Back to the Future for Taxation (www.amacad.org)

The piece argues that the twentieth- and early twenty-first-century dominance of income taxes and broad consumption taxes (like VAT) proved brittle in the face of new economic realities—most notably machine labor and hard-to-measure externalities—and that policy has swung back toward historically common excises and targeted charges. From a 2075 perspective this is framed as a deliberate return to diversified, instrumented taxation designed not just to raise revenue but to internalize harms (health, ecological) and to put human and machine labor on a level playing field. The author stresses that tax design is ultimately constrained by what can be measured and audited, and reviews the recurring “interrogatives” of taxation (what, whom, when, how much, how, where), the streetlight effect of measuring only what’s easy, and longstanding boundary problems that swamp both income and consumption taxes (imputed income, in-kind labor, consumption vs investment). VAT’s anti‑cascading cleverness is noted, but so are its shared conundrums with income taxes. For the AI/ML community the implications are concrete: moving toward excises or activity‑based charges will force new measurement, attribution, and reporting standards for models, inferences, energy/carbon use, and machine-provided services. Expect technical demand for telemetry, verifiable usage accounting, privacy-aware audit trails, per‑inference or per‑model levies, and standardized definitions of “machine labor” versus human work. That will shape system architecture (on‑device vs cloud), model lifecycle accounting, research priorities on efficiency and explainability, and the incentive landscape around deploying AI while pricing ecological and social externalities.