The failure mode for AI when it comes to cartoons would be barely irritating, whereas for medical diagnostics, it's really important. Aside from that, there's no need to be consistently better. If it's not good, run it again. AI can take 10,000 tries at something in the time it takes human experts to do one. Not super valuable when it comes to questions where your final answer has to be right, like lung cancer. Extremely valuable when only one of your answers has to be right and you just throw the rest away, like a few seconds of a cartoon.