Philosophy and Epistemology of Deep Learning

Causality is the most important topic in the history of western science, and since the beginning of the statistical paradigm, its meaning has been reconceptualized many times. Causality entered into the realm of multi-causal and statistical scenarios some centuries ago. Despite widespread critics, today deep learning and machine learning advances are not weakening causality but are creating a new way of finding correlations between indirect factors. This process makes it possible for us to talk about approximate causality, as well as about a situated causality. Full article

This contribution tentatively outlines the presumed conceptual duality between the issues of incompleteness and incomprehensibility—The first being more formal in nature and able to be declined in various ways until specified in the literature as theoretical incompleteness. This is theoretical and not temporary, which is admissible and the completion prosecutable. As considered in the literature, theoretical incompleteness refers to uncertainty principles in physics, incompleteness in mathematics, oracles for the Turing Machine, logical openness as the multiplicity of models focusing on coherence more than the optimum selections, fuzziness, quasiness, e.g., quasi-crystals, quasi-systems, and quasi-periodicity, which are intended as the space of equivalences that allow for coherent processes of emergence. The issue of incomprehensibility cannot be considered without reference to an agent endowed with cognitive abilities. In this article, we consider incomprehensibility as understood here as not generally scientifically explicable, i.e., with the available knowledge, as such incomprehensibility may be temporary, pending theoretical and technological advances, or deemed to be absolute as coincident with eventual definitive, theoretical non-explicability, and incomprehensibility. We considered the theoretically incomprehensibility mostly in three main ways: as the inexhaustibility of the multiplicity of constructivist reality as given by the theoretically incomprehensible endless loop of incomprehensible–comprehensible, and by existential questions. Moreover, theoretical incomprehensibility is intended as evidence of the logical openness of both the world and of understanding itself. The role of theoretical incomprehensibility is intended as a source of theoretical research issues such as paradoxes and paradigm shifts, where it is a matter of having cognitive strategies and approaches to look for, cohabit, combine, and use comprehensibility and (theoretical) incomprehensibility. The usefulness of imaginary numbers comes to mind. Can we support such research for local, temporary, and theoretical incomprehensibility with suitable approaches such as software tools, for instance, that simulate the logical frameworks of incomprehensibility? Is this a step toward a kind of artificial creativity leading to paradigm shifts? The most significant novelty of the article lies in the focus on the concept of theoretical incomprehensibility and distinguishing it from incomprehensibility and considering different forms of understanding. It is a matter of identifying strategies to act and coexist with the theoretically incomprehensible, to represent and use it, for example when dealing with imaginary numbers and quantum contexts where classical comprehensibility is theoretically impossible. Can we think of forms of non-classical understanding? In this article, these topics are developed in conceptual and philosophical ways. Full article

Deep learning and other similar machine learning techniques have a huge advantage over other AI methods: they do function when applied to real-world data, ideally from scratch, without human intervention. However, they have several shortcomings that mere quantitative progress is unlikely to overcome. The paper analyses these shortcomings as resulting from the type of compression achieved by these techniques, which is limited to statistical compression. Two directions for qualitative improvement, inspired by comparison with cognitive processes, are proposed here, in the form of two mechanisms: complexity drop and contrast. These mechanisms are supposed to operate dynamically and not through pre-processing as in neural networks. Their introduction may bring the functioning of AI away from mere reflex and closer to reflection. Full article

Locked and unlocked strategies are at the center of this article, as ways of shedding new light on the cognitive aspects of deep learning machines. The character and the role of these cognitive strategies, which are occurring both in humans and in computational machines, is indeed strictly related to the generation of cognitive outputs, which range from weak to strong level of knowledge creativity. I maintain that these differences lead to important consequences when we analyze computational AI programs, such as AlphaGo, which aim at performing various kinds of abductive hypothetical reasoning. In these cases, the programs are characterized by locked abductive strategies: they deal with weak (even if sometimes amazing) kinds of hypothetical creative reasoning, because they are limited in what I call eco-cognitive openness, which instead qualifies human cognizers who are performing higher kinds of abductive creative reasoning, where cognitive strategies are instead unlocked. Full article