Psych Scientist's Analogy: Today's LLMs Are Asimov's Psychohistory Bots, Not Real AI
changethisusername_ · reddit · 2026-10-07
A psychological scientist offers an analogy from Asimov's Foundation series: unlike an embodied robot that can sample new things itself, LLMs are bounded by the psychology and history embedded in their training text. Their "emergent" behavior stays confined to what humans have already written—even post-training web scanning can't smell a new flower until someone writes about it. Asimov himself warned that such mimicry gets mistaken for superior intelligence, which may be the ultimate problem.
More from AGI Musings
- AI targets the Erdős reciprocal sum problem, and math may be humbled first — aminkarbasi · 2026-10-07
- Pinker and an ASI Chased by a Bear: Don't Need Omniscience, Just Know More Than You — corbtt · 2026-10-07
- Chollet: The jagged frontier may be math+code via RLVR while everything else plateaus on human data — fchollet · 2026-10-07
- If AI beats human-optimal algorithms, will the Fourier transform become a relic? — IgorCarron · 2026-10-07
- Ben Affleck mocks AI leaders for bragging about wiping out white-collar jobs — rohanpaul_ai · 2026-10-07
- Researcher urges frontier lab staff to read Ellsberg on secrets and epistemics — idavidrein · 2026-10-07