The Jowett Society (Friday - Week 2, MT26)

Abstract: Cognitive scientists have developed an impressive array of tests for consciousness (C-tests). But can these tests legitimately be applied to artificial intelligence? Going against a prominent trend in the literature, I argue not. C-tests are well-supported as tests for consciousness in humans. But when a test is applied to a novel population, it stands in need of validation. There are two ways to validate C-tests for AI. On the theory-driven approach, we start with an explanation of consciousness then use this to justify trusting the test across novel populations. On the iterative natural kind approach, we bracket explanatory claims and make incremental inferences from a consensus population to adjacent populations. I argue that neither approach works for AI. Consequently, the results of any putative test for artificial consciousness should not be trusted. Many commentators have argued that if an AI passes a C-test, we have a moral imperative to take that result seriously. I argue that this moral argument is misguided and suggest an alternative way forward.

Please note: The seminar runs from 15:00 to 17:00, drinks are held in the Faculty Hub afterwards.

Registration: If you do not hold a university card, please contact the seminar convenor or admin@philosophy.ox.ac.uk at least two working days before a seminar to register your attendance.

_{Jowett Society Organising Committee: Joshua Loo, Zachary Lang, Xavier Morales Zayas and Sepehr Razavi | Jowett Society Website}