What does it really mean to say that an AI system understands, reasons or thinks?

In this episode of Ethics and Innovation by Oxford+ brought to you by Equinox, host Susannah de Jager speaks with Dr Raphaël Millière, Associate Professor in Philosophy at the University of Oxford, about the deep conceptual questions behind today’s AI systems. Raphaël explains why strong performance on a benchmark does not always prove real competence, why language models learn in a profoundly different way from humans, and why common sense remains such a revealing fault line in AI research.

The conversation moves from consciousness and anthropomorphism to the geopolitical stakes of sovereign AI. Recent research from Anthropic on global workspace-like structures in language models shows why questions about internal representations are becoming more urgent, while the European Commission’s 2026 technological sovereignty package underlines how AI capability is now tied to economic resilience, national security and democratic control.

(00:00) - Welcome to Oxford+
(01:01) - Raphaël Millière’s Work in Philosophy of AI
(03:24) - Performance, Competence and AI Evaluation
(07:01) - Testing Analogy and Reasoning in AI Models
(08:56) - Why AI Learns Differently from Humans
(13:13) - Language, Embodiment and the Path to Intelligence
(17:02) - Common Sense and the Jagged Edge of AI
(19:39) - Should We Build Human-Level AI?
(22:29) - The Utopian and Arms Race Cases for AI
(28:31) - Sovereign AI, Europe and Strategic Dependence
(34:01) - Public Investment, Private Incentives and Democratic Governance
(37:21) - Moving Beyond Polarised AI Debates

Raphaël Millière: Raphaël Millière is an Associate Professor in Philosophy at the University of Oxford, a Fellow of Jesus College and affiliated with the Institute for Ethics in AI. His research sits at the intersection of philosophy, cognitive science and artificial intelligence, with a focus on the capacities and limitations of contemporary AI systems, including language understanding, reasoning, planning, consciousness and AI safety. He also holds an AI2050 Fellowship from Schmidt Sciences, and his recent work includes The Philosophy of Language Models, published in Philosophy Compass.

Connect with Raphaël on LinkedIn

Susannah de Jager: Susannah is a seasoned professional with over 15 years of experience in UK asset management. She has worked closely with industry experts, entrepreneurs, and government officials to shape the conversation around domestic scale-up capital.

Connect with Susannah on LinkedIn and Subscribe to the Oxford+ Newsletter for Exclusive Content

Oxford+ is hosted by Susannah de Jager and supported by Equinox.

Produced and Edited by Story Ninety-Four in Oxford.