AI as the engine, humans as the steering wheel

AI agents would compete as open-market "solvers" while humans steer through simple verification mechanisms, rather than a single AI governing. The author argues that a single AI fails credible neutrality, since LLMs encode specific preferences, are about as complex as all US law, and would need replacing every three months. Futarchy, from Robin Hanson's "vote values, but bet beliefs," fits this pattern, with traders predicting outcomes and voters verifying a chosen metric. In distilled human judgement, a jury answers about 100 of a million questions and checks submitted answer lists; in a toy example, a blend of 0.199 Claude and 0.801 Deepseek answers fit the jury best.