Michael Tomasello on Machine Agency: LLMs Skipped the Infrastructure for Shared Agency
Machine Learning Street Talk · rss · 2026-11-19
Developmental psychologist Michael Tomasello (20 years at Max Planck comparing children and great apes, author of The Evolution of Agency) joins MLST to unpack what it would actually take for a machine to act for its own reasons.
- Agency begins with unpredictability: predictable environments get hardwired behavior; unpredictable ones get decision-makers — from reptile control loops to squirrels simulating falls to apes gathering information before committing
- Shared agency is uniquely human: joint attention, joint goals, perspective-taking, and conventions invented bottom-up (children losing audio on a video call invent gestures)
- LLMs started at the endpoint: Tomasello argues large language models skipped this social infrastructure and can't create conventions; a Turing test is far from sufficient
Also covers the stag hunt, why lying only works among cooperators, whether capital is an agent, and Dennett's intentional stance.
More from AGI Musings
- S.J. Murray essay: stripping friction from AI companionship strips love's power to transform us — AnnaCiaunica · 2026-11-19
- Novel as a form was invented by a woman in Japan 1,000 years ago, prompting AI-era reflections — CatAstro_Piyush · 2026-11-19
- RhoAgent in Nature Biotech brings AI design from proteins to programmable functional RNA — EricTopol · 2026-11-19
- AI medicine researcher invokes Musk's 'idiot index' to argue healthcare needs its own cost-transparency metric — zakkohane · 2026-11-19
- LLMs are language declaring independence from the human substrate: a viral poetic thread — aiamblichus · 2026-11-19
- Does Claude deserve co-authorship? Academics spar over AI-era authorship norms — paulnovosad · 2026-11-19