Power and Progress

In the AI Story

The book's structural argument is that every major technology admits multiple deployment paths. The British industrial revolution could have produced broadly shared prosperity in its first century; it did not, because mill owners chose deployment architectures that concentrated gains and because workers lacked the political power to redirect them. The Luddite response is reread not as irrational resistance to progress but as accurate diagnosis of who was capturing the gains — and strategic failure in the absence of institutional alternatives.

The authors develop the concept of the productivity bandwagon to name the ideological claim that automatically links technical progress to wage growth. The bandwagon is shown to require specific institutional conditions — unionization, progressive taxation, public investment, democratic countervailing power — that were absent before the mid-twentieth century and have been progressively dismantled since. The postwar prosperity that seemed to demonstrate the bandwagon was in fact produced by the post-war social compact operating on the technology.

Applied to AI, the book argues that the current trajectory resembles the installation phase of earlier technological revolutions more than it resembles their deployment phases. Productivity gains are real; worker gains are absent. Labor's share of income continues to fall. The concentration of AI capability in a small number of firms controlling foundation models replicates the ownership structure of nineteenth-century industry rather than the distributed structure of mid-twentieth-century manufacturing.

The book's prescriptive argument — and the most contested part — is that the direction of AI development can be shaped by policy. Redirecting research subsidies toward machine usefulness rather than machine autonomy, reforming tax codes that currently favor capital investment over labor employment, and breaking up AI concentrations through antitrust enforcement are presented as technically feasible interventions whose obstacles are political rather than technological.

Origin

Johnson and Acemoglu began the project in the late 2010s as they watched the AI wave accelerate without corresponding institutional response. The book's approach of threading a thousand years of economic history through a single distributional question was modeled on Robert Allen's work but extended to the full sweep from medieval to digital. It appeared in May 2023 and received the Financial Times Business Book of the Year long list recognition.

Key Ideas

The productivity bandwagon is an empirical contingency, not a law. Shared prosperity requires institutional conditions that can be built or destroyed.

Technology choices are political choices. What gets automated, who owns the results, and how gains are distributed are settled by power, not by technical necessity.

Luddites were often right about distribution. The canonical story of irrational resistance to progress obscures the accurate diagnosis the displaced workers made about who captured the gains.

AI requires democratic intervention. Without redirection — through antitrust, labor organization, tax reform, and public investment — the AI transition will replicate the first century of the industrial revolution rather than the mid-twentieth.

Debates & Critiques

Tyler Cowen and other growth-focused economists argued the book understates the eventual welfare gains of the industrial revolution and the self-correcting capacity of labor markets. Acemoglu's defenders noted that 'eventual' in the British case meant roughly a century, during which life expectancy fell for manufacturing workers and child labor became normalized — a counterfactual no democratic society would now accept.