Le spiegazioni algoritmiche della mente alla prova del" deep learning"

An important aspect of Guglielmo Tamburrini’s philosophical work concerns the role of algorithmic explanations in the study of the mind. For an algorithmic system to count as an explanation of a cognitive phenomenon, it must be designed as the implementation of a theoretical model of the phenomenon to be explained. Thus, mere behavioral equivalence is not sufficient; additional constraints deriving from the assumptions of the theory must be satisfied. The recent successes of AI are the result of an approach that clearly prioritizes technological and engineering aspects over more theoretical considerations: such systems are not intended to advance our understanding of biological cognition. Nevertheless, enthusiasm for these successes has sometimes led to bold conclusions about the functioning of the human mind, for example in domains such as natural language processing or visual perception. AI systems based on machine learning, precisely because of their exclusively applicative nature, are not subject to cognitive constraints and therefore cannot be regarded as models of natural cognition. This does not make machine learning techniques entirely irrelevant from a cognitive standpoint. Their use in cognitive modeling, however, raises new questions about the nature of algorithmic models in the study of the mind and about how such models should be employed.