General challenge for site members

[ttw]

I wrote a program for the HP50g (and an earlier but more limited) for the HP41c that would play rock-scissors-paper against a person. It would learn quickly and win around 70%+ after about a dozen games. I extended it to rock-scissors-paper-lizard-Spock but that performed the same.

I looked at the last choices the opponent made vs the history. For a 3-level history, this takes 27 memory slots. I had a system that rewarded the correct choice and penalized wrong choices. It was hard to beat. I did manage to score evenly against the program using a base 3 pseudo-random number generator with good statistics and a very long cycle (too long for my program to learn.) Running about 100 of my pseudo-random choices a second time would end up losing.