I’m a great admirer of economist Daron Acemoglu, having read several of his books, papers, and articles, and listened to many of his public lectures, and I find that he is one of the clearest thinkers on the road ahead for capitalist free market that retains control by a liberal democracy.
He is one of the most influential and oft-cited economists of the past several decades and the Nobel Committee cemented that status with his Prize in Economics in 2024. He’s a guy worth listening to.
He has written a new book What Happened to Liberal Democracy? which traces how the deal of shared prosperty has started to erode during the data revolution of the past few decades. There are mixed reviews, and a lot of criticism that Acemoglu doesn’t ‘get’ AI from the industry, but I think much of that is just panic that he gets the implications for our economy and society a little too well.
Acemoglu sees the neccessity of reigning in the power of tech companies to do whatever they please, the need to reorient our tax structure to incentivize the use of human labor over capital investment in automation, and the need to carefully craft IP protections for humans’ ‘data exhaust’, the high quality data we produce just by existing and working online. We need to make protections for that IP strong and easily licensable and monetizable by those humans producing it. The implications for data privacy are also equally striking.
Acemoglu argues the project of maintining liberal democratic societies under the new pressures on the democratic capitalist deal of poltical equality and shared prosperity that welds the working class to liberal democratic politics is under increasing pressure, which will only be exascerbated if we allow mass economic movement toward AI models replacing humans in the workforce. If that integration of AI into the economy goes wrong (i.e. continue as is, without checks and rules), and displaces a significant percentage of the workforce (say, 10-15%) then the wheels are going to come off our current social structure. We can see that happening right now.
He had a wide ranging discussion on these topics with Yasha Mounk, which requires a subscription to Persuasion to hear in full. I decided that the end of the conversation was prescriptive enough to serve as a good summary and launching point, so I am presenting that text here for you for critical review and academic purposes (even though I’m a mere baccalaureate with a minor in economics, and thus am but an ant crawling over the edifice that is Acemoglu’s ouevre). I treat it as a launching point for some of my own idea and conclusions, and it is a good description of how the Democratic Party will have to craft and new economic policy framework in coming years if we hope to keep the liberal democracy our ancestors built. It’s a long read, admittedly, but not nearly as much as the full conversation, and considering the massive stakes involved. All bold emphases and bold italic commentary is mine.
For the TLDR crowd, a few take-aways from this dialog and my reaction to it, but you will have WAY more insight and foundation for these conclusions if you RTFT (may it spur you to read further):
- AI threatens a basic breakdown of the liberal democratic political settlement between capital and labor of the last century;
- Massive joblessness if AI development aims merely at replacing human labor will create societal chaos;
- To avoid that result, we must democratically take the reins and create market and state incentives to develop AI to complement and enhance human labor, including reversing our favorable treatment of capital gains and capital investment and instead valorize and favor the return on human labor;
- people call data the new oil powering our economy; they are more correct than they know and we must expropriate that human generated resource from the imperial tech giants who think they own it and return it to the humans who create it with strong property and privacy laws.
Begin Commentary-Enhanced Transcript:
Mounk: So what is working-class liberalism?
Acemoglu: I am hoping to contribute to the articulation of a new governing philosophy and a coalition around it. Working-class liberalism, I think, is the name that appeals to me, because it is going to remain true to liberal values, but prioritize issues that are critical for the working classes. That means jobs, shared prosperity, and pulling back from social engineering, because working classes have very diverse values. Look at black communities: they want many policies that enable economic mobility, but they do not want some of the cultural aspects that are anti-religious. On the other hand, there are going to be other communities that are very progressive in their religious values, communities that have certain ethnic sensibilities or conservative values. We want to allow that richness, but still bring everybody together with a respect for local community, local governance, local self-governance, and some shared moral values and communication, so that the nation can steady itself as a nation. That is a recipe for forming a broad coalition, and, of course, that involves pulling back from social engineering and finding more common ground. Shared prosperity and jobs are going to be a very important part of that common ground.
Then the first one that you mentioned—none of this, in my mind, is feasible if AI goes in a job-destruction direction, which I associate with AGI, artificial general intelligence, because that is the apex of automation in my mind. AGI, as articulated by industry insiders, does not have a very clear definition. There are several definitions around, but essentially it means AI models reaching capabilities comparable to the very best humans, let us say the top one percent of experts, in pretty much every field. If that comes, it will inevitably be associated with massive job displacement, because if AI models are better than humans in pretty much everything, sure, there will remain some human employment—I do not expect all jobs to disappear—but there will be significant displacement. Robots and manufacturing equipment caused the displacement of something like five percent of the tasks that human workers were performing in the United States in the 1980s, ’90s, and 2000s, roughly speaking. If AI does twice as much of that, that is already huge. But with AGI, even if we do not reach AGI, if we come close to it, that could be ten percent, twenty percent, even more. Those kinds of displacements would make a mockery of our aspirations to build shared prosperity, to create meaningful, dignified, well-paying, stable jobs for humans. Without that, we are not going to be able to recreate the support for liberal democracy, and without the support for liberal democracy, nothing would stand.
Ed. MDB: I don’t know that I believe ,as Acemoglu does, that ‘nothing would stand’ if 10-20% of jobs were automated away, because I believe the democratic energy to replace employment income generally for the displaced with Basic Income (it won’t be Universal, so I don’t use the term UBI) or another means (discussed later) of providing income support would be overwhelming. If, however, nothing were done – as seems possible if oligarchism continues to dominate our political economy – then, truly, we would have a complete meltdown, against which, nothing could stand. That way lies chaos.
Acemoglu continues: That is why it is so important that we benefit from AI, but that we redirect it in what I call a pro-worker direction, a human-complementary direction, so that we get the productivity benefits, we get the technological advances, we get science and knowledge improved, but at the same time we deploy the skills and competencies of many different types of human workers.
Mounk: One of the things that strikes me about the debate regarding artificial intelligence and employment is that a lot of the time there are two very extreme poles. Either people say—and there are quotes about this from Elon Musk, and slightly less extreme quotes from Dario Amodei and so on—that all the jobs are going to go, or half of the entry-level jobs are going to be gone in three years, which would go together with a social cataclysm, but I think can sometimes be somewhat naive about the real-world obstacles to adoption of technology. We had Arvind Narayanan on the podcast recently to talk about some of those. On the other hand, some economists say that, because of job reinstatement and Jevons’s paradox and other kinds of things, we are sure that there is always going to be full employment. I think there is a very realistic scenario in which we fall well between those two poles, where we have, certainly temporarily, perhaps permanently, a significant share of the population that cannot find a job. Let it just be fifteen or twenty percent of people who currently have a job. If you look back at the Great Depression, or the Great Recession, or other moments of economic contraction, we were talking about way fewer people, for not very long, and that had absolutely disastrous political consequences. It seems to me that this debate is ill-calibrated in not taking seriously how disastrous the impact of job displacement from artificial intelligence would be, in between the poles of nobody has a job tomorrow and everything is hunky-dory.
Acemoglu: One hundred percent, you got it exactly right. The only thing I would add is that it could be disastrous even with five percent job displacement if it is also associated with the remaining jobs being meaningless, low-paying, and unstable. So it is not just the number of jobs, but also the quality of jobs. In the 1980s, ’90s, and 2000s, there was an increase in joblessness—that is what our earlier discussion of employment-to-population ratio was about. But the bigger effect was that people who were in middle-class jobs paying, in today’s dollars, something like forty or forty-five dollars an hour, went down to precarious jobs paying fifteen or twenty dollars an hour. That was also disastrous. It is really the creation of jobs, and the creation of well-paying, meaningful jobs, that is so important, and none of that is guaranteed. I am flabbergasted that some economists, as well as industry leaders, somehow think there is a law that every technological change will ultimately restore full employment. I do not understand where that law comes from. There is no such law. Technologies do many things. Some technologies automate work and create inequality, and jobs may disappear. Some other technologies create jobs. It depends, of course, on the social safety net. If there is no social safety net, perhaps people will have to go and work for four dollars an hour, but that is not what we are talking about here, that is not what we want. If there is a social safety net, we are definitely not going to be able to pay people meaningful jobs that will put them above it. So it is a very complex situation, and any kind of complacency of the sort you hear from some economists and many technology leaders is really irresponsible, I think.
Ed. MDB: This is where the idea of Basic Income falls down, IMO. People may survive short term with a minimal support system, but people don’t want to survive, they want to live well and aspire to more. They want their own home, kids, nice car, vacations, and savings to cushion the unexpected. None of that is contemplated in Basic Income, and so it might be a useful stabilizer for the economy in the short term, it is NOT a long term solution for long-term loss of employment and the meaning and dignity and ambition that comes with it. THAT is what we will need to replace in society if the maximal dreams of AI and robot boosters come to pass replacing much human employment.
Mounk: Let me connect two things you said, one at the beginning of the conversation and one now. At the beginning, you said that the way we built the middle class historically was in part that, when the Industrial Revolution took off and there was much more demand for human labor, human labor was scarce, and so employers had to compete with each other for workers.
Acemoglu: Yes, after the first phase of the Industrial Revolution, by the way. The early phase of the Industrial Revolution did not really make human labor scarce. It created jobs for five-year-olds to go down the mineshafts, but not meaningful jobs.
Mounk: Right, we had the Enclosures and so on. Now, if you have, to pick a number out of a hat, twenty percent of the population that lose their jobs and do not have new jobs, I think often we imagine, well, that twenty percent is going to be unlucky, and we somehow have to have mechanisms of redistribution to make sure we do not fail them, but the rest are going to be fine. But of course, that also means there are now twenty percent of people sitting on the sidelines, very happy to take any job that becomes available, which would undermine the bargaining position of those people who still do have jobs.
Ed. MDB: Yes, there will be a constant oversupply of human labor if the unemployed are not satisfied with their little survival Basic Income payments – and that’s the point. That’s why people like Musk embrace BI. They know they will be able to mistreat and underpay the remaining workforce with the threat of having to survive on BI.
Acemoglu: One hundred percent. That is why, yes, we did somehow create some jobs in the midst of the displacement from robotic and software systems, reorganizations of workplaces, and globalization, but those were not well-paying jobs, and the bargaining power was not there. People went into jobs in sectors such as fast food and other in-person services that were not paying very well. None of that is guaranteed. You create jobs with the right technology, the right effort, and investment. There is nothing automatic about it.
Mounk: Some people might fear that, whatever this technology turns out to be, that is how the world is going to turn out. You have written, in this book and in other publications, that we can shape this future to some extent, and that is part of the working-class liberalism you are talking about.
Acemoglu: One hundred percent, I think that is a very important part of it. In some sense, what we just talked about says there is no inevitabilism when it comes to full employment. Full employment is not an inevitable thing that will happen to us, and there is no inevitabilism when it comes to what technology will do. Every technology is a menu, and there are many different things on that menu. With AI, the menu is tremendously large; AI is such a heterogeneous bag of things that we have just given a common name to. Most importantly, as we have already talked about, one can use AI, and one is using AI, although slowly, to automate work. There are three and a half million customer service representative workers in the United States, roughly speaking, and I do not see what is going to happen to them but to lose their jobs, ultimately, with AI taking over more and more customer service jobs. It might actually do a bad job, as we have seen in many instances. But there are many other jobs in which we can imagine—and even in customer service, we can imagine—AI being a complement to humans. That is the redirection I am talking about. We need much better electricians dealing with much more complex problems; AI can be a guide to that, can enable them, can help with their training, and, more importantly, help them in real time deal with much more complex problems. AI can make nurses more productive and let them take on more tasks.
Right now we are using AI very doggedly to sideline teachers, from Khan Academy, which tries to put teaching in the hands of students and parents, to automated teaching and automated grading, ways of lecturing from one person being broadcast to others. Those are all ways of reducing some of the teaching functions and taking them away from teachers. But we can also use AI to actually realize the aspiration people have had for over a hundred years of much more personalized, individually targeted education, because we can have much better information about what each student is having difficulty with, and real-time adjustment for that, using AI tools and skilled teachers who know how to use them and are really experts in what they do, which is the communication with students and the inspiration of students, which is such an important part of learning. In every occupation I can give you examples of this.
Mounk: How much of this is about choices that the frontier AI labs should be making in terms of how they design this technology? How much of it is choices that schools, for example, in the case of education, should make about how they do or do not integrate AI technology to reimagine education? And how much of it is tools for economists to make sure, for example, that we shift the tax basis away from very significantly taxing labor in a way that discourages human work, to create incentives against automation?
Acemoglu: The latter one is about policy, and I will come back to that in a second. But when it comes to how we should design the technology, or how we should use the technology, the answer is both. You cannot use a technology in a way that is completely opposed to how it is designed, and you cannot do anything with a technology if people refuse to use it in a particular way. That is why, when I have a chance to talk to business leaders, I try to convey the message that one very powerful way of achieving greater productivity is to recognize that it is their human resources that are so critical for innovation and productivity, and to give them better tools. That is a mindset very different from some business leaders’ emphasis on cut cost, cut cost, cut labor costs. That is a very important part of it. But they cannot do that if the only thing the tech sector produces is automation tools. That is not true, because large language models can be used in many different ways. But I also do not think that the architecture of large language models, and the industry’s focus on developing bigger and more capable large language models, is consistent with this pro-worker AI direction. We need more domain-specific models. You cannot really put any of the large language models in the hands of field technicians or electricians to deal with complex problems, and you cannot rely on them to let nurses take over diagnosis, cure, and prescription decisions. So we need more domain-specific models. We need to develop them in a more reliable way, in a way that can be understood and that works better with human workers. Both of those need to be done, and that is where policy comes in.
I strongly reject two propositions. One is that somehow bureaucrats or government officials can tell technology companies what to do. That is not possible; innovation will come from entrepreneurs, they need to do that. I also reject the simplistic view that somehow we can perfectly guide the technology in other ways. It is a messy process. But we can set a framework for it, and the ultimate purpose of the framework should be to change the focus of the industry. The way I believe—and I do not have evidence for this—is that if tomorrow forty or fifty percent of the tremendously talented engineers and innovators in Silicon Valley, or the tech sector more broadly, said, we are no longer interested in AGI or superintelligence, and what we want to do is create human-complementary tools, that is exactly what we would get.
So the question is, how do we convince them? That is where policy comes in, that is where discourse comes in, that is where the media comes in. You made a comment about the media—I think that has been a real problem. From the very top to the very bottom of the media landscape, for so long, the only thing you could read about AI was either that it was going to be the most amazing thing and we are all so lucky to have the talents of Sam Altman and Elon Musk, or that AI and killer robots were going to end civilization. The useful stuff, of what AI will do and how we can actually make that better, was never part of the conversation. That is what we need to make part of the conversation. We need to make sure that the inspiration for AI engineers is not the science fiction of artificial superintelligence, but understanding and helping humanity.
Ed. MDB: I agree that one cannot usefully guide innovation with bureacratic dicats, but I do believe that we can shape the free market decision space with intelligent incentives and standards, democratically arrived at. If we bend the tax code and the system of scientific and academic incentives, granting, and industrial policy toward positive technologies that increase human productivity, judgment and decision-making, rather than replace it, then we can have a positive effect on the development of AI technology.
Mounk: That is what really strikes me about the debate about AI, and what my next book tries to address: the debate is either at the level of robots are going to kill us, and you are in the world of sci-fi, or it is at the level of looking at water use and data centers and so on. There are obviously a lot of social scientists working on this, but there has really not been a lot of debate in the broader public about what this is going to do to our basic economy, our democratic institutions, our society, at the mid-level. That is the aspiration I have for the book. (Ed. MDB: I look forward to reading that…) To return to your book, we have dealt quite well with the AI part of it. More broadly, what is working-class liberalism going to look like as an economic program, and what is the answer to how we talk about liberalism liberated from the contractarian tradition? You have a few minutes to make your case, and after that, if there are open questions, people will have to go and buy your book.
Acemoglu: I do not have a perfect roadmap of what economic policy under working-class liberalism should be, and some of it is going to evolve over time. But a focus on jobs requires that we provide incentives for job creation, in a way that generates wage growth, and that technological investments do not go in the automation direction unnecessarily. One aspect of it, for example, is correcting the wrongs of our current tax system, where we heavily tax labor via payroll taxes and other obligations on employers, such as fringe benefits and income taxes, while we more or less subsidize capital. If an employer has two pre-tax cost-equivalent ways of performing some task, if they go for labor, they have to pay twenty-five percent plus in taxes; if they go for capital, they pay less than five percent, close to zero percent in some instances. That is a massive subsidy towards automation that makes shared prosperity much harder.
Ed. MDB: Deeply agree! We are incentivizing innovation that prioritizes deployment of massive amounts of capital in our economy, for AI and other purposes. That works in many domains, but it is poison when the free market is incentivized generally to deploy capital merely to replace human labor. We need to rebalance and make some conditions on that incentive. It will take time, experiment, failures, and therefore political courage to get that done. But we can also restore a balance between how we tax capital returns and labor; right now capital returns are vastly more advantaged. We need to do the EXACT OPPOSITE and massively advantage earning by labor, not just for the laborer, but for the companies paying and employing labor.
Acemoglu continues: We also need to set AI on a better path. Part of that is to create an AI agency that provides state-of-the-art, best-practice knowledge and information, just like the National Institutes of Health sets best practices, but also has the power to suggest regulations and the resources to fund certain underfunded areas. We also need new legal infrastructure. Everybody I know in this area agrees that data is going to be the lifeblood of the economy, most likely more important than land. Can you imagine a world in which anybody who wanted to could go and take your plot of land away from you? That would be unthinkable in a market economy, in a functional economy. But that is how we treat data. These companies, more powerful than any organization you can imagine, can expropriate all of your data without paying for it. That is not just inequitable, it completely kills the incentive to invest in data, to create the high-quality data that is going to be so important, especially for the domain-specific models that will help human expertise. So we need new laws and new ways of structuring the economy. We also need other help for workers: training, worker organizations, worker voice. After all, workers are the ones who know how work is organized, they have the tacit knowledge. We need to protect that, and we need to make sure that we use it in the right way. So I think there is a lot to be done.
Ed. MDB: A basic digital privacy and property law that declares clearly that you OWN your data, no matter where it is located, is deeply needed for regular people to survive the future economy (finally a great use for blockchain and distributed ledgers, rather than mere enabling crypto con games!), and whatever its source, your so-called ‘data exhaust’, which is actually incredibly useful, and therefore valuable, data. The AI frontier models are running out of such human-generated data quickly. Future progress in all of AI development will be limited by access to this high-quality data. I suppose this could become the basis of a new economic foundation for a lot of people and foundation of the wealth required for a UBI of sorts; it could be based on the royalty value of human generated data. It might be one big pool of sovreign wealth, equitably distributed, or a micro-payment system that incentivizes those who are excellent at generating useful data. When people say data is the new oil of the market economy, they could be more right than they knew. The problem to be overcome is the the single resource political curse, which will need to be avoided. Up to now, the big tech companies acted like the imperial resource companies of the past (and somewhat in the present…), who would exploit a nation’s resources and return only a pittance to the people; they, of course, now exploit the whole world’s data. We need to expropriate from the tech giants that scarce resource, which they currently view as their own, and privitize it with strong property and privacy rights for the people creating that resource. The big imperial data exploiters are going to fight back to keep their ill-gotten goods, and that is part of the political fight of our generation.
End Transcript.
I hope you made to the end! Congratulations, you are smarter than the average bear, Booboo!
Discover more from Blog for Arizona
Subscribe to get the latest posts sent to your email.