Power and Progress and the Politics of Technological Choice
Daron Acemoglu and Simon Johnson's Power and Progress is a direct challenge to the idea that technology automatically improves ordinary life. Its central claim is political: machines distribute benefits according to the institutions, bargaining power, design choices, and stories that surround them. That makes it one of the more useful AI-era books for thinking about automation without surrendering to either panic or inevitability.
The Book
Power and Progress: Our Thousand-Year Struggle over Technology and Prosperity was published by PublicAffairs in 2023. The publisher lists the book at 560 pages, with Daron Acemoglu and Simon Johnson as contributors. Both authors are MIT economists, and both shared the 2024 Nobel Prize in economics with James A. Robinson for work on institutions and prosperity.
The book is broad historical political economy. It moves through medieval agriculture, industrialization, railways, factories, modern computing, social media, surveillance, and artificial intelligence. The argument is not that technology is bad. It is that the benefits of technology are never self-distributing. Powerful actors shape what gets built, who gets automated, who gains status, who gains political voice, and who is told to wait for progress to arrive later.
That frame matters because AI discourse often treats deployment as if it were weather. A model appears; labor markets must adapt. A platform changes; users must accept the interface. A company builds surveillance or substitution into a workflow; the result is framed as modernization. Acemoglu and Johnson ask readers to slow the sentence down. Who chose this path? Who benefits from this definition of efficiency? What other path was available?
Technology as Choice
The strongest idea in Power and Progress is that innovation has a direction. There is no single line from invention to public good. The same technical capacity can be aimed at worker augmentation, labor replacement, surveillance, manipulation, monopoly, public infrastructure, or shared capability.
This is why the book keeps returning to power. A technology can raise aggregate output while weakening many people. It can create impressive products while degrading work. It can generate new surplus while concentrating voice among owners, managers, and platform operators. The political problem is not whether the machine is clever. It is whether people affected by the machine have any real influence over what it is for.
For AI, that distinction is essential. The same foundation-model stack can help nurses document care, help teachers adapt lessons, help technicians diagnose equipment, and help disabled people navigate institutions. It can also intensify workplace monitoring, deskill junior roles, replace support staff with brittle chatbots, flood public life with synthetic persuasion, and centralize practical judgment inside vendors.
Labor, Voice, and Countervailing Power
The book's labor argument is not nostalgic. Acemoglu and Johnson do not claim that existing jobs should be frozen in place. Their point is that broad prosperity has historically required workers and publics to gain enough power to claim part of the value created by new tools. Without that counterweight, productivity can become an extraction story.
The MIT News account of the book emphasizes this point through the authors' contrast between replacement-oriented automation and machine usefulness. MIT's summary reports their concern that many algorithms are being designed to replace humans as much as possible, while the authors argue for machines that make people more capable rather than merely cheaper to exclude.
The later NBER paper Learning from Ricardo and Thompson extends the same argument into early industrial labor history and AI. Acemoglu and Johnson write there that automation can raise wages when it creates new tasks that increase labor productivity or when complementary sectors hire enough workers, but workers are unlikely to share in productivity growth when they lack power. That is the hinge of the review: AI labor politics cannot be reduced to whether productivity goes up. The question is who gets the institutional leverage to turn productivity into livelihood, autonomy, and skill formation.
The AI-Age Reading
Read in 2026, the book is most useful as an antidote to inevitability. It refuses the story that AI must follow the path selected by the largest model companies, cloud platforms, advertisers, defense customers, and venture-backed automation vendors. Those actors matter, but they are not nature.
The practical AI question is whether systems are built to extend human agency or compress it. A tool that gives a worker better information, broader reach, safer work, or stronger bargaining capacity is politically different from a tool that turns the same worker into a monitored appendage. A public-service model that makes benefit rules easier to understand is different from a model that silently triages applicants through opaque risk categories. A classroom assistant that expands feedback is different from a system that replaces apprenticeship with shallow output.
The book also clarifies why surveillance belongs inside any serious AI labor analysis. Data collection is not only a training input. It is a management relation. When platforms and employers instrument work at high resolution, they create a loop in which workers become measurable, comparable, nudged, predicted, and disciplined. The model may be sold as intelligence, but the institutional effect can be a tighter form of command.
Institutions Decide the Shape of Intelligence
Acemoglu's later NBER paper Institutions, Technology and Prosperity makes explicit what the book implies: institutions, market structures, norms, and ideologies influence which technological possibilities a society pursues and how gains are distributed. That is the bridge between the authors' long-run institutional economics and the immediate politics of AI deployment.
This is where Power and Progress belongs beside books on legibility, classification, platform power, moderation labor, and surveillance capitalism. AI systems do not enter a blank world. They enter procurement rules, antitrust law, workplace hierarchy, professional licensing, union weakness or strength, public-sector austerity, data rights, advertising incentives, and institutional hunger for measurable control.
The review's concrete lesson is that governance has to begin before harm is visible in a dashboard. Who funds the research path? Who defines productivity? Who gets consulted before deployment? Who can appeal automated decisions? Who owns the operational data? Who is paid to verify outputs? Who trains the next generation when entry-level work is automated? These questions decide whether machine capability becomes public capacity or private leverage.
Where the Book Needs Friction
Power and Progress is ambitious, and that ambition creates stress. It covers a thousand years, many sectors, and several kinds of technology. Readers looking for detailed casework on specific AI systems, model architecture, data labor, content moderation, or global supply chains will need other books beside it.
Some reviewers also challenge the book's confidence in its historical and policy synthesis. A review in International Affairs argues that the authors understate variables such as national security, international networks, culture, ideology, and leadership. A review in Industrial and Labor Relations Review treats the book as a major labor-and-technology intervention while locating it in an active debate about institutions, policy, and work.
That criticism is useful. The book is strongest as a steering argument, not as a complete operating manual. It tells readers that technological direction is contestable. It does not settle every question about how to build durable democratic capacity in markets where compute, data, distribution, and capital are already concentrated.
The Site Reading
For this site, Power and Progress is a book about refusing the machine's alibi. When an institution says the model made us do it, the book asks who chose the model, who bought it, who set the objective, who benefits from the savings, and who loses the practical knowledge that used to live in people.
The recurring theme is not anti-technology. It is anti-fatalism. Interfaces, datasets, agents, workflow tools, and automated decisions are social arrangements with technical parts. They make some futures easier to imagine and others harder to fund. They turn certain people into experts, users, subjects, customers, risks, costs, or noise.
The book's best AI-era contribution is its insistence on institutional imagination. A society can ask for pro-worker tools, public-interest infrastructure, appealable automation, data rights, antitrust pressure, procurement discipline, and research paths that augment human skill. Or it can accept a thinner future in which intelligence means replacing judgment, surveillance means management, and progress means the powerful got there first.
Sources
- PublicAffairs / Hachette Book Group, Power and Progress: Our Thousand-Year Struggle over Technology and Prosperity.
- MIT News, "An AI challenge only humans can solve", May 17, 2023.
- MIT Stone Center on Inequality and Shaping the Future of Work, Power and Progress project page.
- Daron Acemoglu and Simon Johnson, National Bureau of Economic Research, "Learning from Ricardo and Thompson: Machinery and Labor in the Early Industrial Revolution, and in the Age of AI", Working Paper 32416, May 2024.
- Daron Acemoglu, National Bureau of Economic Research, "Institutions, Technology and Prosperity", Working Paper 33442, February 2025.
- Mark Zachary Taylor, International Affairs, review of Power and Progress, November 6, 2023.
- Stephen J. Frenkel, Industrial and Labor Relations Review, review of Power and Progress, 2025.
Book links are paid affiliate links. As an Amazon Associate I earn from qualifying purchases.