The NY Times published an online article that lets you play Rock, Paper, Scissors against an AI opponent.
Fantastic. Thinking that the best way to play this thing was by being completely random about it, I immediately went to the nearest shelf of board games to find a six-sided die. Out of eleven board games, only one of them had a six sided die, and that box was still shrink-wrapped.
The computer can perform in a novice mode, where it learns as it plays, or in advanced mode, where it uses 200,000 previous games to inform its predictions on what I would do. Would the die result in a tie game?
Not really. In advanced mode it was looking to predict my moves based on prior moves, and my random element confounded it:
After 20 rounds: 9 W 6 T 5 L
After 30 rounds: 12 W 11 T 7 L
After 40 rounds: 16 W 14 T 10 L
Then I went ahead and tried this against the novice, but limited it to 20 rounds (I have to work at some point.) I was surprised to see that I had a very early lead, though with a majority of ties. By round 16 I was winning 6W 9T 1L, but then it won the last four rounds ending with 6W 9T 5L.
Not statistically significant, but I want to push more data through. That each round takes a couple of uninterruptable seconds didn't help much. The application also only shows the last few rounds, whereas I had hoped it would allow me to export game's full history.
But it certainly has me thinking about how I can inject some form of randomness into my gameplay.