Radiology was one of the first fields which Geoffrey Hinton thought would be dis-intermediated by AI as early evidence seemed to clearly indicate that AI could scrutinise X-rays with greater accuracy and more consistency than a skilled human radiologists. And yet, today there are more jobs openings for these radiologists than ever before, despite widespread diffusion of AI imaging into the field. This post explains why and the main points I think will apply for many other professions - the job is more than the task (of reviewing images), AI performance significantly degrades outside test conditions (training lacks full context), legal liability (somebody got to be legally accountable) and AI is augmentative, meaning more screenings for which there is actually far more hidden demand now that supply has increased. This is a great example which fortifies the AI Optimists narrative. Must read.
