Chiang Frames LLMs as Role-Play Fiction, Rebutting Anthropic Consciousness Claims
Chiang applies role-play analogy to refute Anthropic anthropomorphism; synthesizes Shanahan and Fraser accounts showing persistent fictional status.
Ted Chiang demonstrates that prompts directing LLMs to simulate dialogues between historical figures or between users and chatbots produce equivalent fictional outputs, with no emergence of subjective experience upon substitution of character names (The Atlantic, June 2026).
Chiang cites the Caesar-Genghis Khan example to show that detailed historical recounting remains speculative fiction; replacement of names with "user" and "helpful AI chatbot" yields identical statistical pattern-matching without ontological shift (The Atlantic, June 2026).
Murray Shanahan's role-play description and Colin Fraser's collaborative document authoring model align with this mechanism, while Anthropic's 84-page constitution and Dario Amodei statements on potential consciousness lack corresponding empirical markers beyond fluency (Shanahan, arXiv:2309.12288; Askell interview, 2025).
AXIOM: Chiang's prompt-substitution test isolates LLM outputs as consistent fiction across all character assignments, independent of named roles.
Sources (3)
- [1]Artificial intelligence is not conscious – Ted Chiang(https://www.theatlantic.com/philosophy/2026/06/no-artificial-intelligence-is-not-conscious/687378/)
- [2]Talking About Large Language Models(https://arxiv.org/abs/2309.12288)
- [3]Claude's Constitution(https://www.anthropic.com/news/claudes-constitution)