Jonathan H. Choi (Washington University) & Paul Connell (Wisconsin), AI Deceleration (or Acceleration) by Taxation:

Even the chief executives of America’s leading AI companies cannot agree on whether their products will substitute for or complement human labor. Yet most policy proposals, from data center bans to bonus depreciation for AI hardware, commit policymakers ex ante to just one view of how AI will integrate with workers. By extending a leading macroeconomic model of worker automation, we show that any fixed policy is unacceptably fragile: an AI tax calibrated to a world in which AI on net substitutes for human labor produces large welfare losses in a world in which AI is a net complement, and vice versa.

We propose an adaptive Pigouvian rule that responds to observed wage dynamics by taxing automation when real wages fall and subsidizing it when they rise.

We show that this rule is the unique policy in our comparison set with positive welfare gains in both worlds, and it exhibits other attractive qualities that are emphasized in the policy robustness literature. From a coalition-building perspective, it is the only policy mechanism that both AI decelerationists and accelerationists should commit to ex ante. We propose implementing this adaptive rule through a two-pronged reform: (1) equalize the existing capital-labor tax wedge, and (2) enact a national AI-usage sales tax whose rate (which may be negative) is indexed to AI-induced unemployment (or change in real wages) by an independent labor-market body.

Editor’s Note:  If you would like to receive a daily email with links to tax posts on TaxProf Blog, email me here.