Either in 1978 or around 1980 or 81, I recall reading an article on AI-based image recognition with a micro like the Altair or IMSAI. All in machine code, of course, toggled in. The guy called it Legion, because it was based on a bunch of cells all making decisions. I think it may have been a 5x5 grid of cells, each with its own algorithm for voting yes or no. He would then show it a shape (enter some data 8^), and Legion would, based on the vote, tell him something (I forget what). Then he would tell Legion the real answer. The cells that voted correctly earned points, those that voted incorrectly lost points. As a cell lost points, it would eventually cross 0, at which point it was a failure, and was cast out. A different cell (algorithm) would take its place. The idea was that hopefully, the cells would eventually populate with algorithms with a high success rate. He miscoded a test, though, and only the cells that hit 0 exactly were cast out (into pigs, perhaps), but those that jumped from +1 to -1 kept voting. A program dump showd that some of these cells had actually been performing a good service. Anyone recall (much less have) this article? It was very well written (a bit tongue in cheek)... -Miles