Potenziamento morale mediato dall'intelligenza artificiale: verso un modello di deliberazione morale assistita.

The growing integration of artificial intelligence into human cognitive and decision-making processes requires us to consider not only the risks arising from the progressive delegation of functions traditionally attributed to human beings, but also the opportunities these technologies may offer for strengthening our deliberative capacities. Against this background, this thesis investigates whether, and under what conditions, artificial intelligence can serve as an instrument of moral enhancement without compromising individual autonomy, freedom, and agency. After tracing the development of artificial intelligence up to the contemporary generative paradigm and examining its principal technical, epistemological, and anthropological risks, the thesis turns to the contemporary debate on human moral improvement, focusing in particular on the proposals advanced by Thomas Douglas, Ingmar Persson and Julian Savulescu, as well as on John Harris’s critical response. An analysis of the limitations of the biomedical paradigm provides the basis for shifting the object of intervention from the content of moral choice to the process through which such choices are formed. On these grounds, the thesis advances a procedural conception according to which making an agent morally better does not require directing them towards a predetermined outcome, but rather strengthening the capacities that enable them to acquire and assess relevant information, compare reasons and alternatives, identify biases and inconsistencies, adopt different perspectives, and critically revise their own judgments. The aim thus becomes that of improving the conditions under which deliberation takes place while preserving the agent’s ability to recognize themselves as the author of the final decision. Artificial intelligence is subsequently examined as a possible external instrument of deliberative assistance, with particular attention to the conditions distinguishing genuine from merely apparent improvement and to the risks associated with delegation, nudging, and value alignment. A comparison between a subjectivist configuration, represented by Julian Savulescu and Hannah Maslen’s agent-tailored moral AI, and an objectivist configuration, represented by Michał Klincewicz’s moral reasoner, ultimately leads to the original proposal developed in this thesis: the Constructivist Moral Advisor. The constructivist model seeks to occupy a position between two opposing risks: that of a system that merely reflects the user’s antecedent moral convictions and that of an artificial intelligence that purports to determine the morally correct answer independently of the deliberative process. Its role is neither to replace the agent nor to prescribe what ought to be done, but to structure a pluralistic, critical, and non-coercive deliberative space in which reasons, principles, consequences, and alternative perspectives can be made explicit, compared, and subjected to justification. Artificial intelligence is thereby reconceived as an ethical interlocutor and a form of deliberative support capable, at least in principle, of contributing to more reflective and publicly justifiable moral reasoning, while leaving ultimate responsibility for, and authorship of, moral judgment with the human agent.