Economist Alex Tabarrok on AI's economic impact: productivity gains will be broad, jobs won't collapse — but change will take longer than technologists think

Alex Tabarrok on AI, the Baumol Effect, and the pace of economic change

Alex Tabarrok, economist at George Mason University and co-author of the Marginal Revolution blog, argues that AI will broadly distribute productivity gains across the economy — but that technologists are systematically too optimistic about how quickly those gains will show up in jobs data and GDP.

The Baumol effect and why services keep getting more expensive

The conversation opens with the Baumol effect — the long-running pattern where manufactured goods get cheaper while services like healthcare and education keep getting more expensive. Tabarrok's explanation is structural: service industries, particularly education, have seen almost no productivity growth in centuries. A professor today does roughly what Pythagoras did, with marginally better tools. Because wages in flat-productivity sectors still have to compete with wages in improving sectors, prices in services must rise regardless of regulation.

Tabarrok pushes back on the regulatory-capture explanation that circulates in policy debates. Healthcare costs were rising before Medicare and Medicaid existed. Cobblers and car repair shops aren't expensive because of occupational licensing — they're expensive because labor costs have risen everywhere else. Consumers keep buying more healthcare and more education precisely because they're getting wealthier, which is itself evidence that price growth in services reflects abundance elsewhere, not market failure.

“The models seem to be getting more powerful and cheaper at a faster rate than any other technology that I have ever seen... I trust the technologists when they say that the technology is going to keep getting better quite rapidly. Where I think the technologists are not quite right is that it's gonna take much longer than they think for this to start affecting jobs and the economy... Never in my life have I felt that the window of what is possible is as large as it is today.”

What AI could change

Tabarrok sees AI's biggest economic opportunity in medicine, not software. A 5% reduction in cancer mortality, he says, would be worth trillions. Drug discovery and diagnostic breakthroughs are well within reach, and the division of labor in healthcare will deepen further — more specialization, not fewer physicians.

On jobs broadly, the data so far shows AI increasing employment, not reducing it. Tabarrok is relaxed about the job apocalypse scenario on first principles: if AI does eliminate large categories of work, the result is an economy producing vastly more wealth, which is a distribution problem rather than a destruction problem. Growing-pie problems, he argues, are solvable in ways that shrinking-pie problems are not.

Value capture and the open-source ceiling

Tabarrok is skeptical that frontier AI labs will sustain durable pricing power. The underlying technology is linear algebra; the ideas that made it powerful were few and are now widely understood. OpenAI and Anthropic are roughly six months ahead of open-source alternatives, and for most use cases the open-source models are already sufficient. Models are getting more powerful and cheaper faster than any prior technology he can recall, which suggests gains will be widespread rather than concentrated at the frontier. Early employees at the leading labs will be fabulously wealthy; the technology itself will not stay proprietary for long.

Where technologists get the timing wrong

Tabarrok's sharpest disagreement with the Silicon Valley consensus is on speed. He trusts the technical forecasts — the models will keep improving rapidly, and the possibility of superintelligence is not insane — but the economic diffusion will be slower than technologists expect. Electricity took decades to reshape production structures even after the technology was clearly superior. The same lag should be expected here. The window of what is possible, he says, is wider than at any point in his career, in both directions: the upside case and the catastrophic-risk case are both non-trivial, and neither should be dismissed.