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

Eric Elmoznino

Université de Montréal, Mila

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

Eric Elmoznino is a doctoral researcher at Mila and the Université de Montréal working on consciousness and machine intelligence. His research connects two ideas the field usually keeps apart, that attention selects information for its usability, and that the connection between consciousness and competence runs through that selection. His published work argues that content becomes conscious-relevant, in the functional sense, when attention selects it for maximal usefulness to the task at hand, a formulation with direct consequences for how machine systems are assessed.

Elmoznino wrote the first drafts of the global workspace implementation sections of the Butlin and Long indicator-properties report, the collaboration this site examines in the nineteen researcher checklist, specifying how multiple specialized systems, a limited-capacity workspace, and global broadcast could be implemented in current AI.

The position his work defends is that the workspace architecture is not merely an analogy for large language models and their relatives. It is an implementable specification, and the interesting question is which systems realize how much of it. That is the engineering half of the site’s measurement program, stated by one of the people drafting it.

Known for. Global workspace implementation analysis in AI, attention and usable information, consciousness and competence, indicator properties collaboration

Coverage on this site

Related researchers