Intrinsic Barriers to Explaining Deep Foundation Models