Computer overlords

Oh, no. NPR, you did it again. I've said before that journalists make terrible science reporters, but this is just bad. A computer beat two Jeopardy champions pretty easily. It was an interesting game to watch and an amazing feat of technology, but Martin Ford, a computer programmer and writer decided that winning at a game show is enough evidence to show that computers can take over medicine, emphasizing on radiology. He said, "Radiologists basically focus on looking at visual images from medical devices — things like X-rays or CAT scans, that type of thing. Now, machines are getting much better at analyzing that type of visual information."

Arrgh! Radiology is not just looking at images. It isn't even just pattern recognition. It's putting together information from a vast number of sources (patient history, previous scans, pathology, physical properties of the imaging technique, variances in image acquisition, positioning of patient, etc.) To be an effective radiologist, you can't just look at every image with the same formula (if you can, who needs a radiologist?). That's a start, but personal experience at interpreting images and intuition about what the image and patient history points to adds a lot. It's that intuition that looks more closely at an image or orders a different view that makes the diagnosis.

I don't doubt that computers will eventually able to do this, but just because one won at Jeopardy, doesn't mean that it can start replacing radiologists. In reality, if computers get good enough to accurately interpret all types of images (at least with the same accuracy of radiologists), then they'll be good enough to take an effective history, perform a physical exam, order and interpret appropriate lab tests, perform procedures or even surgery and prescribe medications. The only medical specialty that would be safe is psychiatry (that is, until it starts using real science also). I don't see that happening in my lifetime, and if it does, we'll be living in a pretty different world.