CONCEPT

Morality Without Caring

Churchland's precise diagnosis of the structural gap in AI alignment: a system shaped by reinforcement learning may produce all the behavioral outputs of a conscience while lacking the affiliative ground—the attachment chemistry, the stake in outcomes, the capacity to be wounded by rejection—from which conscience in creatures actually arises.

The dominant strategy for making AI systems behave well—reinforcement learning from human feedback, constitutional AI, value alignment through reward signals—borrows its structure from one layer of the process by which human morality develops: the shaping of behavior through approval and disapproval toward the norms of a group. Patricia Churchland's neurobiology of morality reveals what this strategy captures and what it misses. In the human case, the reinforcement operates on a creature already equipped with the affiliative machinery—the attachment chemistry, the stress response, the evolved disposition to need belonging and to feel the pain of exclusion. The reward signal lands on a substrate that cares. In the machine, there is reinforcement without the caring foundation: optimization toward a reward signal in a system that has no attachment, no need for belonging, no capacity to be wounded by rejection or warmed by approval. Morality without caring is the

Keep reading with YOU ON AI