Technical Safety & Governance Lab

Neural networks can make predictions about complex systems, sometimes with striking accuracy, while leaving us unable to explain what they have learned. If we come to rely on them more and more, we need to ask what kind of understanding we are gaining by studying them—and what we may lose through this lack of understanding.

At TSG Lab, we study how these models work, how to test them, and how people can remain in control as they are given more responsibility. Understanding AI systems is a scientific question in its own right. It also matters for deciding when an AI system can be trusted and how it can be governed.

The Lab is based in the Department of Engineering Science at the University of Oxford. We are also part of the Oxford Martin AI Governance Initiative.