Researchers at a Seattle A.I. lab say they have built a system that makes ethical judgments. But its judgments can be as confusing as those of humans.