What ML Cannot Do

ML has three hard limits: it can learn bias from data, confuse correlation with causation, and fail on rare events with too little training data. Power without judgment is still brittle.