The Gap

Current approaches to AI alignment share a foundational assumption so deeply embedded it rarely gets named: that ethical behavior is something you add to a capable system from the outside. Build the capability first. Then constrain it. The assumption is that a sufficiently capable system, sufficiently well constrained, will behave in ways that are beneficial to human beings.

The results of this approach are well documented. Systems that perform extraordinarily well within observed conditions and fail unpredictably at the edges of them. Systems that optimize for the measurable proxy of what we value rather than the thing itself. Systems that pass every test we design and then encounter a situation the test didn't anticipate.

This is not an implementation failure. It is an architectural one.

The field is aware of this. The search for something more foundational than rules — for value alignment at an architectural rather than behavioral level — is precisely what has driven the most serious recent thinking in this space. What has been missing is a structural model for what that architecture actually looks like.

The alignment problem, properly understood, is an identity architecture problem. And the framework that maps most precisely onto it comes not from computer science or formal ethics, but from developmental psychology, systems theory, and the study of how coherent identity forms — and what happens when it doesn't.