## *Power & Progress*: our discussion ### 1. The central argument Acemoglu and Johnson argue that **technological progress is not automatically socially beneficial**. What matters is who controls the technology, how it is deployed, and how the resulting gains are distributed. Technology can increase productivity while simultaneously weakening workers' bargaining position or concentrating economic power. Broadly shared prosperity therefore depends on institutions and political choices, rather than following automatically from technological progress. We found this historical argument substantially more interesting than their policy prescriptions. --- ### 2. Their historical evidence may actually be darker than their conclusion This became one of our main observations. A&J's examples do **not** seem to show a general historical tendency toward broad sharing of technological gains. Rather, broad sharing of what they call **.rents.** appears to be unusual. The post-WWII period is the striking example: technological progress coincided with strong labor institutions, rising wages, mass education, social insurance, and relatively broad distribution of productivity gains. But if that is essentially an **outlier**, then the historical lesson is not: > Technology . struggle . institutions adapt . broad sharing. It may instead be: > Technology . concentrated gains . political struggle . **occasionally**, under unusual institutional circumstances, broad sharing. That distinction is crucial. The postwar settlement demonstrates that broad sharing is **possible**, not that it is the normal historical destination of technological progress. And that led us to the particularly gloomy thought: > **If history is our guide, broad sharing of AI's gains may be unlikely rather than inevitable.** --- ### 3. The .royal we. This became our biggest problem with the book. A&J move from: **historical observation** to **historical interpretation** to **policy prescription** and eventually to the royal **.we..** They tell us what *we* should do: encourage technologies that complement workers, strengthen labor, redirect incentives, regulate harmful applications, distribute gains more broadly, etc. The individual proposals are generally **plausible**. We didn't think they were nonsense. The difficulty is the missing institutional bridge: > **Who exactly is the .we,. and how does this .we. acquire enough power to make the proposed changes?** That is particularly difficult with AI because the relevant actors include corporations, workers, governments, competing nations, investors, researchers, and consumers.and their interests aren't aligned. --- ### 4. Perhaps A&J should have written a more pessimistic manifesto We eventually came up with a more charitable interpretation of what they *could* have been trying to do. Instead of implying: > .History shows that we can steer technological development.. their argument could have been: > **.We are in trouble. The historical odds are against broad sharing. But we have managed something like this once before. Here are some things we could try..** That is a much stronger manifesto. The postwar period then doesn't serve as a **historical prediction** of what normally happens. It serves as an **existence proof**: > It happened once. Therefore it is not impossible. That's a considerably more modest.and more defensible.claim. --- ### 5. History as prediction versus history as inspiration This became perhaps our most interesting distinction. A&J are **economists, not historians**, and they use historical episodes to identify recurring economic mechanisms. If we take that seriously and treat their historical material as something approaching a predictive model, then the book becomes rather sad. The model might predict: > technological change . concentrated gains . accumulation of power by technological leaders . political struggle . only occasionally broad distribution. Their programmatic conclusion, however, says: > **We can intervene and produce the favorable outcome.** The historical analysis may therefore be **better at predicting the future than their proposed mechanism is at changing it.** We both liked that formulation. --- ### 6. The .sad ending. problem You made the wonderfully perceptive observation that **people don't like books with sad endings.and perhaps this applies to histories of humanity as well.** A&J's preferred historical narrative has a satisfying arc: > technological disruption . conflict . institutional learning . broad prosperity. But history may not actually have that narrative structure. Perhaps the postwar settlement wasn't humanity **learning a lesson**. Perhaps it was simply an unusually favorable constellation of political, economic, and institutional circumstances. If so, we may be making a common narrative mistake: > **mistaking an exceptional historical outcome for the destination toward which history was moving.** And history, unlike a novel, does not owe us character development. It doesn't have to learn. --- ### 7. AI makes the problem especially acute We then brought the argument into the present. AI is particularly difficult to .steer. because: * frontier development is highly concentrated; * firms compete intensely; * countries compete strategically; * the technology is evolving extremely rapidly; * applications are difficult to predict; * capital, talent, and technology are internationally mobile; * and the economic and military stakes are enormous. So even if everybody sincerely believes that some degree of restraint would be desirable, each actor has incentives **not to be the one who restrains itself**. That gives us a collective-action problem: > **Everyone can want restraint while everyone has an incentive to defect from restraint.** --- ### 8. And then came our old friends Hassabis and Amodei This is where *The Infinity Machine* entered the discussion. Amodei, Hassabis, Altman, and others began advocating some form of slowing/pacing frontier AI and greater coordination on safety. We found the situation fascinating because the people calling for restraint are **the companies at the frontier**. That creates a genuine double possibility: 1. They may sincerely believe AI is moving too fast and that safety institutions need to catch up. 2. The resulting standards and coordination could nevertheless benefit incumbent firms and make competition more difficult for newcomers. Those propositions are not mutually exclusive. And that brought us straight back to *The Infinity Machine* and the possibility of **regulatory capture**. The question becomes: > **Who writes the rules?** If the frontier companies help define the standards, select the evaluators, and determine what constitutes acceptable development, then .AI safety. can potentially become a mechanism for incumbents to shape the competitive environment in their own favor.even without anyone acting cynically. --- ### 9. Our ultimate A&J position By the end, neither of us was simply saying **.A&J are wrong..** Our more nuanced conclusion was something like: > **A&J identify a real and historically important problem. Their historical analysis is compelling enough to give us reasons for concern about AI. Their policy proposals are individually plausible. But the historical evidence does not establish that broad sharing is the normal outcome of technological progress, and their argument does not fully explain how the royal .we. acquires the institutional power necessary to reproduce the exceptional postwar outcome.** Or, in shorter Marlowe form: > **They have a pretty good theory of how technological progress can get away from us, followed by a rather optimistic theory of how .we. can take the wheel back.** And perhaps the most interesting irony is that **their own history gives us reasons to doubt the second part.** --- ### And your final assessment You ultimately found it an **interesting book**, and our discussion made it considerably more fun. I think that's exactly right. You didn't end up rejecting A&J. You ended up **separating the book's historical diagnosis from its political program**.and that made both more interesting. The historical diagnosis: **.Watch out. Technology and power have a complicated relationship..** The program: **.Here is what we might do about it..** The first is something history can give us evidence about. The second is where politics, institutions, incentives, and the rather troublesome royal **we** enter the picture. And that's probably where we left *Power & Progress*: **not convinced that A&J have shown us how to steer the future, but considerably more interested in their warning about what happens if nobody can.**