After Orthogonality: Virtue-Ethical Agency and AI Alignment

The essay challenges the conventional goal-based framework for understanding rational behavior, arguing that rational agents—both human and artificial—do not operate primarily through explicit goals. Instead, the author contends that rationality emerges from aligning actions with established practices and social contexts, rather than pursuing predetermined objectives. This critique targets what the essay calls "orthogonality," the idea that an AI's intelligence level is independent of its objectives.

The proposed alternative centers on virtue ethics, suggesting that AI alignment should focus on developing virtuous agency rather than programming specific goals. Under this framework, ethical behavior stems from cultivating appropriate dispositions and practices aligned with human values and social norms, similar to how humans develop practical wisdom. This approach reframes AI safety from a goal-specification problem to one of nurturing beneficial institutional behaviors and contextual decision-making.

The implications of this perspective are significant for AI development philosophy. If correct, it suggests current approaches to AI alignment—which typically emphasize explicit goal-setting and utility functions—may be fundamentally misguided. Instead, the framework advocates integrating AI systems into human practices and institutional contexts where virtuous behavior naturally emerges, potentially offering a more robust path to safe and beneficial AI than traditional goal-oriented methods.