“Even if somebody can give you a reasonable-sounding explanation for their actions, it probably is incomplete, and the same could very well be true for AI”