Shanahan's role-play paper: describe LLMs with programs, not minds

gerardsans · x · 2026-09-23

Replying to the agent-consciousness debate, @gerardsans cites Murray Shanahan et al.'s paper "Role-Play with Large Language Models" (arXiv:2305.16367). It argues that as dialogue agents act increasingly human-like, we need high-level descriptions that avoid anthropomorphism: framing agent behavior as role-play lets us use folk-psychological terms without ascribing traits LLMs lack. The paper applies this to apparent deception and apparent self-awareness — use programs, not minds.

Original post →

More from AGI Musings

AGI Musings channel →