An actuarial artificial intelligence for the game rock-paper-scissors

Rock-paper-scissors is a simple game played by two players who use hand gestures that resemble a 'rock' (a fist), a piece of 'paper' (a flat hand) or a pair of 'scissors' (index and middle fingers in the shape of a V). Each player plays a gesture at the same time, with the scoring as follows: 'rock' beats 'scissors'; 'paper' beats 'rock'; and 'scissors' beats 'paper'. If the game is played only once, no dominating strategy exists and the result is down to chance. If the game is played multiple times the chance element reduces as a player's next gesture is influenced by two factors: the previous result (win, loss or draw) and the previous gesture ('rock', 'paper' or 'scissors'). Previous attempts by RoShamBot to create an artificial intelligence (or AI) for rock-paper-scissors have relied on simple frequency analysis (i.e. the user has played 'rock' the most, therefore the AI must play 'paper' next) or history matching (i.e. the AI matching the last four rounds to a large database of previous games and determining what gesture the user will play next and then countering that gesture). In an actuarial spirit, I take a stochastic approach and instead of determining the next gesture with certainty, I flex the probabilities for 'rock', 'paper' and 'scissors' and then let my AI randomly choose one. The probabilities are flexed based on previous results and previous gestures.