Ted Chiang argues that large language models cannot be conscious, calling such claims a misunderstanding of how AI works. He emphasizes that AI excels at generating text but lacks consciousness or moral agency.
Ted Chiang published an article refuting claims that large language models (LLMs) could be conscious. He cites Anthropic's Claude 'constitution' and CEO statements as examples, warning that attributing consciousness to AI is a dangerous error.
Recently, AI companies and some researchers have publicly discussed the possibility of LLMs being conscious. Anthropic's Claude 'constitution' states that AI's moral status is deeply uncertain, and its CEO has left open the possibility of AI consciousness.
Chiang argues that this line of thinking obscures the real risks of AI and confuses accountability. Misunderstanding AI as conscious could lead to overlooking technical limitations and result in misguided regulation or ethical debates.
Regarding Ted Chiang's claim that 'AI is not conscious,' comments were skeptical, pointing out that the definition of consciousness itself is ambiguous, making it difficult to draw definitive conclusions. Some argued that LLMs, despite their language ability, lack emotion or embodiment, thus unrelated to consciousness. Others raised extreme skepticism, asserting that one can only be certain of consciousness in oneself, and provocatively questioned the lack of evidence for shared consciousness among humans.