How can AI reshape employment—not just by replacing tasks but by amplifying human expertise?
Editor’s note: This episode is a VoxTalks Economics podcast by CEPR, and is available on Spotify, Apple Podcasts, or wherever you get your podcasts.
Interested in the economics of AI? Check out our latest series of podcasts, Ideas in Development, which features ten episodes on AI in low- and middle-income countries.
The 2025 CEPR Policy Forum in Paris brought together economists to examine the evolving world of work. A major theme was the impact of artificial intelligence (AI) on labour markets. In his compelling keynote, David Autor presented a provocative and hopeful vision of how AI could reshape employment—not just by replacing tasks, but by amplifying human expertise and potentially rebuilding the middle class. David joined VoxTalks to discuss the key insights from his keynote.
“AI is an amazing tool that has great potential and we should recognise that it opens a range of possibilities, from pretty good to pretty bad.”
Why we must think beyond automation
At the heart of Autor’s argument is a distinction between two types of AI tools: automation and collaboration.
“An automation tool is something that eliminates expertise… A collaboration tool is a force multiplier for expertise.”
He warns that much of AI development is currently skewed toward automation, often with disappointing or even dangerous results. A better future lies in AI designed to enhance, not replace, human judgment.
Expertise: The true currency of the modern labour market
A major concept introduced in the episode is that of ‘mass expertise’. Autor defines expertise not as formal education but as “the domain-specific knowledge or competency to do some practical and valuable thing.” This could be anything from diagnosing a patient to designing a kitchen or coding an app.
Crucially, the value of expertise hinges on two conditions: it must be useful and it must be scarce. If deployed poorly, AI could devalue expertise by making tasks too easy or generic.
“If everyone is expert, no one is expert.”
This shift in the valuation of expertise, Autor argues, is not hypothetical—it’s already happening. In grocery stores, for example, cashiers’ productivity has increased due to automation, but their wages have not risen.
The paradox of productivity: More jobs, less pay?
Autor challenges the intuitive belief that as jobs become more skilled, employment in those roles should increase. In fact, he reveals, the opposite often happens. Using decades of empirical data, Autor and co-author Neil Thompson find that as jobs become more expert—through the removal of routine, low-skilled tasks—wages rise, but employment does not.
Conversely, jobs that become less expert tend to attract more workers but pay less. For instance, Uber driving saw a 240% increase in employment in the US, yet wages declined.
“It's kind of counterintuitive… that if you take away some of the expertise in a job, then employment in that job goes up.”
This trade-off suggests that while automation might expand access to some occupations, it risks diluting the value of those roles if expertise is stripped away.
AI design as a political and institutional choice
Throughout the episode, Autor stresses that the impacts of AI are not deterministic. “How AI design is designed is a choice”, he says. The structure of our institutions—professional guilds, licensing bodies, firms, and education systems—will determine who benefits.
He highlights how restrictive institutions protect elite professions. In the US, for example, the American Medical Association has resisted expanding the role of nurse practitioners for decades. Autor argues that empowering more people to do expert work with the aid of AI is not only possible, but necessary for broadening access to economic opportunity.
“The good scenario, if we use AI well, would be to enable more people who are not the educated elites… to do more valuable, expert, judgmental work.”
Rethinking occupational bundles and task reallocation
One of the episode’s most original ideas is that jobs are not single tasks, but rather ‘bundles’ of tasks. When automation removes one task, what remains can fundamentally change the nature of the job.
“You should care about what part of the bundle is being done by the machine, and what part remains for you.”
A striking example is from construction, where rebar tying—once a skilled activity—can now be done with a $2,000 tool. Workers resent this not because they dislike tools, but because it devalues a form of expertise they have mastered.
“It’s not collaborating with them, it’s competing with them.”
Time, trust, and the myth of rapid displacement
Despite the urgency around AI, Autor believes change will come more gradually than some predict. Technologies take decades to be fully adopted, even if invented today. For instance, he notes that even if autonomous vehicles were perfected tomorrow, it would take 25 years to replace the global fleet.
Furthermore, he is skeptical about the breathless optimism surrounding AI’s capabilities. Rather than a sudden employment apocalypse, he foresees a long transition where some professions transform or fade while others evolve in tandem with new tools.
Rebuilding the middle class with AI
At its most hopeful, Autor’s message is that AI could reverse four decades of job polarisation.
“A good scenario for AI is that it would lower the barriers to entry by enabling more people with the right training and judgment to do a larger set of care tasks, of legal tasks, of software tasks, of education tasks.”
This would mean breaking the monopoly of elite professions and allowing workers without university degrees to access better-paid, more meaningful jobs—aided by AI that collaborates rather than replaces.
A cautious optimism
Ultimately, Autor calls for both optimism and humility in the face of AI’s rise. He urges us to resist both utopian and dystopian visions, and instead make deliberate, institutional choices that shape how AI supports work.
“We should be optimistic and pessimistic simultaneously. We should be cognisant of the risks and the downsides, and never think that everybody benefits, but recognise there are really good things that could come.”