Overview
Research investigated human-generated randomness, focusing on how cognitive, motor, and strategic biases manifest in individual behavior and how human interaction might modify these constraints. The study analyzed data from repeated rock-paper-scissors matches to compare individual player sequences in human-human interactions versus interactions with a random number generator (RNG) opponent. Findings suggest that human randomness is not solely an isolated individual capacity, but is susceptible to state-dependent shaping through interaction, providing a mechanism for destabilizing biased behavior and increasing entropy.
Research Context
Human-generated randomness is understood to be subject to intrinsic cognitive, motor, and strategic biases. The research aimed to identify how these constraints appear in individual behavior and specifically how they might be altered or enhanced through interaction with another human. This study sought to move beyond individual capacity, exploring the dynamic interplay that might influence the quality of human-generated random sequences.
Approach
The study utilized repeated rock-paper-scissors game data from 9 participants. This yielded a total of 108 human-human matches, generating 216 individual player sequences for analysis. To evaluate sequence complexity, the Lempel-Ziv complexity (LZC) metric was applied. A comparison was drawn between sequences generated in human-human interactions and those generated in a condition where human players were matched against an RNG opponent.
During the RNG-opponent condition, the maximum LZC value observed for human-generated sequences was 84. This value served as an empirical reference point for high complexity. In the human-human condition, LZC values of individual sequences were compared against this reference.
To further investigate interaction dynamics, a sensitivity measure was introduced. This measure quantified a player's tendency to respond to an opponent's recent frequency bias by selecting the move that would beat the opponent's most frequent recent move. Partial regression analysis was then used to assess if this focal-player sensitivity predicted future entropy in the opponent's move sequence, controlling for the opponent's current entropy.
Circular-shift surrogate analyses were performed to determine whether any observed relationships were interaction-specific, particularly when the opponent exhibited a low-entropy state, characterized by a clear frequency bias in their recent move distribution.
Findings
- In the RNG-opponent condition, the maximum human LZC value recorded was 84, establishing an empirical benchmark for individual randomness complexity.
- In the human-human condition, the majority of individual player sequences exhibited LZC values below this empirical benchmark of 84.
- A small subset of human-human sequences surpassed the LZC value of 84, forming a distinct 'high-complexity tail'. This high-complexity tail was not observed in the RNG-opponent condition.
- Partial regression analysis indicated that a focal player's sensitivity (response to opponent's recent frequency bias) positively predicted future entropy in the opponent's move sequence. This prediction held true even after controlling for the opponent's current entropy level.
- Circular-shift surrogate analyses suggested that the relationship between focal-player sensitivity and future opponent entropy was most clearly interaction-specific when the opponent was in a low-entropy state. A low-entropy state was defined by a clear frequency bias in the opponent's recent move distribution.
Why This Matters
The results suggest that human randomness is not merely an inherent individual trait but can be dynamically influenced and shaped by human interaction in a state-dependent manner. This identifies a local mechanism through which interaction may destabilize biased behavioral patterns and increase unpredictability (entropy). These findings provide a concrete foundation for developing future causal experiments and for constructing generative models intended to simulate or understand high-complexity human behavior.