A corporate spat between two tech giants last week was just the latest volley in a conflict over whether dangerous mistakes are being made in A.I. training.