"The idea is not that AI should replace human involvement"
Wink, wink. Maybe it's not the intent, but that's how it seems to turn out. The computer is only as good as the people who tell it what to do. Eventually it gets to the point where the computer is training the people. What happens when it decides to go on extended holiday, even quit? 🙂
Neural networks (the principle behind much of pattern recognition) are, rather, only as good as the data sets used to "train" the networks (and the design of the networks). The networks "learn" to recognize features by themselves -- the are not
instructed to look for this, or look for that. Generally a portion of a very large data set (e.g. photomicrographs showing, say, prostate cancer or no prostate cancer are used to train the network. The other half of the photos (again with known diagnoses) are used to test the trained network. The results are direct -- you can see how well the neural net does compared to human experts presented with the same test set. I'm oversimplifying, but that's the gist.
Papers on neural-nets and machine learning are appearing more frequently with regard to PCa diagnosis, a good application because it is all based on visual patterns (histopathology). The invention of neural networks was a big step forward in AI. In the best cases, these systems can outperform experts in some fields because the networks incorporate "knowledge" that the experts aren't aware of, or don't apply as consistently or as well, especially in the tough cases.
For example, if you're training a pattern-recognition system to distinguish cats from rabbits, it will have incorporated the "fact" that "long pointy things on the head favors rabbit", by having learned that rabbits have something there, and in all positions, with ears up and back, and in all body orientations. But it has never been given, e.g., a formula like "if the ears are x times as long as this other measurement, score x points in favor of rabbit" It's also learning all sorts of "facts" about
body and limb shapes and proportions without being explicitly instructed how. It's knowledge is distributed through the network. A new era: artificial intelligence and machine learning in prostate cancer
 Commentary: Automated diagnosis and gleason grading of prostate cancer – are artificial intelligence systems ready for prime time?
Post Edited (DjinTonic) : 1/10/2020 7:37:08 AM (GMT-7)