AI operates across systems at a pace no human oversight function can match. So who's accountable when it goes wrong, asks Paul Loftus.