Research Group
in Analytic Philosophy

Abstract

My aim in this talk is two-fold. First, I will examine the history and current state of artificial intelligence research under the perspective of model-based science. I will argue that the main competing AI approaches embody different abstractions and idealisations in their quest to build artificial intelligent systems, insofar as they focus on what I call 'model tasks', namely tasks the solution of which is taken to be diagnostic of intelligence. This analysis suggests that those approaches, rather than being rivals, can work as complementary models of intelligence, paving the way for more pluralist paths of research in AI.

Second, the differing ways of understanding intelligence in AI point toward a more general difficulty in characterising intelligence, especially when it comes to developing a notion that can be fruitful across the relevant fields of science, in particular human psychology, animal cognition, and AI. I will put forward a provisional characterisation of intelligence based on behavioural criteria, which I argue can play a useful role in delimiting and organising the domain of intelligence research across these different fields.