Sobre a viabilidade da representação de conhecimento moral em sistemas inteligentes: uma abordagem baseada em lógica não-monotônica

Abstract

This thesis investigates the capability of artificial intelligence (AI) systems to engage in morally significant decision-making. The general hypothesis posits that although AI is becoming ubiquitous in contexts requiring moral judgments, we cannot attribute moral responsibility to autonomous systems because moral knowledge cannot be fully represented or simulated by logic. The specific hypotheses explore various nuances of this central issue. It is argued that autonomous systems lack "moral motivation" and that, despite AI's ability to automate some human thought processes such as creativity and decision-making, it cannot generate authentic moral knowledge due to the impact of data bias and the conditional and contextual nature of knowledge. The thesis also maintains that it is not possible to establish a basis for moral knowledge without considering the element of belief, nor to artificially simulate belief to produce moral knowledge. To examine these hypotheses, the thesis utilises moral dilemmas involving complex decision-making situations, modelled through modal logic equations to assess AI's capability to resolve these dilemmas. The analysis shows that while AI can simulate responses to dilemmas based on logical premises, there is always a significant gap due to the absence of moral motivation and contextual understanding, crucial for genuinely moral decisions. Incorporating the thoughts of Stuart Mill and Jesse Prinz, the research deepens understanding of how traditional philosophical concepts can be applied to modern problems, illustrating that empathy, human moral motivation, and ethical judgement are aspects that current AI cannot replicate or autonomously incorporate. The conclusion reinforces that AI, no matter how advanced, cannot replace human moral judgement. The scientific contributions of this research are substantial, promoting the debate on AI ethics by clarifying the limits of automating moral decisions.