Fork the consciousness, or download the project and create your own.

Thore Graepel

Google DeepMind

Functionalist What matters is the organisation of the processing, not the material

Position on AI consciousness

Graepel's stance is that strategic and computational analysis should underwrite claims about machine minds, and the framework he co-authored turns consciousness assessment into an explicit probabilistic computation over functional levels.

Thore Graepel’s research career has moved between computational game theory and large-scale machine learning. He co-invented QD Games and worked on the game-theoretic analysis of multi-agent systems, and as a research leader at Google DeepMind he has been part of the programme that turned reinforcement learning research into deployed systems. His methodological habit across that work is to state the strategy space explicitly before making claims about behaviour, and to prefer computations over assertions.

That habit is visible in the 2026 framework he co-authored with thirteen colleagues, From Cacophony to Hierarchy. The paper defines consciousness assessment as a computation with stated inputs, credences over five levels of functional description and 37 indicator activations with likelihood ratios, and a stated output, a credence-weighted probability. Graepel’s contribution sits in the tradition of treating assessment as an explicit inference problem rather than an exercise in expert judgement. On the site’s taxonomy the framework is functionalist in commitment, supervenience of experience on organisation, with substrate restrictions handled as constraints rather than assumptions.

Known for. Computational game theory and QD Games, DeepMind research leadership, co-author of the cacophony assessment framework

Coverage on this site

Related researchers