All the recent studies I’ve read actually show the opposite - that even models that are no longer considered useful are as good or better at diagnosis than the mean human physician.