Carl Benedict Frey is an Oxford economist and historian who heads the Future of Work program at the Oxford Martin School. He's known for his 2013 research on automation and employment, and his latest book How Progress Ends Technology, Innovation, and the Fate of Nations examines why technological advancement doesn't automatically guarantee economic prosperity.

Speaking with World Economic Forum digital editor Puja Chabria at Davos 2025, Frey argues that without proper institutional frameworks, AI risks following the same disappointing trajectory as the 1990s computer revolution - impressive technology that failed to deliver sustained productivity growth. The conversation covers market concentration in AI, the need for institutional flexibility, and why emerging economies might actually benefit more from current AI developments than developed nations.

Why the Computer Revolution Failed to Deliver Sustained Growth

The 1990s computer and internet revolution only produced "a decade-long productivity upsurge mostly confined to the United States" despite revolutionary capabilities like connecting global scientists and streamlining research processes.

Academic incentives demonstrate the problem: "publish or perish" means researchers use productivity tools to "do more things" rather than "drill deeper," spreading attention across more projects and reducing breakthrough potential.

"The technology itself is not all that matters. Institutions and incentives matter too" - Carl Benedict Frey explains why technological capability alone doesn't guarantee economic transformation.

Market Concentration Threatens AI Innovation Potential

Large firms use AI "for process improvements and automation" while smaller players "are more likely to push the boundaries and develop entirely new products and industries."

Compliance costs like GDPR have helped large tech firms "capture a larger share of the market" while young companies struggle with regulatory burdens.

"Most material progress comes from doing new and previously inconceivable things" - without airplanes, rockets, vaccines, antibiotics, and computers, automation alone would have only given us "productive agriculture" and "cheap textiles."

The scaling era of AI is ending as human amateurs beat top Go programs by exposing them to untrained positions, showing even superhuman performance has limitations in new circumstances.

Europe's Self-Imposed Trade Barriers and Innovation Challenges

European barriers to trade in services within the EU "amount to something like 110%" - equivalent to "Trump liberation day tariffs self-imposed on services inside the European Union."

This fragmentation caps "the return to innovation in Europe" by limiting market size, while the US and China benefit from "large harmonized markets to scale into."

Europe is "increasingly squeezed between the United States and China" and must harmonize its single market in services to compete in the digital economy.

The Zero-Sum Politics of Stagnant Growth

"Economies that are not growing eventually become a zero sum game where the only way for me to improve my standard of living is at the expense of someone else."

This creates "a mindset that if somebody's better off I must have lost" leading to "very contentious and polarized politics as we're having today."

"If AI does not live up to the expectation in terms of delivering productivity growth fairly soon, I think we are in a lot of trouble" - Carl Benedict Frey warns of intensifying political patterns.

High AI infrastructure investment is "pushing up interest rates" at a time when "debt levels are already very high," potentially squeezing public services further.

Emerging Economies' AI Advantage in Services

AI acts "like GPS technology for taxi services" in knowledge work, reducing barriers to entry and making location less important for service delivery.

An accountant in New York City "earns an order of magnitude more" than one in Manila, but AI could reduce this productivity differential and shift work toward lower-cost locations.

Machine translation is making "language barriers less of an issue" as "most trade in services have historically been confined to English-speaking countries."

This creates opportunity for "service-led growth" similar to how manufacturing migration "saw 800 million people being lifted out of poverty" in countries like China.

The Coming Paradigm Shift Beyond Large Language Models

"It's far from clear that the future of AI is large language models. It might be small language models. It might be symbolic AI. It might be a fusion" of different approaches.

Future AI models need to be "much more similar to humans which are much more energy efficient and much more data efficient capable of learning from only a few examples."

If "the AI bubble bursts," reduced capital flow will "force firms to actually invest more in data, compute, energy saving AI" and make "energy less of a bottleneck."

Historical precedent shows course correction takes time - the first industrial revolution required "around seven decades for the benefits to trickle down to the broader population."