The Pill and the Dish: Robust AI Policy for Two Possible Futures

Governments are writing rules for artificial intelligence while its direction is still deeply uncertain. This report sets AI aside and follows two imaginary technologies with similar economics: an augmentation pill that people gradually allow to act on their behalf, and a lab-grown "DishBrain" that may someday have interests of its own. Each is traced through a series of stages. At every stage the report asks which policies would help, and whether those policies would still work at later stages. The results are then mapped back to AI. Both pathways cross a visibility threshold: a point past which the causes of a system's behavior can no longer be established from outside. Two familiar tools fail at that threshold. One is labeling machine-made work. The other is assigning blame to whoever caused a harm. The report proposes eighteen policies. Eleven of them hold on both pathways because they attach to things that stay observable: physical inputs, named responsible parties, measurable effects, and the system as a whole. Each policy is tested for durability past the threshold, and also for cost, for who bears that cost, and for its potential for abuse.