Ben Recht: utility maximization is inescapable—we must learn when it's misapplied
beenwrekt · x · 2026-10-01
In a blog post from Class 9 of his graduate seminar "Forecasting: A Critical Retrospective", Berkeley professor Ben Recht dissects expected utility maximization.
The core idea is simple—model expected benefit of acting vs. not acting, pick the larger—and it underpins classification rules, stochastic optimization, probabilistic thinking, and the philosophy of many powerful people in government and industry. The problem: the numbers are all made-up forecasts. Outside casinos we rarely know precise odds, and forecasting future values is even more uncertain than the odds themselves. His stance: utility maximization is inescapable, but we should learn when it's misapplied—a Keynesian critique of false precision in decision-making.
More from AGI Musings
- Educators debate whether kids should embrace AI early — erikphoel · 2026-10-01
- Only 31% of Americans now see college as very important as AI economy fuels trades training — NinaDSchick · 2026-10-01
- Beyond making agents want the right things: an 'alignment compiler' for multi-agent alignment — xuanalogue · 2026-10-01
- SynthID Bio Touted as Key to Responsible AI Synthetic Biology as Methods Scale Up — davidstutz92 · 2026-10-01
- AI for Biology Compared to Generative Media in Early 2023 as Scale Stays Hard to Predict — davidstutz92 · 2026-10-01
- 'General' intelligence is a myth: human minds are narrow savanna specializations, not yardsticks — YogeshMalik · 2026-10-01