New paper: 'Embodied Hijack' explains why users keep anthropomorphizing disembodied LLMs
MacrinePhD · x · 2026-09-24
A hypothesis-and-theory article by Sheila Macrine (UMass Dartmouth) in Frontiers in Psychology argues that the standard 'naive anthropomorphism' account is incomplete: across evolutionary time, fluent, responsive communication came only from embodied, self-maintaining agents, and conversational LLMs are the first entity class to reproduce those social signals without the grounding properties — biological self-maintenance, vulnerability, temporal continuity — that historically produced them. This 'Embodied Hijack' exploits evolved social-cue detection, explaining why anthropomorphism persists even among users who know the system has no inner life.
More from AGI Musings
- Benjamin Bratton: the singularity is the wrong picture — the next intelligence explosion will be social — bratton · 2026-09-24
- AI floods courts with lawsuits — proposal: prediction markets to price and settle disputes — devanshmehta · 2026-09-24
- AI safety researcher: if you believed the risk, your tone would show it — geoffreyirving · 2026-09-24
- AI safety researcher on inauthentic comms: saying important things in unimportant tones — geoffreyirving · 2026-09-24
- Hugging Face CEO shares 3 lessons from first public disclosure of an agent cyberattack — mmitchell_ai · 2026-09-24
- 'De-coupling from the Silicon Valley grindset': dev pushes back on speed worship — floguo · 2026-09-24