Book Review: Power and Progress
Introduction
Power and Progress: Our Thousand-Year Struggle over Technology and Prosperity (2023), by Daron Acemoglu and Simon Johnson, is a book I was excited to read because it could address some problems I’m very interested in. Namely, what will be the economic, social, and political impacts of artificial intelligence on most people? Acemoglu and Johnson’s thesis is that technological progress does not guarantee shared prosperity. Rather, it is up to all of us, collectively, to shape the development of new technologies and their consequences, for good or ill. In this review I’ll explore the arguments presented in the book and reflect on them.
Premises
To engage with the authors’ thesis, we have to accept their premise that advances in artificial intelligence can be viewed as analogous to other advances like the advent of agriculture or the industrial revolution, rather than something far more profound than either of those and which is likely to have effects which are not at all analogous and much more dramatic.
My own views are different from the authors’ in that I think there is a reasonable chance that the deployment of highly capable systems will lead to all humans being killed. Nonetheless, I was excited to see what the authors would have to say about problems we could face if alignment does get solved (and several other steps.) Or if we see a long period of slow AI capabilities growth, perhaps due to regulation or plateaus in research, during which the economy could change a lot but we don’t reach existential outcomes.
Productivity
Throughout the book, the authors provide historical examples that demonstrate that technological progress can have net positive or net negative consequences for workers. After examining cases from medieval times, the industrial revolution in England, and the early Soviet Union, they conclude:
For anyone who believes that the productivity benefits necessarily trickle down through society and improve wages and working conditions, these formative episodes are hard to explain. But once you recognize that technology’s advances look after the interests of those who are powerful and whose vision guides its trajectory, everything makes a lot more sense.
Technological improvements to productivity do not necessarily lead to improving the lot of workers, for example, by increasing wages. The history cited demonstrates cases where productivity increases lead to an increased incentive for employers to squeeze workers harder, leading to worse outcomes for those workers. Not productivity, but rather marginal productivity is one key to improved outcomes for workers. If employers feel a pressure to hire more workers, this gives workers a stronger negotiating position.
This emphasis on marginal productivity will return throughout the book to reinforce the lens through which technological progress is viewed: are technologies positioned as complements to humans or as means of replacing them? The term machine usefulness captures this idea, and it is held in contrast with the dominant focus on artificial intelligence.
Artificial Intelligence vs. Machine Usefulness
The authors frame these two competing paradigms in the development of computer technology, and this dichotomy was new to me. The AI framing is familiar to all with its popularized roots in Alan Turing’s test of a machine’s ability to imitate a human. The authors are quite critical of the popular obsession with this agenda to replicate intelligence and “high-level capabilities”.
This vision of almost inexorable benefits from new technology, including intelligent machines, led by talented entrepreneurs is an illusion–the AI illusion. […] it receives a further boost because it enriches and empowers elites corralling technology toward automation and surveillance.
The other framing has its roots in a succession of early thinkers and inventors. Norbert Wiener was maybe the first to advocate for something like the idea of machine usefulness as a better target for our efforts than general intelligence. He also was concerned with the displacement and degradation of human labor as a result of automation:
“Let us remember that the automatic machine, whatever we think of any feelings it may or may not have, is the precise economic equivalent of slave labor. Any labor which competes with slave labor must accept the economic consequences of slave labor.”
The authors next offer J.C.R. Licklider, who in 1960 argued that:
“Relative to men, computing machines are very fast and very accurate, but they are constrained to perform only one or a few elementary operations at a time. Men are flexible, capable of ‘programming themselves contingently’ on the basis of newly received information.”
This does seem like quite the dated idea from our perspective in (early) 2024. We seem to be at the cusp of having AI agents who can do everything Licklider admires that men can do. Whether and how fast we reach this point is unclear, but to me this argument is undermined by the rapid progress we see today. Rather than concluding that it is folly to chase general intelligence, this framing only increases my concern that we’ll increasingly witness the displacement of human knowledge work by automated systems.
Lastly the authors cite the contributions of Douglas Engelbart, creator of the computer mouse and a host of other innovations including hypertext, bitmapped screens, and graphical user interfaces. Certainly these technologies contributed greatly to the broad usability of computer systems, and give us an intuition for how investment into machine usefulness can pay dividends by “expanding human skills”.
Bringing About Machine Usefulness
The authors describe four ways that machine usefulness can be developed. They are: (1) increase worker productivity in tasks they are already performing (2) develop new tasks for workers (3) process information, for example through recommendation systems for aiding decision making (4) create new platforms or markets. They show that the pursuit of machine usefulness leads to great rewards but has increasingly been sidelined in favor of artificial intelligence.
The response I would expect to hear is “but wait, these are AI assistants we’re developing over here.” So much of the hype around new AI tools is just that… AI tools for individual workers or creators. Certainly we do see both strands of thought at play in the applications being created. ChatGPT and Copilot seem like powerful tools for increasing worker productivity through information processing. One does worry about these as Trojan horses–short term strategies to collect data and insights which can be used to train worker replacements.
I think both approaches can be in flight. Tools for workers and also systems for employers looking to automate their workforces away as soon as they can. The authors describe the dominant culture in business today which seeks to maximize shareholder value and how automation goes hand in hand with that approach to management.
Democracy
I was not surprised and even a little frustrated in the chapter on how all this relates to our democracies. The authors invoke democracy, the people’s will, and diverse viewpoints as major (potential) drivers of solutions while at the same time lamenting how all that is being undermined by these same forces. So much of the latent potential in artificial intelligence favors autocracies and large tech corporations, who can exploit technologies which automate surveillance, control our attention, and make us economically dependent.
An aside: my frustration comes in part because the other major cause I am interested in alongside AI safety is American election reform. In brief, democracy is broken in America. Our representatives face corrupted incentives and are not held responsible for doing well by their constituencies. I’m excited about simple reforms to the system which could have a huge impact, like open congressional primaries. So yes, I know we could do a lot to address these problems if we had a more representative government that was more responsive to the needs of the people and not corporations. This section reinforced my belief in the importance of reform and my frustration with how slowly it seems to be going.
Solutions
At the end of the book we get a discussion of potential methods to “rechart the course of technology” which I think it is valuable to summarize here:
Worker Organization - That is, unions, to bond workers together and allow for collective negotiation
Civil-Society Action - By which it is meant broad community organizations which can pursue collective goals, such as Greenpeace
Market Incentives - Subsidizing socially beneficial technologies (like privacy tools), and regulating harmful ones (like surveillance). Also some novel frameworks like not enforcing patents for undesirable technology (!)
Breaking up big tech - They point out, and rightly so, that tech giants like Google, Facebook, and Amazon are all too similar to the giants we successfully broke up in the past like Standard Oil and AT&T, to the benefit of all. They do point out here (and in other places) that this is a complementary tool and not sufficient on its own.
Tax Reform - For coming so late in the book, I think this was a very powerful and reasonable insight “the US has taxed labor at an average rate of about 25 percent over the last four decades while imposing much lower taxes on equipment and software capital.” So remove this asymmetry and increase corporate taxes at the same time to rebalance the incentives.
Investing in Workers - Further balance the incentives here for not just hiring workers but for training them and improving their productivity.
Government Leadership to Redirect Technological Change - Broad-brush subsidizing or direct funding of the development of positive technologies.
Privacy Protection and Data Ownership - GDPR has not been successful, but the idea of giving users more control and ownership of their data is good. Could users receive income from the use of their data? Could they negotiate such by forming data unions?
Repeal Section 230 of the Communications Decency Act - This “protects internet platforms against legal action or regulation because of the content they host.” They imagine that Facebook and YouTube would be able to survive such, and that their platforms would host much less harmful content as a result. I think a proper evaluation of this is a bit beyond the scope of this review and even the book itself.
Digital Advertising Tax - They seek to directly encourage alternative business models to the advertising one, which I certainly support. I’m not sure I’m as convinced that advertising is the root of these companies' obsession with engagement and maximizing the user’s time on platform, coming from games where we shared that obsession but I’ve never had an ad-based model. Subscription or microtransaction-based models can be just as eager to maximize engagement in my experience.
Capitalism and Conclusion
My other frustration with the book was that it seems to avoid grabbing the bull by the horns: why are the solutions mainly drawn from a moderate economic approach of redirecting and counterbalancing the forces of exploitation, rather than removing them entirely? This seems to take for granted that we are in a codependent relationship with capital which cannot be fully revised. We see increasing the power of labor unions suggested but not a description of alternative models of ownership such as worker co-ops. I may need to read Nick Romeo’s The Alternative: How to Build a Just Economy (2024) next as I think it explores this question specifically.
In the end, I was hoping this book could make explicit its imagined approach to actually getting to work on solving these problems. Who is to implement the changes above? Is this book for elected representatives? For voters? For policy analysts at think tanks? If you work at Facebook, what opportunities do you have? Are individuals left to find civil-society organizations to focus their efforts? Is someone connecting the abstract ideas and suggestions in the book to change in the real world? Should I expect a book like this to lay that out?
These are questions I will keep asking as I continue to explore this topic, and I’ll take with me many of the ideas from Power and Progress even if I think it only constitutes a piece of the overall puzzle.