Grady Booch to Joshua Bengio: Consciousness Talk About LLMs Is Bad Science
Grady_Booch · x · 2026-09-27
Grady Booch pushed back at Joshua Bengio, arguing he's still 'engaging in bad science': words like think, reason, and consciousness are 'suitcase words' loaded with emotional baggage. Booch proposes testing precise claims instead — e.g., whether LLMs can generate Martin-Löf random numbers, which he says they are architecturally incapable of — where evidence and falsification actually work. A notable clash over the scientific rigor of AI consciousness discourse.
More from AGI Musings
- Jeremy Howard: the look-it-up learning loop is exponentially more powerful with LLMs — jeremyphoward · 2026-09-27
- Personal computing pioneers overestimated coding appeal; AI unlocks tiny personal software — pixlpa · 2026-09-27
- Systems thinkers are flocking to vibecoding, and it's only the beginning — eschadiol · 2026-09-27
- Yacine: we can now turn any measurable task into a general-purpose solver — yacineMTB · 2026-09-27
- AI co-scientists can hypothesize, design experiments and analyze data, but humans still decide what makes sense, Nature reports — irinarish · 2026-09-27
- Petabytes of agent logs nobody reads: researchers warn of AI oversight collapse — birchlse · 2026-09-27