dioscuri: 'local' vs 'global' superpersuasion are very different AI risks
dioscuri · x · 2026-10-29
Responding to discussion of AI "superpersuasion," dioscuri offers a useful distinction:
- Local superpersuasion: getting one specific person to believe one specific thing
- Global superpersuasion: finding whatever levers across a population achieve your goal — "I may not sell John THIS timeshare, but I can sell others enough that it doesn't matter"
The underlying point: people underestimate their own suggestibility, and AI doesn't need full superpersuasion to meaningfully shape human decisions at scale in high-stakes situations. Risk assessments shouldn't fixate only on persuading specific individuals.
More from AGI Musings
- Applied cybernetics: keep yourself in the outer loop as agents automate the inner ones — curious_vii · 2026-10-29
- AI may replicate human skills, but not human connection or lived experience — LuizaJarovsky · 2026-10-29
- UK civil servants convened AI expert panel to shape policy, insider reveals — robertskmiles · 2026-10-29
- Roll-up founder plans AI "CFO" to let 10-person accounting firms serve 40 clients — alex_verem · 2026-10-29
- Columnist argues LLMs can't write novel content as humanities make a comeback — StewartalsopIII · 2026-10-29
- Radiology First to Be Fully Automated? Prediction: Nearly Autonomous by 2027 — MaziyarPanahi · 2026-10-29