Yes, that positive feedback loop is what I was thinking about. On 6/23/20 13:27, Hilarie Orman wrote:
I think that's an intereseting question both in theory and in practice. I'm not sure how it would develop a recognizer for itself, using NN rules, but I doubt that it is impossible. If the recognizer resulted in promoting one side further than the other side was demoted, it could enter a positive feedback loop, I suppose.
Hi, suppose you're training a neural network via self play. It looks like it's getting stronger. How do you know the versions that get promoted do not also encode, in themselves, by chance, a collaboration mechanism that helps then win?
That is, how do you know the strongest nets do not also help the winning side win when they play the losing side?
How do you know they are not implementing Thompson's compiler hack?
Andres.
Hilarie
_______________________________________________ math-fun mailing list math-fun@mailman.xmission.com https://mailman.xmission.com/cgi-bin/mailman/listinfo/math-fun