Errors like this prove that so-called artificial intelligences really don’t understand what they’re doing.