Who watches the AI watchman?

Researchers at the University of Waterloo are using applied mathematics and machine learning to ensure that increasingly autonomous AI systems—such as those managing power grids or autonomous vehicles—can be trusted and verified. Their team developed neural-networks trained to satisfy mathematical stability proofs and a logic-based system to rigorously verify those proofs, providing a framework for safer AI controllers. The work is being released as open-source and forms part of Waterloo’s broader push in transparent and responsible AI.