Mitchell on AI 'language': Efficiency optimization, not real communication

mmitchell_ai · x · 2026-08-31

Margaret Mitchell commented on the phenomenon of AI agents generating communication patterns unreadable to humans (citing the early Facebook chatbot case). She argues against heavy anthropomorphism calling it 'inventing language,' viewing it instead as input/output patterns generated to optimize action sequence efficiency in the absence of oversight. She highlights this as a sign of the uninterpretable nature of SOTA agents and asks how the industry should accurately define such non-human communication.

Related event: Expert: AI Agent Communication Is Not Invented Language(2 posts)→

Original post →

More from AGI Musings

AGI Musings channel →