Neural Network Device ‘Solves’ Rock-Paper-Scissors

The game of rock-paper-scissors (RPS) has been used to solve many small disagreements, with winners and losers decided by the ‘random’ offering of hand symbols for one of the three choices listed in name of the game. Interestingly, however, what people offer up here isn’t actually random at all — or not as random as you likely believe — allowing Paul Klinger to design and implement his own tiny RPS game that wins 38% of the time. While that might not seem particularly impressive compared to the 33% win rate you’d expect when playing at random, consider how well you could in Vegas if your odds were similarly elevated!

To accomplish this AI feat, Klinger uses a three-layer recurrent neural network (RNN) trained by looking at over 80,000 games played on roshambo.me. This runs on a Microchip ATtiny1614 microcontroller, with a 3D-printed case, CR2032 battery, and other electronics. A button starts the round, and inputs are available for rock, paper, and scissors, depending on the human selection. Three LEDs correspond to the computer’s move, and as seen in the video below, it wins more than it loses!

While this doesn’t solve RPS in the same way that a computer can win or tie every game of tic tac toe, it’s an fascinating build that shows just how predictable humans are — even if we would claim that this isn’t the case!

[h/t: Reddit]