Bernie Sanders' recent proposal that the public should own half of artificial intelligence is unlikely to become policy anytime soon, but it reflects a broader debate that is gaining momentum among economists, technology researchers, and policymakers: How can Americans benefit if AI creates trillions of dollars in new economic value? There is no shortage of ideas, most untested. But the risks, and rising public opposition to AI, make the question an important one.

AI wealth has accumulated quickly in the stock market, but many Americans are still limited in how much they benefit from that growth. Recent survey work indicates that a majority of U.S. workers now want to hold corporations more accountable via an AI sovereign wealth fund. There have been unconfirmed reports that ahead of a highly anticipated IPO, OpenAI has discussed offering the government a 5% equity stake. Meanwhile, Jeff Bezos recently told CNBC that the best policy idea to level the economic playing field is simply eliminating federal income taxes for the bottom-half of earners in the U.S.

Recent survey data indicates that this question is embedded in a rapid change in public sentiment towards AI. An Emerson College poll released this week found that only 27% of Americans support data centers being built in or near their community, with 63% opposed. Public sentiment has soured substantially in less than a year. A similar poll conducted in December 2025 found that while 33% said that they would support such developments, only 42% voiced their opposition. Many Americans feel as though they have nothing to gain and everything to lose from AI.

"When I see the data center proposal, I don't see progress," said Will Hollingsworth, a Northeast Ohio resident, speaking at an April public comment session regarding a proposed 257-acre data center campus in Portage County. "I see a gamble where the big tech companies get the gold while Portage County foots the bill."

"We're being asked to sacrifice the lifeblood of our city so that a trillion-dollar company can save a fraction of a cent on its margins," Hollingsworth said in comments that went viral. "We're being asked to drain our reservoirs so [that] a chatbot can write a poem or so [that] our sheriff can generate a picture of himself standing next to Bigfoot."

Among economists, tech industry researchers, and public policy experts, there are multiple proposals that address Hollingsworth's sentiment that the potential outcomes from AI development are dramatically skewed in favor of corporations. These include partial public ownership models in AI and other shared equity mechanisms.

Computer scientist Jaron Lanier, who currently holds the Office of the Chief Technical Officer Prime Unifying Scientist at Microsoft Research, has argued for a model sometimes referred to as "data dignity," where people receive compensation for the information and contributions that help create AI systems.

"I spent some time with Sen. Sanders when he visited the AI community at Stanford," Lanier said. "Whether [his proposal] would be a good idea depends on the nature of the government that would be responsible for routing benefits to people," he said. If the government were to simply become "just another AI company," he prefers what he calls a more "distributed economic model."

"Good data and supervision," Lanier says, can result in enough real money having a significant impact on people's lives.

"But if the future is to be the normative Silicon Valley one, where people will be fictionally treated as becoming useless because their contributions have been anonymized and dismissed in favor of pretending AI did all the work, then it is better for some kind of support to come through a government structure with a participatory/democratic element," Lanier said.

Paying people directly for AI training data

The challenge is figuring out how such a system would work. AI models are trained on enormous amounts of information from millions or billions of sources. Determining which individual contributions created value, and how much they should be paid, might run into the same criticisms that dogged efforts to compensate people for search histories — while the sums are massive in the aggregate, the economic value of any single individual's data is low.

Raul Castro Fernandez, an assistant professor of computer science at the University of Chicago who has recently written about how to fairly compensation the public for AI, refutes the argument that it's infeasible to accurately track (and compensate) the enormous amounts of data points collected by AI models from human contributors. "The strongest version of profit sharing is not a tax but a compensation system tied to the human contributions that make AI systems valuable in the first place," Fernandez said.

"They [the AI companies] already estimate how much data matters through scaling laws," Fernandez said. "A plausible mechanism would look less like calculating the exact value of every individual 'token' and more like a collective-management system, analogous in spirit to music royalties: AI companies would pay a share of model profits into a pool, the aggregate share would be anchored by evidence about how much model performance depends on data, and payments would be distributed across creators, publishers, platforms, or other intermediaries according to audited measures of data contribution," he said.

But Nicholas Vincent and Brent Hecht, researchers from Simon Fraser University and Northwestern University, respectively, caution against this approach. In a 2023 study examining whether it's possible to adequately compensate individuals for their contributions to an AI system, Vincent and Hecht argue that attaching valuations to each person's data can be extremely subjective and potentially counterintuitive.

"Seemingly minor design choices can seriously change the distribution of data values, a serious concern for any human-AI system seeking to incorporate such values for payments or other purposes," they state. "If a technology is reliant on the collective contributions of millions or billions of people, we already know each individual value will be very small, so why bother spending time and energy performing [potentially costly] data value estimation?" they concluded.

Creating new powerful unions for the 21st century

Direct payouts might not be the only way to achieve a more equitable data sourcing process, however. Matt Prewitt, president of RadicalxChange Foundation and one of the two authors of that policy paper, advocates for creating a new class of legal rights that give people powers to shape how AI works, a 21st century version of unions in which "people cannot sign away these rights on an individual level. Instead, people must join together into associations to exercise these rights."

This would create a new class of regulated associations that have "a very serious seat at the table with AI companies, and that have the power to gain shares, remuneration, governance, and power," Prewitt said.

Economist and technologist Glen Weyl, a principle researcher at Microsoft and founder of RadicalxChange, argues that the goal should not necessarily be government or public ownership of companies. According to RadicalxChange, efforts to "either divide and fractionalize ownership (i.e., give more or different people a share of conventional ownership); or consolidate ownership (i.e., place it in the hands of some representative of the public, like the state)," are able to do some good but are best seen as "only band-aids."

"Fractionalizing ownership just 'spreads around' the same old extractive incentives of conventional ownership, while consolidating ownership 'puts all the eggs in one basket,' intensifying the risks of institutional capture and illegitimate representation," RadicalxChange staffers wrote in a policy piece arguing for new models centered on common ownership.

New corporate taxes, less work hours

Others say the mechanisms already exist for policymakers to create a more equitable AI economy without resorting to untested ideas. According to Dean Baker, an economist and co-founder of the Center for Economic and Policy Research, these include stronger corporate taxes, antitrust enforcement, and labor protections.

While Baker said he remains unconvinced that there will be mass displacement of human labor by AI, he added that he would still fall back on "old remedies."

These remedies could include "a workable corporate income tax at a higher rate, for all companies," Baker said. But he added the form of payment could be new. "The best way to do this is require companies to turn over non-voting shares equal to the targeted tax rate (e.g. 25% of shares for a 25% tax rate)," he said.

Additionally, Baker says when it is rigorously applied, anti-trust enforcement provides a plausible route to fairer economic distribution of AI profits. Baker offered the analogy of cheap Chinese products displacing the blue-collar workforce. "We let in Chinese manufactured goods to screw large segments of the blue-collar workforce. We should not have protectionism to keep Elon Musk and Mark Zuckerberg ridiculously rich," he said.

Not repeating the policy mistakes of the past, including the lax attitude towards the rise of social media and the earlier global outsourcing era, is high on the radar of some of the most-senior policymakers in the U.S., who are betting that AI, and the jobs aspect in particular, will grow as an electoral and social issue in the years ahead.

While the idea of a universal basic income — or a "universal high income" program as Elon Musk calls it — to combat mass unemployment has been bandied about for years, there is a simpler labor market mechanism for distributing future economic efficiencies created by AI that already has a global precedent.

The answer isn't no work, but less work.

"We set the 40-hour work week 90 years ago and it has not changed since," Baker said. "Other countries have shortened the work week, work year. If AI is going to give us the promised boom in productivity, let's lower the threshold to 32 hours, or possibly even lower. We can also double the overtime premium to 100% rather than 50%," Baker said.