Deep learning is a perfectly useful engineering solution.
As a tool for basic research, particularly in the biomedical realm, it's awful. You get a system that performs pretty well most of the time, but tells you nothing of interest.
That's fine, as far as it goes, but "performs pretty well" is so much more useful in industry than "we know how X biological system works" that entire fields that are interested in the mechanics of how perceptual biological systems work and may be fixed are getting eaten alive. To our collective detriment, I think.
As a tool for basic research, particularly in the biomedical realm, it's awful. You get a system that performs pretty well most of the time, but tells you nothing of interest.
That's fine, as far as it goes, but "performs pretty well" is so much more useful in industry than "we know how X biological system works" that entire fields that are interested in the mechanics of how perceptual biological systems work and may be fixed are getting eaten alive. To our collective detriment, I think.