Psychologists studying expertise noticed something odd: when you ask a chess grandmaster why they moved a piece, they give you reasons — but those reasons are often reconstructed after the fact, not the actual cause of the move. Jonathan Schooler called this 'verbal overshadowing' — the act of putting an implicit skill into words can actively degrade it. This is worth sitting with when you reach for AI-assisted explanation. When a model translates your half-formed technical intuition into crisp, confident prose, it isn't just cleaning up your syntax. It's replacing an analog signal — one that still contained genuine uncertainty and open edges — with a digital one that sounds finished. The philosopher of mind Hubert Dreyfus spent his career arguing that the deepest human competencies live below the threshold of explicit representation, and that systems which can only operate on explicit rules will always miss something. He was mostly talking about early AI. He's still right about what we risk losing when we outsource articulation too early in the thinking process.

In the last week, when did you let a generated explanation stand in for an understanding you hadn't fully earned yet?

Drawing from Phenomenology / Philosophy of Mind — Hubert Dreyfus