Representations and ontologies: what tools do we need to explain deep neural networks?

Dimitri Coelho Mollo (Umeå University)

Deep neural networks have increasingly become objects of scientific and philosophical investigation in their own right. In this talk, I examine the possibilities and challenges of applying key concepts from the cognitive sciences to the domain of advanced connectionist systems. In particular, I focus on whether and in what way the notions of internal representation and cognitive ontology can play a role in the project of better understanding how today’s large language models work.