It has no ability to ground output in causality. It does not understand what causes humans to learn. It’s a correlation and randomness engine.