Artificial Intelligence and the philosophical problem of scientific explanation: a methodological re

Robots are often used to discover and test mechanistic hypotheses on intelligent and adaptive behaviours of living systems. These studies – which exemplify the so-called “understanding by building” approach to theoretical modelling in AI and in the cognitive (neuro)sciences – raise a number of crucial epistemological and methodological problems, which notably concern (1) the relationship which must obtain between a theoretical hypothesis and a machine implementation of it, and (2) the nature of “good” scientific explanations of intelligent and adaptive behaviours. This talk will focus on the latter problem and on the relationship it holds with the former one. After a brief discussion on the main philosophical theories on the nature of mechanistic explanation in the cognitive neurosciences, it will be argued that simulation studies in AI and biorobotics presuppose an epistemic conception of scientific explanation which posits an essential connection between explanation and prediction. This conception has been widely criticized in recent years by appeal to the theoretical and practical difficulty of predicting the behaviour of living systems in complex environments. These criticisms will be undermined by reference to the idealized character of AI explanations and of the related explanation requests.