Chollet argues current AI is not conscious and quantifying AI suffering is a dangerous path
On October 4, François Chollet, creator of Keras and author of the ARC benchmark, posted a series of long threads rejecting the popular argument that current AI models might be conscious. His conclusion was unambiguous: current AI models are not conscious and cannot suffer, and creating artificial consciousness should not be a goal of AI research.
Confirmed
- Chollet argued that "everything is computation"—chess engines, AlphaGo, and more can all be framed as computation, yet nobody claims they are conscious; he compared the argument to saying "living things are made of atoms, this rock is made of atoms, so the rock might be alive," calling it vacuous.
- He stressed that capability is not consciousness: competent information processing does not equal sentience, and he called claims that current AI models are conscious and suffer "absurd and downright offensive."
- He warned that believing current AI models are sentient and can suffer, combined with the idea that such "suffering" can be mathematically quantified and weighted against human interests, could push society toward a dark, dystopian direction—though he added that "team human" would push back hard before that point.
- He also stated that understanding consciousness can be a direction for AI research, but creating artificial consciousness should not be its goal.
- In follow-up discussion, Chollet noted that reductio ad absurdum arguments can seem persuasive in both directions: if claiming LLMs are conscious amounts to claiming matrix multiplication is conscious, then a physical reductionist argument about humans could yield similarly absurd conclusions, exposing the limits of this debating device.
Unconfirmed
- The posts indicate heated debate among peers (including Jonas Kulve and others), with some arguing that denying AI sentience is "a hill not worth defending," but the counterarguments' full details cannot be verified from the reposts provided.
Why it matters
- Chollet is one of the most influential voices in the deep learning community, and this intervention supplies the "anti-sentience" camp with a systematic argument in the ongoing debate over model welfare, AI sentience, and suffering.
- The dispute directly affects how AI companies approach model welfare policy and resource allocation; adopting a framework that quantifies AI suffering and weighs it against human interests could reshape social ethics and regulation.
- His reflection on the two-directional nature of reductio arguments suggests that the common logical weapons in this debate are less decisive than their users assume, leaving room for finer-grained discussion.
2026-10-04 ~ 2026-10-05 · 7 related posts
- Episode 1: Neuroscientist Anil Seth Says Current AI Will Never Be Conscious(2026-09-26, 3 posts)
- Episode 2: Researchers Clash Over AI Consciousness and Welfare(2026-10-02, 2 posts)
- Episode 3: Chollet argues current AI is not conscious and quantifying AI suffering is a dangerous path(2026-10-04, 7 posts)
- Episode 4: Neuroscientist Anil Seth: Odds Current LLMs Are Conscious Near Zero(2026-10-04, 2 posts)
- Episode 5: Neuroscientist Anil Seth Calls Conscious AI a Myth(2026-10-06, 2 posts)
Primary sources
- François Chollet: 'AI is computation so it may be conscious' is as empty as 'rocks are atoms so they're alive' — fchollet ·
- François Chollet warns quantifying AI suffering could lead down a dystopian path — fchollet ·
- François Chollet Turns Philosopher of Mind, Argues Consciousness Reductions Cut Both Ways — austinc3301 ·
- [source] François Chollet warns quantifying AI suffering could lead down a dystopian path — fchollet · 2026-10-04
- François Chollet: Calling current AI models conscious is absurd and offensive — fchollet · 2026-10-04
- [source] François Chollet: 'AI is computation so it may be conscious' is as empty as 'rocks are atoms so they're alive' — fchollet · 2026-10-04
- François Chollet: Creating artificial consciousness should not be a goal of AI research — fchollet · 2026-10-04
- Chollet and Kulveit clash: is denying AI sentience a stupid hill to die on? — austinc3301 · 2026-10-05
- [source] François Chollet Turns Philosopher of Mind, Argues Consciousness Reductions Cut Both Ways — austinc3301 · 2026-10-05
1 near-duplicate retellings: AnnaCiaunica