Kelly’s AI Tax Bill Struggles to Link AI Activity to Economic Harm

Summary

Sen. Mark Kelly’s Make AI Work for Americans Act proposes three taxes intended to capture gains from artificial intelligence and fund support for workers and communities affected by economic disruption. The proposal would tax “computational processing,” define one unit as 10 kilobytes of binary data, impose a 5% levy on covered digital advertising revenue attributable to ads shown to US users, and apply a 50% tax to covered income above 40% of gross receipts. The analysis argues that the computational tax measures data volume rather than the work required to generate an output: a short output may require substantial processing, while a longer one may be relatively cheap. Even a better compute measure would remain an indirect proxy for job displacement, wage effects, or other labor-market costs. The digital advertising tax is easier to administer but does not establish that AI generated the revenue, reduced labor costs, or displaced workers. The excess-profits tax uses AI-related activity, revenue thresholds, and substantial electricity use to determine coverage, but does not isolate profits caused by AI or worker displacement; firms with employment-reducing AI uses could also fall outside the rules. The author says these chained proxies require stronger evidence and justification than the proposal supplies. As an interim alternative, Congress could seek a public claim on AI-company profits rather than pretend that bytes, advertising dollars, or electricity use directly measure economic harm. Better evidence could eventually link specific AI uses or model classes to measurable costs, but the article concludes that current evidence is insufficient for a direct AI-harm tax.