An essay published in The Gradient makes a provocative case: rational people don't have goals, and rational AIs shouldn't either. The author argues that human actions are rational not because they aim at some final end, but because they align with practices—networks of actions and action-dispositions that shape how we behave over time.
This view directly challenges common assumptions in AI alignment. If rationality is about cultivating the right practices rather than pursuing goals, then alignment work should focus on building AI systems with virtuous dispositions, not just specifying reward functions or objectives. The essay frames this as a virtue-ethical alternative to the dominant goal-based paradigm.
The argument is philosophical rather than technical, and it does not offer implementation details. But it suggests that the field's focus on goals may be part of the problem, and that a shift toward practice-based agency could open a different path for safe AI development.