AI Policy Should Focus on Diffusion, Not Redistribution

A new “progressive vision” risks slowing the gains it hopes to share.

Courtesy Xavi Torrent/Getty Images

Last week, a group of Democratic strategists, former Obama and Biden administration officials, and center-left technologists launched the Center for Shared AI Prosperity, a new initiative aimed at ensuring that the gains from artificial intelligence are broadly shared. The group’s board is a sort of “who’s who” of the tech-friendly, left-wing policy establishment: Dylan Matthews, once of Vox, Zohran Mamdani advisor Morris Katz, and progressive data darling David Shor all make an appearance.

The initiative’s organizers are explicit about their ambitions to advance a “progressive vision” for the economy in the AI age. Yet their preliminary proposals—including expanding the social safety net, raising taxes on capital, and strengthening labor protections—resemble the sort of left-wing ideas already commonly floated by some of the familiar names. What’s new, in other words, isn’t the substance of these ideas, but the effort to repackage them as a response to artificial intelligence.

The center’s emphasis is explicitly redistributive. Its founders argue that artificial intelligence could concentrate wealth and economic power, weaken labor’s bargaining position, and produce large-scale dislocation. They contend that policymakers therefore need a dedicated policy agenda for distributing AI’s gains more broadly through a mix of greater tax redistribution and more generous transfer payments. As the group puts it, there’s currently no “serious and actionable redistributive economic framework for AI.”

Yet the center misdiagnoses the central economic challenge posed by AI. The first-order task isn’t to redistribute AI’s gains, but to ensure that the technology diffuses widely enough across firms, workers, and public institutions to generate those gains in the first place. The key constraints on that diffusion, and the larger productivity gains associated with it, are likely to be supply-side factors such as energy, data centers, and the capital investment needed to deploy AI at scale.

The real danger of a distribution-first approach is that, in trying to cushion the economic fallout from AI, policymakers enact policies that discourage investment and slow adoption. Taxes, regulations, and labor-market interventions that reduce the returns to investment would undermine the growth and job creation that make broadly shared prosperity possible.

The China shock offers a useful analogy. Research by economists David Autor, David Dorn, and Gordon Hanson has shown that import competition from China imposed significant and lasting costs on some workers and communities. The right lesson was that economic change can be disruptive, and policymakers need better ways to help people adapt.

Too often, however, the episode was treated as a case that trade itself is a bad idea. The risk is that policymakers make the same mistake with AI.

Doing so would mean sacrificing the massive economic upsides from AI. Analysts at Goldman Sachs estimate that widespread adoption of generative AI could raise global GDP by roughly 7 percent over the next decade. Even if that figure proves optimistic, relatively modest increases in productivity growth would have significant effects on living standards.

There are also ways for AI to broaden opportunity directly. By putting sophisticated capabilities into the hands of ordinary workers and small businesses, the technology can make specialized knowledge and skills far more accessible. More people will be able to do higher-value work, start businesses, and increase their productivity rather than simply rely on post-market redistribution to share the gains.

The same logic applies to government itself. Public administration remains slow, labor-intensive, and highly inefficient. If AI can help governments issue permits faster, reduce backlogs, detect fraud, and deliver services more effectively, the benefits will be widely shared while strengthening state capacity. This may be the most underappreciated dimension of the AI debate.

The first priority, then, should be to maximize the gains associated with AI technology. That requires making it easier to build AI-related infrastructure, including data centers and energy projects, reducing unnecessary regulatory barriers to adoption, and preserving the incentives that drive innovation, entrepreneurship, and investment.

Only then does the question of distribution become salient. Such questions are not unimportant. But there will be far less to distribute if policy undermines the technology’s growth potential.

This tendency to prioritize redistribution over growth isn’t unique to AI policy. In housing, energy, and labor markets, parts of the modern Left have often treated economic growth as secondary to the distribution of economic outcomes. The risk is that AI policy follows the same pattern: focusing disproportionately on how to divide the gains while paying insufficient attention to the conditions needed to generate them.

The Center for Shared AI Prosperity reflects that broader instinct. Yet if policymakers get the sequencing wrong—focusing first on distribution rather than diffusion—they risk slowing the growth and productivity gains on which broadly shared prosperity depends.

The real challenge isn’t to redistribute the gains from AI. It’s to ensure that those gains are generated and diffused as broadly as possible across the economy. Get that right, and AI could usher in a more prosperous and broadly shared future. Get it wrong, and AI policy may amount to little more than old politics dressed up in new packaging.