As artificial intelligence moves closer to taking over more economic tasks, two economists are arguing that the biggest open questions may not be technical, but distributive. In a recent interview, Alex Imas of Google DeepMind and the University of Chicago and Phil Trammell of Epoch and Stanford discussed what could remain scarce in a world shaped by AGI, and what that scarcity could mean for wages, profits, and taxation.
Their conversation focused on a central policy issue: if AI generates large new wealth, who should capture it, and how should that wealth be shared across workers, investors, and countries?
Imas said one likely source of lasting value is what he called the relational sector, meaning goods and services that people value in part because a human is directly involved. He pointed to examples such as live performance or hospitality, where the fact that a person is doing the work matters to consumers. In his view, even if automation makes many other goods abundant, human involvement itself could stay scarce.
That argument set up a broader question about whether human-to-human services could become a major part of the economy even if machines can eventually do nearly all physical production. The interview explored a possible split between a machine economy that produces most material goods and a human economy that serves preferences for human-made experiences.
Imas cautioned, however, that economists should be careful about making precise forecasts. He argued that individual predictions about labor markets and AI may be less useful than aggregated approaches such as prediction markets. He cited the large disagreement among economists when forecasting the labor market and said better data would improve the quality of economic planning.
He also drew on history, saying debates about automation and work have been going on for two centuries. He noted that David Ricardo foresaw both the benefits of industrialization and the risk of mass unemployment, but missed the way cheaper goods would create new demand and new kinds of jobs. For Imas, that example shows why outcomes under AI remain uncertain.
Trammell added that some parts of production may already be moving toward full automation, even if that shift is not always visible from the final product alone. He said economists should look down entire supply chains, not just at the final step, to understand how much labor still contributes to value. In some sectors, he said, the network-wide share of capital could eventually approach one if every stage can be automated and no human input is intrinsically valued.
But Trammell said the effects of that transition on the overall economy are not straightforward. If some sectors become fully automated while others retain a premium for human involvement, the result could be very different from a simple collapse in labor’s share.
The interview also touched on a longstanding economic puzzle. For decades, labor has received a little over 60 percent of national income in many advanced economies, with the rest going to capital owners. Imas said that stability has surprised economists, especially given the scale of industrial automation already in place. He noted that some researchers believe the labor share may be flatter than commonly thought once accounting changes are taken into account.
One reason the discussion matters for policymakers is that AI could change the basis for taxation and redistribution. If capital captures a larger slice of AI-driven output, governments may need to reconsider how they tax profits, ownership, and automated production, while also thinking about how countries outside the AI supply chain can participate in the gains.
The interview did not settle those questions, but it framed them as urgent. As AI advances, the economists suggested, the key issue may not only be how much wealth is created, but which parts of human labor remain valuable enough to command a premium, and how that value is distributed.