A single physical law can be written in mathematically equivalent ways that are psychologically very different: as local forces, as a minimum principle over whole paths, or as a field equation. Though these formulations predict identical results, they suggest different generalizations when extended to new domains, so a scientist's choice of formulation is not neutral.
Feynman imagines a mother counting her son's 28 blocks each day to verify none are lost; when some go missing she learns to account for them by weighing a toy box or measuring the bathwater's rise, inferring hidden blocks from indirect quantities. The story concretely shows that a conserved number need not correspond to visibly countable objects, only to a calculated total that always comes out the same.
Feynman describes the method as: first guess the law, then compute its consequences, then compare those consequences to experiment; if they disagree, the guess is wrong, regardless of the guesser's name, elegance, or intuition. No amount of beauty or authority protects a guess that fails observation.
Feynman holds that fundamental laws are strikingly simple in statement while the phenomena they generate are intricate; simplicity lives in the underlying rule, not in the observed behavior. He also notes that we cannot know in advance whether this simplicity will persist as we probe deeper, so it is an observed character of known laws, not a guaranteed feature of nature.