Research
Research projects
- Harnesses, Bodies and Deaths: Notes from an EveningWhat does a system need beyond raw intelligence in order to adapt? An open evening of conversation explored why agent swarms drown in bureaucracy, why language models understand stories but not numbers, and why a well-designed intelligence might need a way to die.
- How the most successful founders take their strategies from biology.Why should a founder care how a slime mould survives a famine? This piece, based on Adam Timlett's talk and the discussion that followed, looks at what startups can learn from biology about risk: shock absorption and evasion, assets with hidden uses, and why the businesses that last hold not the best card but the thickest hand.
- On the Origin of RiskWe manage risk by holding our shape. Organisms lose shape deliberately, and collect on it. Timlett argues the difference is formal, not temperamental: shape conservation is built into the mathematics of prediction and optimisation, and it hides an entire class of strategy — including why agility pays even when the forecast is exactly right, and why a cell signalling channel that fails 8% of the time may be storing options rather than losing information.
- Can AI predict your next move?An essay on the limits of prediction: chaos theory, computational irreducibility, and what changes once autonomous agents start modelling each other. Along the way it asks whether Artificial Super Intelligence could ever out-predict a human — and concludes that the more pressing worry is not the AI, but the people controlling it.
- What is complexity?An inquiry into whether complexity is a property of systems or of our knowledge of them. We treat it as the gap between bounded models and unbounded worlds - and ask not how to close it, but how a system detects that its model has expired and pivots.
- Sema: when the hash is the wordSema is a decentralised protocol giving agents a shared, persistent vocabulary for coordination and verifiable agent-to-agent contracts. Tracking which Sema hashes agents use also exposes their reasoning traces for optimisation and auditing.