
Taming AI
Computer scientist Matt Jones of Swansea University argues that artificial intelligence needs to be tamed rather than merely restrained, using analogies from dragon-training films to fire and cars to frame the problem. He walks through design approaches such as Human-Centred AI and Responsible AI, explaining why anthropomorphizing systems misleads users and why concepts like trustworthiness, legibility and explainability matter in practice. Case studies include the COMPAS sentencing algorithm, the Boeing 737 MAX software failures, and the Chernobyl disaster as a lesson in guardrail design, alongside a discussion of the wolves-versus-dogs explainability problem and the risks of AI sycophancy. Jones also addresses Asimov's laws, constitutional AI, automation bias, and what both individuals and governments can do to keep humans meaningfully in the loop. Recorded at Barnard's Inn Hall, London, as part of Gresham College's public lecture series, the talk closes by considering what regulation and governance should look like as AI systems grow more powerful.