Redistributed

AI will make us all more productive. We've heard this promise before — about computers in the 1980s, about the internet in the 1990s, about smartphones in the 2000s. Each time, the productivity gains were real. Each time, most workers saw little of the benefit.

This isn't a puzzle. It's redistribution.

The Promise

Every AI company makes the same pitch: their tools will boost your productivity. Write emails faster. Generate code in seconds. Analyze documents in minutes instead of hours. McKinsey estimates that generative AI could add $2.6 to $4.4 trillion annually to the global economy (McKinsey, 2023).

The implied bargain: technology makes workers more productive; more productive workers are more valuable; more valuable workers earn more. Economists call this the marginal productivity theory of wages — you're paid what you contribute.

If this theory held, the past 40 years would have been a golden age for workers. Productivity has grown steadily. Computers, the internet, mobile technology, cloud computing — each wave made workers measurably more efficient.

The golden age never arrived.

The Evidence

Since 1979, US productivity has grown 64.6%. Hourly compensation for non-supervisory workers has grown 17.3% (Economic Policy Institute, 2024). The gap between what workers produce and what they receive has widened into a chasm.

This isn't a recent development. The divergence began in the late 1970s and accelerated through each technology wave. More output per hour. Same wages. Somewhere, the extra value went — just not to the people generating it.

Where did it go? Corporate profits are at record highs as a share of GDP. The labor share of income — the portion of economic output going to workers rather than capital — has declined from 65% in 1980 to under 60% today (Bureau of Labor Statistics, 2023). Stock buybacks have returned trillions to shareholders. Executive compensation has grown 1,460% since 1978, while typical worker pay has grown 18% (Economic Policy Institute, 2022).

The productivity gains were real. The redistribution was equally real.

The Original Paradox

In 1987, economist Robert Solow quipped: "You can see the computer age everywhere but in the productivity statistics." This became known as the Solow Paradox — despite massive investment in computers, measured productivity growth remained sluggish.

Eventually, productivity did accelerate — briefly. From 1995 to 2004, US productivity grew at 2.8% annually, matching the postwar boom. Economists credited information technology. The paradox seemed resolved.

Then productivity growth collapsed again. Since 2004, it has averaged about 1.4% annually (Bureau of Labor Statistics, 2024). Two decades of smartphones, cloud computing, machine learning, and now generative AI — and productivity growth remains historically weak.

This creates a puzzle only if you assume productivity gains must appear in the aggregate statistics. But they don't. Productivity can grow at individual firms while aggregate productivity stagnates. The gains can be real but unevenly distributed. They can show up as profits rather than output. Measurement itself may fail to capture new forms of value.

Brynjolfsson and colleagues (2023) offer evidence that AI does increase productivity — dramatically in some cases. Their study of customer service workers using AI assistance found a 14% average productivity increase, with the largest gains (35%) for the least experienced workers. The AI effectively redistributed knowledge from the best performers to everyone else. The AI Coding Assistants research shows similar patterns in software development. But a finding about within-workplace leveling is not the same as a finding about economy-wide redistribution — and the more important question is what happened next.

But this raises a question the study doesn't answer: where do those productivity gains go?

Where Gains Don't Reach Workers

Four mechanisms explain why productivity gains flow upward instead of outward:

Market power. When a few firms dominate an industry, they capture productivity gains as profits rather than passing them to consumers (through lower prices) or workers (through higher wages). The tech sector is among the most concentrated in the economy. Google handles 92% of global search. Meta controls the social graph. Microsoft dominates productivity software. Amazon runs half of cloud computing. These firms don't face pressure to share productivity gains — they can simply keep them (Philippon, 2019).

Weakened labor institutions. In 1980, 23% of US workers were union members. By 2024, that figure is 10% — and only 6% in the private sector. Collective bargaining once forced productivity gains to be shared. Without it, gains accrue to whoever has leverage — and that's not workers. The decline in union density correlates almost exactly with the productivity-wage gap (Western & Rosenfeld, 2011).

Ownership structure. Productivity gains flow to whoever owns the productive assets. When those assets are physical plants and equipment, workers had some leverage — the factory needs people to operate it. When the assets are software, data, and algorithms, workers are more easily replaced. The shift from physical to intangible capital concentrates ownership in fewer hands (Haskel & Westlake, 2018).

Shareholder primacy. Since the 1980s, corporate governance has explicitly prioritized shareholder returns. Stock buybacks — illegal before 1982 — now return hundreds of billions annually to shareholders. In 2022, S&P 500 companies spent $923 billion on buybacks (S&P Global, 2023). That's money that could have gone to wages, R&D, or capital investment. Instead, it went to people who own stock — disproportionately the wealthy.

None of these are technological factors. All are policy choices.

"So-So Automation"

Daron Acemoglu (MIT) argues that the productivity paradox reflects a deeper problem: we're building the wrong kind of AI (Acemoglu, 2021).

He distinguishes three types of automation:

Productive automation — machines do tasks better than humans, generating genuine productivity gains. Think of automated manufacturing that produces more goods with fewer defects.

Labor-complementing automation — AI augments human capabilities rather than replacing them. A diagnostic tool helps doctors catch diseases earlier. The worker becomes more productive; their labor is more valuable.

"So-so" automation — machines replace workers in tasks where they're only marginally better or no better at all. Self-checkout kiosks are slower than cashiers but cheaper. Automated call centers frustrate customers but reduce labor costs. The technology replaces jobs without improving outcomes.

So-so automation is redistribution without productivity. Workers lose income; shareholders gain it; total output doesn't increase. The company's margins improve. The economy doesn't.

Why do firms choose so-so automation? Because the incentives push that way:

Tax treatment favors capital. The US tax code subsidizes capital investment through accelerated depreciation, investment credits, and favorable treatment of capital gains. Labor is taxed heavily (payroll taxes). This tilts the calculation toward automation even when it's not more productive.
Short-term pressure. Public companies face quarterly earnings pressure. Replacing workers with machines reduces costs immediately; whether it improves productivity shows up later (or never). Wall Street rewards cost-cutting.
Concentration enables laziness. Dominant firms don't need to be efficient — they can maintain margins through market power. So-so automation is good enough when you don't face real competition.

Who Captures the Surplus

When AI does increase productivity, the surplus goes somewhere. Tracing it reveals the redistribution:

Shareholders. The S&P 500 has returned an average of 10.7% annually over the past 30 years. This reflects productivity gains captured as profits. But stock ownership is wildly unequal — the wealthiest 10% of Americans own 93% of stocks (Federal Reserve, 2023). When productivity gains flow to shareholders, they flow to the already-wealthy.

Executives. CEO compensation at the top 350 US firms was 399 times the typical worker's pay in 2021. In 1965, it was 21 times. This isn't because CEOs became 19 times more productive — it reflects shifts in bargaining power, social norms, and corporate governance that allow executives to capture more of the value their firms create (Economic Policy Institute, 2022).

Platform owners. Digital platforms are productivity multipliers — they allow a small workforce to serve massive user bases. But the productivity of those platforms accrues almost entirely to their owners. Uber makes drivers more efficient at finding rides; Uber captures the margin. Amazon makes sellers more efficient at reaching customers; Amazon takes its cut. The platform owners didn't create the productivity — they positioned themselves to tax it (Srnicek, 2017). This dynamic is explored in The API Economy and Pricing Power.

Landlords (in a broad sense). Some productivity gains flow to whoever controls bottleneck resources. In tech, that's increasingly cloud providers, chip manufacturers, and AI model developers. Nvidia's market cap grew from $300 billion to over $2 trillion in 2023-2024 as AI boomed. They didn't capture the productivity of AI — they captured the toll for accessing the hardware that makes AI possible.

Historical Parallels

Previous automation waves followed similar patterns.

The first industrial revolution concentrated wealth in factory owners while workers faced dangerous conditions and subsistence wages. Living standards for workers didn't rise meaningfully for 40-50 years after industrialization began — the "Engels pause." Eventually, labor organizing, regulation, and competitive labor markets forced broader distribution. But it required struggle, not automatic adjustment.

Electrification transformed manufacturing productivity in the early 20th century. But as David Nye documents in Electrifying America, the benefits took decades to spread. Rural America lacked electricity until the 1930s — not because of technology, but politics. The Rural Electrification Administration brought power to farms only after political mobilization demanded it.

The computer revolution was supposed to automate offices out of existence. Instead, it created new categories of work while eliminating others. But the distribution was uneven — high-skill workers gained; middle-skill workers lost ground; the hollowing of the middle class accelerated.

Each wave eventually produced broadly shared prosperity — but only after political contestation forced redistribution. Technology created the surplus. Politics determined who got it.

What Redistribution Would Require

If productivity gains are real but captured at the top, the question becomes: what would it take to distribute them differently?

Changing tax incentives. Acemoglu and Restrepo (2020) propose rebalancing the tax treatment of capital and labor. Remove subsidies for automation. Reduce payroll taxes that make labor artificially expensive. Consider direct taxes on labor-replacing automation. The goal isn't to stop technological progress — it's to stop tilting the playing field against workers.

Strengthening labor power. When workers have collective bargaining power, they can demand their share of productivity gains. The collapse of unions correlates almost perfectly with the productivity-wage divergence. Sectoral bargaining (common in Europe), works councils, and labor representation on corporate boards all give workers voice in how technology is deployed and how gains are distributed.

Breaking up concentration. Monopolies and oligopolies capture productivity gains as profits. Competition forces those gains to flow to workers (through higher wages) or consumers (through lower prices). Serious antitrust enforcement in tech would change who captures AI's benefits.

Rethinking ownership. If productivity gains flow to owners, broaden ownership. Worker cooperatives, profit-sharing schemes, employee stock ownership, and sovereign wealth funds are all mechanisms for spreading ownership of productive assets. Alaska's Permanent Fund distributes oil revenues to all residents — a model that could apply to AI-generated wealth. AI Cooperatives and proposals for Universal Basic Income and AI represent concrete efforts to put these ideas into practice.

Public alternatives. Mariana Mazzucato (2021) argues that public investment creates the conditions for innovation — but private capture collects the returns. Public AI infrastructure, publicly funded research with public-benefit requirements, and public alternatives to corporate AI platforms could ensure that some productivity gains flow directly to the public.

These aren't radical proposals. They're the mechanisms that produced broadly shared prosperity in the mid-20th century. They were dismantled deliberately. They could be rebuilt deliberately — but the political forces that dismantled them have only grown stronger in the interim, and the companies capturing AI's productivity gains are among the most powerful lobbying forces in Washington.

The Redistribution Lens

The productivity paradox isn't really a paradox. It's a distribution question misframed as a measurement puzzle.

Productivity gains are real. AI does make workers more efficient. The studies show it. The corporate profits show it. The stock market shows it.

But markets don't automatically share gains. The marginal productivity theory of wages assumes competitive markets, free entry and exit, and workers with bargaining power. None of those conditions hold in the industries deploying AI. Without them, productivity gains accrue to whoever has leverage.

This isn't market failure — it's markets working as designed. The rules — tax treatment, labor law, antitrust, corporate governance — all tilt toward capital. AI operates within those rules. It accelerates existing dynamics. If the rules favor concentration and capital, AI will accelerate concentration and capital accumulation.

Policy determines distribution. Technology creates the surplus. Institutions determine where it goes. The same AI, in a society with strong unions, progressive taxation, and robust antitrust, would produce very different distributional outcomes.

When someone tells you AI will make you more productive, ask: productive for whom? The gains are real. The question is whether you'll see any of them.

Sources

Acemoglu, Daron. "Harms of AI." Cambridge Journal of Economics, 2021. Cambridge
Acemoglu, Daron and Pascual Restrepo. "Robots and Jobs: Evidence from US Labor Markets." Journal of Political Economy, 2020. JPE
Autor, David. "Work of the Past, Work of the Future." AEA Papers and Proceedings, 2019. AEA
Brynjolfsson, Erik et al. "Generative AI at Work." NBER Working Paper, 2023. NBER
Bureau of Labor Statistics. "Labor Productivity and Costs." 2024. BLS
Economic Policy Institute. "The Productivity-Pay Gap." 2024. EPI
Economic Policy Institute. "CEO Pay Has Skyrocketed 1,460% Since 1978." 2022. EPI
Federal Reserve. "Survey of Consumer Finances." 2023. Fed
Haskel, Jonathan and Stian Westlake. Capitalism Without Capital: The Rise of the Intangible Economy. Princeton University Press, 2018. Publisher
Mazzucato, Mariana. Mission Economy: A Moonshot Guide to Changing Capitalism. Penguin, 2021. Publisher
McKinsey Global Institute. "The Economic Potential of Generative AI." 2023. McKinsey
Philippon, Thomas. The Great Reversal: How America Gave Up on Free Markets. Harvard University Press, 2019. Publisher
S&P Global. "S&P 500 Buybacks." 2023. S&P
Srnicek, Nick. Platform Capitalism. Polity, 2017. Publisher
Western, Bruce and Jake Rosenfeld. "Unions, Norms, and the Rise in US Wage Inequality." American Sociological Review, 2011. SAGE

General

The Productivity Paradox