Rock-Paper-Scissors study: Human randomness improves through interaction with another player

A new study from Song-Ju Kim, Shoma Ohara, and Hiroaki Kurokawa, posted on arXiv, explores how human-generated randomness in Rock-Paper-Scissors (RPS) can be modified through interaction. Analyzing 9 participants across 108 human-human matches and 216 individual sequences, the team used Lempel-Ziv complexity (LZC) to quantify randomness. In a control condition where players faced a random number generator (RNG), the maximum human LZC was 84. However, in human-human matches, a small but distinct tail of sequences exceeded that threshold, indicating that interaction can push players beyond their individual cognitive limits for randomness.

The researchers introduced a sensitivity metric that tracks whether a player responds to the opponent's recent move frequency bias by choosing the counter-move. Partial regression showed that a focal player's sensitivity positively predicted future entropy in the opponent's moves, even after controlling for current entropy. Circular-shift surrogate analyses confirmed this effect was strongest when the opponent's entropy was low (i.e., they had a clear bias). The findings reveal a local mechanism where interaction destabilizes biased behavior and increases entropy, providing a concrete basis for future causal experiments and generative models of high-complexity human behavior.