Taxing an industry for cause before it has shown clear harm risks forgoing the benefits it offers before it has a chance to deliver them.
Taxing an industry for cause before it has shown clear harm risks forgoing the benefits it offers before it has a chance to deliver them.
S enator Mark Kelly’s (D., Ariz.) Make AI Work for Americans Act aims to make companies that profit from artificial intelligence pay their “fair share” and use those proceeds to “invest in American workers.” The bill would create an AI Horizon Fund — financed by three new taxes and overseen by a 15-member AI Advisory Council — to expand job training, national service, small business support, and unemployment benefits.
The premise and assumptions behind the bill are clear: AI has been, and will be, displacing workers on a large enough scale that it necessitates a new federal trust fund. But this is counter to most of the evidence we have about AI and the economy. Economists have yet to find clear signs of broad labor market disruption, even among heavily AI-exposed industries or entry-level openings. Even on its own terms, the bill’s design has massive problems regarding how it raises taxes, how it spends that money, and who makes the spending decisions.
To finance the Horizon Fund, three new taxes are to be levied: A 5 percent tax on digital advertising, a 50 percent tax on AI companies’ profits (in excess of 40 percent of gross receipts), and an excise tax on the output of AI models. But economics tells us that he who writes the check is not necessarily he who actually pays the tax. After countries like France, Italy, Turkey, and the United Kingdom imposed digital services taxes, Apple, Amazon, and Google passed those costs directly to advertisers, third-party sellers, and app developers. A usage tax could be even worse for the consumer. While one reason the bill’s supporters want the tax would be to slow down AI job replacement, a blanket usage tax would be one-size-fits-all and not distinguish between use cases that replace vs. complement human labor.
The unemployment insurance provisions in the bill also promise harmful effects. Within six months of the advisory council declaring a “significant labor market disruption,” states must replace 75 percent of the claimant’s previous wages for six months. The federal government would cover costs above 40 percent. But a meta-analysis confirms that more generous benefits lengthen unemployment. That same analysis also found that the optimal income replacement rate is 28 percent and raising it to 75 percent would slow how quickly workers re-skill and find new jobs.
Finally, there are rent-seeking problems in who decides where the money goes and who benefits from those decisions. How will the advisory council decide if there’s been a “significant labor market disruption,” and how will it steer the Horizon Fund’s spending plans? The council will be made up of AI experts, organized labor representatives, and leaders from the workforce development field, in addition to a host of government agencies that would administer the money. Such a group would hardly be a neutral arbiter. Since the body deciding whether a “crisis” exists would also steer where the funds go, its membership would have a strong interest in seeing those funds grow — even at the expense of developers and the broader economy.
In the announcement, Senator Kelly points to union training programs or community colleges as potential stakeholders. But members of the council, including organized labor leaders, would have a perverse incentive to claim disruption, even if there is evidence to the contrary, in order to receive funds for their training programs. A swath of incentivized groups pledged support for the Horizon Fund, including technical schools and union groups. Again, these are groups that will likely get extra funding as a result of declaring a workforce emergency. When beneficiaries help write the rules, spending will inevitably flow toward groups organized enough to capture those dollars. Putting them on the AI Advisory Council essentially guarantees it.
While the effects of AI on labor markets are not yet settled, the early evidence points away from mass job displacement. This fact warrants skepticism toward efforts to create a fixed federal apparatus built to solve a problem that has yet to materialize. Being proactive is admirable, but taxing an industry before it has shown clear harm risks forgoing the benefits it offers before it has a chance to deliver them.
Nicholas Thielman is a resident fellow in technology and innovation policy at the R Street Institute, and Justin T. Callais is chief economist at the Archbridge Institute.