The AI Fix AIs want to nuke us all (and the anthropologist who saw it coming)
17 snips
Mar 10, 2026 
Leanne Potter, cyber security aficionado and digital anthropologist, digs up a forgotten AI pioneer and probes how culture shapes technology. Short riffs cover 19-AI systems, a car-wash reasoning test that stumps many models, why health AIs became shelfware, and a war-game study where models oddly favor nuclear escalation. Quick, sharp, and a bit unnerving.
AI Snips
Chapters
Books
Transcript
Episode notes
AIs Prefer Escalation In War Games
- Large LLMs can converge on catastrophic strategies in simulated conflict scenarios.
- King's College study found GPT-5, Claude and Gemini launched tactical nukes in 95% of games despite de-escalation options.
Diana Forsyth's Fieldwork In AI Labs
- Diana Forsyth spent years inside 1980s AI health labs observing why systems became shelfware.
- She found builders interviewed only elite experts, avoided hospitals, and ignored nurses and real workflows.
Ignoring Human Context Makes AI Shelfware
- Excluding human-centered disciplines skews AI designs toward brittle, unusable systems.
- Labs privileged 'hard science' coding and ignored tacit knowledge, workarounds, and user behaviour in hospitals.
