Schmidhuber: Turing Test is a bad intelligence measure — no AGI without mastery of the real world
SchmidhuberAI · x · 2026-09-23
In a reply to Elon Musk and François Chollet, Jürgen Schmidhuber restates his long-held view: today's only well-working AI lives behind the screen — summarizing documents, playing chess, proving theorems, generating images and programs — yet no AI-controlled robot can match a plumber or even a capuchin monkey, because the physical world is far more demanding.
- Passing the Turing Test is far easier than True AI in the physical world, so it's a poor measure of intelligence
- AI software research alone isn't enough; it must be combined with machines and robotics
- "No AGI without mastery of the real world"
The argument comes from Sec. 20 of his "Annotated History of Modern AI and Deep Learning" (survey arXiv:2212.11279).
More from AGI Musings
- Philosopher Véliz: AI presents predictions as facts, raising deep ethical issues — CarissaVeliz · 2026-09-23
- Model release cadence shrinks from 73 to 18 days, sparking safety calls — tristanbob · 2026-09-23
- AI Twitter Jokes That EA's Insect Rights Obsession Is a Secret Jain Cabal — tszzl · 2026-09-23
- AI agents are becoming marketplaces - and neutral brokering is the winning incentive — nicolechirps · 2026-09-23
- Personal GPUs: 'Everyone will own their own AI like a PC,' argues researcher — 0xsachi · 2026-09-23
- Don't be fooled by this summer of AI hype: scrutiny deflates the claims — slhamlet · 2026-09-23