place. I think, again, it comes back to one of the points I made earlier on and […] some reasons that it’s really not technology, but it’s policy decisions. There’s policy choices about what the benefits of any growth increase look like in terms of how they’re spread across society.
Rick Brink: Thank you. And by the way, as a final echo onto that, one thing I should have mentioned as you were making your comments is that part of the UBI discussion is also attaching some level of work to it. Anything, if we do have an aging society, is there a home healthcare, can you care for a family member? As you’ve mentioned in the past when we’ve talked, GDP doesn’t capture a lot of the things we do in our own home that is productive. The idea of home healthcare for your own family or for others or other types of work—I think of Roosevelt’s Works program post-–[Great] Depression and what that was used towards. It’s not all doom and gloom. And again, this is the random musing section, not the findings of Inigo’s research section. So, we can all take a breath.
First and foremost, thank you very much Inigo for doing this.
It is always such a blast to get on here. It’s been probably three years since you and I did a one-on-one back when we did Disruptors, so we have to try to make this a little more frequent. because I really enjoy it. And listen, everyone, thank you so much for the support of the series and for joining this one as well. We hope you enjoyed it. Again, I encourage you to read the paper. There’s a link, you can get to it, you should. There’s a lot of great detail there. But hopefully you enjoyed this as a complement to get to the why and what Inigo and the team were thinking as they put it together, some of the observations, takeaways, “ahas” for all of us and some of the cross chatter that maybe didn’t make it into the paper. Please download that as well. Thank you for joining this episode. We have another one right on the horizon, so please join that as well, and I look forward to seeing you there. Thanks, everyone.
