Kanjun Qiu quote-posted the argument and called it “actually reasonable.” Albrecht also drew a boundary around his claim: he was discussing LLMs, not agents that use tools and operate over time, which he said might present a different question.
Competence is not the same as experience
François Chollet made a related case from another direction. He argued that competence in information processing does not imply sentience, comparing current AI systems with calculators, chess engines and self-driving cars. In a follow-up, he said a static input-output program lacks traits commonly associated with consciousness, including information integration, interoception, temporal binding and embodiment.
Consciousness researcher Anil Seth said he broadly agreed with Chollet that current AI is “vanishingly unlikely to be conscious,” while adding that complete certainty is not possible.
The analogy has limits
Jay Hack challenged the lookup-table argument. He said it does not distinguish a function from its implementation and that a lookup table lacks the causality and state transitions of the algorithm that computes it.
Jason Wolfe took a more agnostic position, writing that science cannot currently answer the question and that conscious AI remains a matter of research and debate. The disagreement leaves two issues open: whether a system’s input-output behavior is enough to assess consciousness, and whether the answer changes when a model becomes part of a stateful, tool-using agent.
