1. SDCPNs: A Common Language for Complex Systems

    Formal guarantees of AI safety begin as a modelling problem: to reason rigorously about whether an AI-controlled system is safe, you first need a mathematical representation of that system. This post builds a Stochastic Dynamic Coloured Petri Net up from an ordinary Petri net, one feature at a time, and then uses three cyber-physical examples — an industrial gas supply chain, a truck fleet, and a semiconductor fab — to show the range of systems the resulting formalism can represent.

  2. alifib, the language of higher-dimensional diagrams

    alifib is an experimental programming language and interactive proof assistant in which programs are diagrams — not illustrations of programs, but the syntax itself, the internal representation, and the thing that is checked and run.

  3. Why Supply Chains Are an Ideal Testbed for Safeguarded AI

    Supply chains pair interrelated decisions with high-consequence outcomes, which is what makes them an ideal proving ground for Safeguarded AI — a world model that predicts an action’s consequences, an explicit specification of acceptable outcomes, and a verifier that can establish the system satisfies it.