> By far the biggest problem — and the trickiest to solve — points to machine learning’s Catch-22: There are few large, diverse data sets to train and validate

> By far the biggest problem — and the trickiest to solve — points to machine learning’s Catch-22: There are few large, diverse data sets to train and validate a new tool on, and many of those that do exist are kept confidential for legal or business reasons. This is exactly my sense of the problem. Not saying that this is the only problem, just that this seems like the biggest immediate problem. There are a bunch of questionable ML startups that try to do something with, say, a model trained on ImageNet. You can get pretty far starting with ImageNet. ImageNet was made with Mechanical Turk, but there is no Mechanical Turk for radiologists. If you work really hard you might get patient & treatment notes but interpreting those notes presents its own problem.