Introduction
While classical economic theory often assumes individuals act as purely self-interested agents, human behavior frequently deviates toward fairness and cooperation. This "fairness bias" has been extensively documented in dyadic interactions like the Ultimatum Game. However, many significant societal decisions—ranging from collective bargaining to international environmental agreements—occur within group settings. This paper investigates the evolution of fairness in a Multiplayer Ultimatum Game (MUG), examining how group decision rules and the introduction of artificial agents influence collective outcomes.
Methodology
The study utilizes a multi-methodological approach combining behavioral experiments, evolutionary game theory (EGT), and hybrid population modeling.
- Behavioral Experiments: 400 participants were recruited via Amazon Mechanical Turk to play the MUG. In this game, a Proposer offers a portion of a resource to a group of $N-1$ Responders. An offer is accepted if at least $M$ responders agree; otherwise, no one receives anything.
- Evolutionary Game Theory: An analytical model was developed to simulate strategy evolution (Pro-social, Asocial, and Generous) in a population where individuals learn through imitation.
- Hybrid Population Modeling: The EGT model was extended to include "hardcoded" pro-social agents—bots that consistently play a fair strategy—to observe their impact on the broader human population.
Results and Discussion
The Impact of Group Decision Rules
Experimental results indicate that the strictness of the voting rule ($M$) is a primary driver of fairness. As the requirement for consensus increases, Proposers tend to offer more egalitarian splits to ensure acceptance.
In groups of $N=10$, increasing the required votes from $M=1$ to $M=9$ resulted in the average proposal rising from 0.265 to 0.354. This suggests that institutional design—specifically the rules governing collective agreement—can effectively nudge individuals toward fairer behavior.
Catalyzing Fairness with Artificial Agents
A key finding of the study is the disproportionate impact of a small number of pro-social bots. In environments where "soft" rules might otherwise favor selfish behavior, the introduction of non-adaptive fair agents can shift the entire population's equilibrium.
As shown in the modeling results, introducing approximately 10% hardcoded pro-social agents is sufficient to steer over 80% of the population toward pro-social strategies in the long run. These agents serve as "social anchors," providing a stable and successful strategy for others to emulate, thereby enlarging the "basin of attraction" for fairness.
Conclusion
This research demonstrates that collective fairness is highly sensitive to the rules of engagement. Stricter agreement requirements naturally foster more egalitarian outcomes among humans. Furthermore, in hybrid populations, artificial agents can be strategically deployed to promote social cooperation. Future research should consider the impact of adaptive agents, reputation systems, and endogenous rule-making on these evolutionary dynamics.
Summary updated