The Race — Building Democracy in the Age of AI. People race toward a shared finish line.
The Race
← Trends

Supporting case

A new MIT explainer returns to a central fiscal question: does the existing tax code subsidize automation?

What happened

MIT's Stone Center published an October 2026 explanation of Daron Acemoglu, Andrea Manera and Pascual Restrepo's 2020 research. This is recovered fiscal-architecture evidence, not a new 2026 estimate of AI job losses. The original paper calculates effective labor and capital taxes and models how their relative treatment changes firms' decisions to automate tasks.

Its mechanism is that favorable treatment of equipment and software can encourage substitution even when productivity gains are small. In its calibrated model, correcting capital/labor tax distortions—or adding a targeted automation tax when broader reform is constrained—raises employment. The paper's headline 4.02% employment gain under optimal taxes is a model counterfactual, not an observed outcome or forecast validated for today's AI. Original paper ↗.

Tax preferences can change employers’ incentives to automate tasks, so fiscal policy can shape how AI gains and transition costs are distributed. Model counterfactuals identify a possible mechanism; they do not establish current AI employment effects or enacted reform.

Dates and implementation stage

Development 2026-10

Supporting sources

Related trends