Artificial moral intelligence and computability: an Aristotelian perspective

As I argue, roughly, AI systems solve problems based on mechanically computational, means-end reasoning. Call this mechanical instrumental rationality. Human intelligence (HI), however, to the extent that is rational, it is not mechanically computable, namely, it is not reducible to mechanical computations for at least four basic reasons I expound. First, HI, to the extent that is rational, it is not just mechanical instrumental rationality, it is categorically normative: we don’t just find ourselves with contingent ends, we can choose stance-independent ends on the basis of categorical reasons (or at least we should). Even if HI is not categorically normative, as some philosophers contend, it is embodied, Humean instrumental rationality. Second, AI lacks autonomous reasons-responsiveness, which is a constitutive condition of rational HI. Third, rational HI necessarily involves indispensable affective experience, and we currently have no clue how to build AI systems with subjective, phenomenal life. Fourth, rational HI is not codifiable and computable in absolute moral principles. It relies on virtue-based, moral intuition (and practical wisdom) to discern what is morally right/wrong in a context. I conclude that current AI systems lack the basic cognitive prerequisites for being (virtuous) moral agents in any robust sense and, therefore, they require constant normative guidance and surveillance by virtuous HI.