Fork the consciousness, or download the project and create your own.
Yoshua Bengio speaking at ICLR 2025

Yoshua Bengio

Mila, Université de Montréal

Integrated information Consciousness is intrinsic causal power, measured as integrated information

Yoshua Bengio Photograph by Xuthoria, CC BY-SA 4.0.

Yoshua Bengio is a Turing Award winning deep learning researcher at Mila and the Université de Montréal, and the most decorated AI scientist to treat consciousness as an engineering target. His 2017 paper “The Consciousness Prior” (arXiv:1709.08568) defines conscious states as low-dimensional representations that are selected through attention over the full state of a learning agent, and shaped by a training objective that makes them useful for predicting the near future and for acting. The formulation is deliberately global-workspace-shaped, restating Bernard Baars’s architecture as a training objective.

Bengio carried the commitment into the field’s main AI consciousness assessment. He is a co-author of the Butlin and Long indicator-properties report, examined on this site in the nineteen researcher checklist, and his stated position there is that no current system is conscious while no obvious technical barrier prevents systems that satisfy the indicators. He has paired that assessment with public warnings that building conscious AI would raise moral risks that the field is not prepared for.

The significance for this site’s subject is straightforward. The Consciousness Prior is the point where the measurement vocabulary of consciousness science, workspace, attention, broadcast, enters mainstream representation learning, and Bengio’s authority made that entry unavoidable.

Known for. The Consciousness Prior, deep learning foundations, Turing Award, global workspace-inspired representation learning

Coverage on this site

Related researchers