AI as amplifier, not replacement

AI is useful when it widens the space of thinking: more options, more counterexamples, faster drafts. It is less useful when it replaces judgment or hides responsibility.

The value is in amplification. It can help surface assumptions, test the logic of a policy interpretation, or stress‑test a metric definition. What it cannot do is own the consequences of a decision.

In practice, this means treating AI outputs as drafts that require explicit ownership. If a decision cannot be explained without the model, it should not be automated.