When should governments start taxing AI? Not when AI becomes capable. Not when unemployment rises. Growiec, Prettner, and Szkróbka derive the trigger point from an optimal control model: tax AI when cognitive workers begin considering switching to manual jobs.
The economy has four factors: human manual labor, human cognitive work, physical capital, and AI. AI substitutes for cognitive work specifically. As AI capability grows, the marginal value of cognitive labor drops. At some point, a cognitive worker earns more by switching to manual work — the substitution pressure hasn't eliminated their job yet, but has compressed their wage below the manual alternative.
This switching threshold is the optimal moment to introduce AI taxation. Before it, AI is purely additive — it augments cognitive output without distorting labor allocation. After it, AI drives occupational displacement: workers move not because their jobs disappeared but because the wage differential reversed. The tax corrects this distortion by keeping the cognitive wage above the switching point, preserving the allocation that the pre-AI equilibrium supported.
The structural point: the right trigger for AI taxation is not a capability milestone or an unemployment rate but a behavioral signal — the moment workers start comparing their options. The economy announces its own distortion through occupational switching, and optimal policy responds to the announcement rather than trying to predict it. The tax is not on AI's existence but on the labor market's revealed preference for rearrangement.