Projects & Innovation Innovation initiative

Next-generation trustworthy operational AI

Operational AI you can actually deploy, because every decision it makes can be validated against explicit rules before it acts. Initiated and developed by us with a research consortium.

  • Multi-agent AI
  • Federated AI
  • Neuro-symbolic reasoning
  • Causal AI
  • Edge AI
  • Continual learning

Project details.

Pending Project facts for this page Status, programme, ElyonX role, duration, and whether funding value and partner names are public. Send these and they replace this block. PROJ-02

The challenge

Operational environments do not reject AI because it is inaccurate. They reject it because nobody can say in advance what it will do, and nobody can reconstruct afterwards why it did what it did. A model that is right ninety-nine times and catastrophically wrong once is not deployable in a network or a plant.

The usual answer is to keep a person in the approval path, which removes most of the value. The real requirement is a system that can act on its own and still produce an argument for every action it took.

Neural model proposes an action Rule layer explicit constraints safety & service limits knowledge graph checked before anything runs Execute approved action runs Blocked & logged with the specific rule that caught it

Our approach

A symbolic layer sits between the model and the world. Explicit constraints, service and safety limits, and a knowledge graph check every proposed action before anything executes. Approved actions run; rejected actions are logged with the specific rule that caught them.

Learning happens across sites that cannot share data, using federated training so no participant ever exposes its raw data, and causal models sit alongside the learned ones so the system reasons about what an action will cause rather than only what has correlated.

Related