Recognising artificial mathematical intelligence
Abstract: The study of intelligence in animals has two potential pitfalls. First, the setting for experiments may not be suitable for animals to exhibit their characteristic intelligence. Second, the observed behaviour may be misrepresented in terms of intelligence. Often this involves ascribing unjustifiably high level of intelligence to animals. In this talk, my focus is on artificial intelligence (AI) and how its study could avoid related pitfalls. In particular, I focus on artificial mathematical intelligence. Related to the first pitfall, I ask whether mathematical AI applications are systematically deprived of the chance to exhibit their characteristic intelligence (if any). Related to the second pitfall, I ask whether there is a danger in misrepresenting the processing of current mathematical AI applications in terms of intelligence. I argue that unlike in animal studies, the second pitfall is significantly more prominent than the first one. However, I argue that the first pitfall may become more serious as mathematical AI applications reach a sufficiently high level. As a proposed solution to both present and future problems, I propose a community-based approach to recognising artificial mathematical intelligence.
The talk will be hybrid in Oslo and on Zoom.
Contact Laura.Crosilla @ ifikk.uio.no for a Zoom-invite.