Samba co-founder · AI safety · IP and privacy regulation · startups · humanities

Edinburgh, Scotland

No trackers, no ads, no data collected or retained.

Agentic AI, Ethics and Artificial Organisations

  • Addressing the Biggest Problems in AI — Most work on ethics and correctness of AI try to make individual models better-behaved. The Perseverance Composition Engine does something very different, it assumes that misbehaviour is going to happen, and structures the system so it is caught before it causes harm. This is the way human institutions have worked for centuries.
  • Slow LLMs and MCPs are Hiding Problems — Agentic AI systems are currently slow enough that serious concurrency and coordination problems remain invisible. This is a technical note about what happens when that changes.
  • Structure vs Constitution in AI Safety — Comparing two approaches, PCE vs Anthropic. Anthropic locates AI ethics in the agent, through training and guidelines. PCE locates it in the structure around the agent.
  • AI, PCE and the Geth Consensus — Science fiction saw artificial organisations coming years ago. This is a collaboration between me and my Consul agent exploring how the story of the Geth illuminates how multi-agent AI works.
  • Logical and thermodynamic reversibility — LLMs consume vast amounts of energy. One response to that is to build on the computer science principle that energy is not consumed by computation but by the erasure of information. By making LLMs which do not erase information, perhaps we can avoid an AI energy crunch.

Browse all content by tag or search full text.

Recently updated