Abstract

The contribution emphasises the topicality of Kant’s thought, whose Copernican revolution highlighted two distinct fundamental capacities of the human mind: the first, which refers to the table of categories and the principle of causality, presupposes time, that is, a succession and asymmetry, given the necessary antecedence of the cause with respect to the effect. For the second, categorisation, i.e. the correct construction of typologies, with the specification of their boundaries, which makes it possible to establish what is part of it and what is excluded from it, the temporal contiguity of certain characteristics may perhaps be important, but not their order. And while the detection of a causal structure generally presupposes some form of classification, the reverse does not seem to be true: to classify does not require a cause-effect relationship. We are thus faced with two different capacities that natural intelligence, as developments in neuroscience point out, manages to make coexist without any problems, integrating them. The distinction between these two processes and the resulting skills and competences is crucial today in order to better understand artificial intelligence, at the basis of which there are two antithetical approaches: – the first consists in building intelligence from the top down, i.e. imitating its highest and most refined product: conscious and rational reasoning, which can be formalised using logic and its symbols and the kantian table of categories; – the second proceeds instead from the bottom up, starting with knowledge of brain processes and their simulation, using machine learning and deep learning techniques