Redistributed

Book: Daron Acemoglu and Simon Johnson. Power and Progress: Our Thousand-Year Struggle Over Technology and Prosperity. PublicAffairs, 2023. Link: Publisher

Why This Book Matters

In October 2024, Daron Acemoglu won the Nobel Prize in Economics — in part for the body of work that Power and Progress distills into a single, sweeping argument. The book's thesis is deceptively simple: technology does not automatically make everyone better off. It never has. Whether the gains from innovation are broadly shared or narrowly hoarded depends on who holds power and how they wield it.

This should be obvious. It is not. The assumption that technological progress lifts all boats — what Acemoglu and Johnson call the "productivity bandwagon" — remains the default setting of policy debate, Silicon Valley pitch decks, and op-ed pages. Power and Progress dismantles that assumption with a thousand years of evidence.

For anyone tracking how AI redistributes wealth, work, and power, this is the foundational text. Not because it has all the answers, but because it asks the right questions with historical rigor that most AI commentary lacks entirely.

The Argument

The Productivity Bandwagon Is a Myth

The book opens by attacking the most persistent assumption in economics: that productivity gains from new technology eventually reach everyone. Acemoglu and Johnson call this the "productivity bandwagon" and argue it is historically false.

The mechanism people assume works like this: new technology raises productivity, which raises output, which raises wages, which raises living standards. Each step seems logical. The problem is that it keeps not happening — at least not automatically.

The English agricultural revolution of the 18th century increased food production dramatically. Landowners prospered. Agricultural workers were displaced, impoverished, and herded into factory towns where conditions were abysmal. The Industrial Revolution made Britain the wealthiest nation on Earth while its workers' real wages stagnated or declined for nearly sixty years. As Acemoglu and Johnson put it, the bandwagon "went off the road."

History Repeats, But Not Inevitably

The book moves through a series of historical episodes — medieval mills, the printing press, the factory system, electrification, railroads — and in each case finds the same pattern. New technology concentrates wealth and power in the hands of those who control it. Broad-based prosperity follows only when three conditions are met:

Workers organize to demand a share of the gains.
Governments intervene to set rules, standards, and redistributive policies.
Social norms shift to make exploitation politically unsustainable.

None of these happen automatically. Each required political struggle, often lasting decades. The eight-hour workday, workplace safety standards, universal education, progressive taxation — these were not natural consequences of industrialization. They were hard-won concessions extracted from people who had every incentive to resist them.

The AI Parallel

This is where the book pivots to the present. Acemoglu and Johnson argue that current AI development is following the historically familiar pattern: concentrating gains at the top while displacing workers at the bottom.

But they go further. The problem isn't just distribution — it's direction. Current AI development is oriented toward worker replacement rather than worker augmentation. This isn't an inevitable feature of the technology. It's a choice, driven by specific incentives:

Tax policy that subsidizes capital investment over labor
Market concentration that allows dominant firms to automate without competitive pressure to improve quality
Tech culture that prizes "beating the human" as a benchmark
Shareholder pressure that rewards headcount reduction over productivity improvement

The result is what Acemoglu calls "so-so automation" — technology that replaces workers without meaningfully improving productivity. Self-checkout kiosks are slower and more error-prone than cashiers, but cheaper for the store. Automated customer service is worse than human agents, but costs a fraction. AI-generated content fills the page, but rarely matches what a skilled writer produces.

So-so automation is profitable for individual firms. It is destructive for workers. And it contributes almost nothing to the productivity growth that is supposed to make everyone better off.

What Would "Good" AI Look Like?

Acemoglu and Johnson don't argue against AI. They argue for a different kind of AI — one designed to make workers more productive rather than redundant.

Their examples are suggestive: diagnostic AI that helps doctors catch diseases they would have missed. Translation tools that let specialists communicate across languages. Scheduling systems that improve logistics without eliminating dispatchers. In each case, the technology complements human judgment rather than substituting for it. The AI Coding Assistants space offers a live test case: developers using LLMs report genuine productivity gains while retaining professional judgment, at least for now.

The distinction matters because labor-complementing technology increases the value of human work. When AI makes a doctor more effective, the doctor's skills become more valuable, not less. When AI replaces the doctor with a chatbot, the skills become worthless — and the chatbot, inevitably, is worse.

What It Gets Right

Historical depth that most AI commentary lacks. The book's greatest strength is its refusal to treat AI as unprecedented. Every claim about technology and prosperity can be checked against the historical record. Acemoglu and Johnson check it — relentlessly.

The "productivity bandwagon" framing is devastatingly effective. Naming the assumption makes it visible. Once you see the bandwagon fallacy, you notice it everywhere: in policy papers that assume automation will "create new jobs," in corporate announcements that promise technology will "enhance the customer experience," in economic models that treat productivity growth as automatically welfare-improving.

Structural incentives, not just outcomes. The book doesn't just say "technology concentrates wealth." It shows why — and identifies specific policy levers (tax code, labor standards, antitrust, R&D funding) that shape which kind of technology gets built. This moves the conversation from lamentation to action.

The Nobel committee agreed. The Swedish Academy cited Acemoglu's work on institutions and prosperity when awarding the 2024 prize. This isn't an argument from authority — the historical evidence stands on its own — but it does indicate the scholarly weight behind these claims.

Critiques and Limitations

The policy optimism may be misplaced. Acemoglu and Johnson's prescriptions — tax reform, antitrust enforcement, labor standards, redirected R&D — are sensible on paper. They also require a functioning democratic politics capable of overriding concentrated corporate interests. In an era of political polarization, regulatory capture, and industry-funded think tanks, the path from "good idea" to "enacted policy" is far longer than the book acknowledges.

The binary framing is too clean. "Good automation" (labor-complementing) versus "bad automation" (labor-replacing) is analytically useful but overly neat. In practice, most AI applications do both simultaneously. A coding assistant augments experienced developers and threatens entry-level ones. Generative AI helps designers iterate faster and eliminates junior design roles. The same tool can complement and replace, depending on who uses it and how firms respond.

Speed and scope may break the historical analogies. The Industrial Revolution took decades to reshape labor markets, during which workers organized and governments adapted. AI capabilities are evolving on a timeline of months. The political response times that allowed historical course corrections may not be available this time. Acemoglu and Johnson gesture at this risk but don't fully reckon with it.

Global dynamics are underexplored. The analysis is primarily Western, and primarily American. Automation incentives differ sharply across countries with different labor costs, regulatory environments, and political systems. Policy enacted in the US may simply shift AI-driven displacement to countries with less bargaining power — a redistribution problem the book does not adequately address. The Sovereign AI Movement and Europe's AI Paradox illustrate how different regions are attempting to chart their own courses.

The Redistribution Question

This is, fundamentally, the redistribution book for the AI age. Its contribution is threefold:

It historicizes redistribution. Every major technology has redistributed wealth, work, and power. The question was never whether redistribution happens, but in which direction and who decides. Framing AI as another chapter in this ongoing story removes the false novelty that paralyzes policy thinking.

It identifies the mechanisms. Redistribution from labor to capital through so-so automation isn't mysterious. It operates through tax incentives, market concentration, and research priorities. These are levers. They can be pulled differently.

It insists on agency. The most important argument in the book is that the current trajectory is a choice. As Virginia Eubanks has documented in the welfare system and Ursula Franklin articulated in her concept of prescriptive technologies, systems that appear neutral embed decisions about who benefits and who bears the cost. Acemoglu and Johnson extend this insight to the full sweep of economic history: technology serves whoever controls it, unless countervailing forces intervene.

The redistribution frame also reveals what the book leaves out. It has relatively little to say about the redistribution of data — the raw material of AI — or about the global labor that produces training data and moderates AI outputs. Mary Gray and Siddharth Suri's Ghost Work documents these hidden workers; Kate Crawford's Atlas of AI maps the material supply chains. Power and Progress provides the economic framework, but the full picture requires these complementary lenses.

How to Use This Book

When someone says "AI will create new jobs": point them to the sixty years of Industrial Revolution wage stagnation. New jobs eventually appeared — after political struggle reshaped the terms.

When evaluating AI deployment: ask whether it complements or replaces human capability. If it replaces workers without significant productivity gains, it's so-so automation — profitable for the firm, destructive for the workforce.

When thinking about policy: the book identifies four levers — tax incentives, labor standards, antitrust enforcement, and R&D direction. Any serious proposal should address at least one.

When countering technological determinism: the historical record demolishes the idea that technology's effects are predetermined. Outcomes depend on institutions, power, and political struggle. They always have.

Sources

Acemoglu, Daron, and Simon Johnson. Power and Progress: Our Thousand-Year Struggle Over Technology and Prosperity. PublicAffairs, 2023. Publisher
Acemoglu, Daron. "Harms of AI." Cambridge Journal of Economics, Volume 45, Issue 4, July 2021. Cambridge
The Nobel Prize. "The Prize in Economic Sciences 2024." Nobel Prize, October 2024. NobelPrize.org
Gray, Mary L., and Siddharth Suri. Ghost Work: How to Stop Silicon Valley from Building a New Global Underclass. Houghton Mifflin Harcourt, 2019. Publisher
Crawford, Kate. Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence. Yale University Press, 2021. Publisher

General

Book Review: Power and Progress — Our Thousand-Year Struggle Over Technology and Prosperity