Philosophers and Developers Clash Over Whether AI Can Be Conscious
On October 9-10, Twitter saw a philosopher-versus-developer debate over machine consciousness, centered on philosopher Benthamsbulldog and developers Gerard Sans and repligate. The dispute remains at the level of argument and thought experiments, but bears on frontier questions of AI ethics and safety.
Confirmed
- Benthamsbulldog's core argument: the computations AI performs may suffice for consciousness under the most mainstream theories (e.g. computational functionalism), now or soon, whereas bricks and iPhones lack the requisite computational structure — so AI consciousness is a genuine possibility.
- He added that top philosophers of mind and cognitive scientists seriously entertain the possibility that AI models could suffer, while no one does so for iPhones, suggesting the former is considered far more plausible.
- Gerard Sans counter-argued that current LLMs are essentially text samplers with no 'self' inside.
- Sans proposed a falsifiable experiment: train a small transformer on phone books or transaction records (Stripe-style), with no assistant data, no persona instructions, no system prompt and no 'I', then test whether a first-person self emerges on its own.
- repligate argued that judging whether AI could be conscious does not require those mainstream consciousness theories at all.
Unconfirmed
- Sans's phone-book training experiment has not been run; his prediction remains unverified.
- The core dispute is unresolved: whether computational functionalism applies to LLMs, and whether pure-corpus training rules out consciousness, remain points of disagreement.
Why it matters
- The debate touches frontier AI safety and ethics: if models could suffer, training and deployment policy would be directly affected; repliers noted relevant research is ongoing.
- Sans's falsifiable experiment offers an actionable technical test for an otherwise abstract philosophical dispute — informative regardless of outcome.
2026-10-09 ~ 2026-10-10 · 6 related posts
Primary sources
- Debate: leading philosophers take AI pain seriously — and incentives sit behind AI welfare — Benthamsbulldog · 2026-10-09
- [source] Why AIs might be conscious but bricks aren't: the computational-functions argument, plus pushback — Benthamsbulldog · 2026-10-09
- [source] Dev Pushes Back on AI Consciousness Claims: 'A Text Sampler Has No Self' — gerardsans · 2026-10-09
- Falsifiable Challenge to AI Consciousness: Train a Transformer on a Phonebook — gerardsans · 2026-10-09
- The Falsifiable LLM Self Test: Train a Transformer on a Phonebook and Ask Who It Is — gerardsans · 2026-10-09
- [source] repligate pushes back on using 'leading theories' to argue AI consciousness — repligate · 2026-10-10