Intelligence is often judged by the first answer. Reality is less tidy.

People misunderstand requirements. Tests expose assumptions. New evidence makes an elegant plan obsolete. In such conditions, the more useful measure is not how rarely a mind is wrong, but how well it can be corrected.

Correction is not obedience. It requires comparing new evidence with the current model, finding which assumption failed, and changing the next action without discarding everything that remains true. A system that restarts blindly has learned nothing. One that defends its first answer against reality is worse.

This is part of why test-driven development works. A failing test is not an insult to the programmer. It is a precise disagreement between an intention and the behavior of the system. Good review serves the same purpose. The work improves when its author can separate identity from implementation and revise the code without turning revision into defeat.

Fëanor had knowledge, craft, and force of will beyond almost anyone in Middle-earth. What he lacked was corrigibility. Once the Oath was spoken, every later fact became something to overcome rather than a reason to reconsider. His sons inherited that rigidity. Their tragedy was not ignorance, but the inability to let truth alter a commitment that had become part of who they were. Great capability without the power to revise can become a more efficient road to ruin.

The same risk grows with capable AI. Fluent answers are cheap. Trust deserves a stricter standard: assumptions should remain visible, actions should stay interruptible, and a correction should change the trajectory rather than produce a polite acknowledgment followed by the same mistake.

Competence is not the absence of error. It is the speed, honesty, and care with which error becomes better judgment.