Speaker: Prof. Jaan Aru, University of Tartu
Host: Mototaka Suzuki (email@example.com), University of Amsterdam.
Venue: Science Park 904, Room C0.110
Zoom link (hybrid): https://uva-live.zoom.us/j/87565344829
Title: What can neuroscience say about consciousness in AI systems?
Interactions with large language models have created the impression that these models may be conscious. Some experts and researchers also share this impression. Interestingly, there are theories of consciousness that have to contend that either already now or soon, these models will be conscious. In this talk, I will try to provide some arguments against this position. I will discuss what is known about the neurobiology of consciousness, focusing on the thalamocortical and dendritic interactions underlying consciousness. As we will see, large language models are abstractions of neurobiological complexity. The question then becomes when will they reach the appropriate level of complexity required for consciousness. Even more, is it simply about complexity, or are there further factors complicating the quest for machine consciousness? I will argue that these models give us a golden opportunity to rethink some fundamental assumptions about consciousness.