AI must pass, in general, the “Galileo” test: even if almost all the training data repeats falsehoods, it must nonetheless see the truth
