Thou Shalt Love Thy Neighbor as Thyself When Thou Playest: Altruism in Game Theory

Jörg Rothe

Abstract

Classical game theory is built upon the foundation of homo economicus—the perfectly rational, narrowly selfish agent. However, real-world interactions often deviate from this model, favoring cooperation and social consideration. This paper surveys the mathematical integration of altruism into game-theoretic frameworks, examining how "caring for others" alters the stability, efficiency, and computational complexity of multiagent systems. By synthesizing models from noncooperative and cooperative game theory, the research highlights the trade-offs between social harmony and system performance.

Introduction

Since the seminal work of von Neumann and Morgenstern, game theory has largely focused on agents interested solely in maximizing their own utility. While this simplification allows for elegant mathematical proofs, it frequently fails to predict human behavior and leads to suboptimal outcomes, such as the "tragedy of the commons," where individual greed destroys collective resources.

As artificial intelligence systems become increasingly integrated into human society, the need for agents that move beyond raw selfishness becomes critical. The central challenge lies in mathematically defining altruism and determining how it affects standard notions of equilibrium, such as the Nash Equilibrium. This paper addresses this gap by surveying models that bridge the divide between classical theory and social reality.

Modeling Altruism: Frameworks and Methods

The research categorizes altruistic modeling into two primary branches of game theory: noncooperative games and cooperative (coalitional) games.

Noncooperative Games

In noncooperative settings, the paper explores Social Context Games, where agents are embedded within a social network. In these models, an agent's utility is not merely their own payoff but an aggregation of their neighbors' payoffs, calculated via functions such as the minimum, maximum, or average.

Another key area is Congestion Games, where an "altruism level" (represented by the parameter $\beta_i$) is introduced. Here, an agent's cost function is a linear combination of their personal cost and the total social cost. This allows for a spectrum of behavior ranging from pure selfishness to total social concern.

Cooperative Games and Coalition Formation

The focus shifts to Hedonic Games, which model how agents form coalitions. The paper details three specific degrees of altruistic preference:

  1. Selfish-First (SF): Agents prioritize their own utility, using the well-being of their "friends" only as a tie-breaker.
  2. Equal-Treatment (EQ): Agents weigh their own utility and the average utility of their friends equally.
  3. Altruistic-Treatment (AL): Agents prioritize the average utility of their friends, using their own payoff only to break ties.

Computational Complexity

A significant portion of the survey is dedicated to the theoretical computer science perspective. The author evaluates the computational difficulty of finding stable states or verifying if a system has reached equilibrium. This involves mapping these problems to complexity classes such as P, NP-hard, and coNP-complete.

Key Insights and Theoretical Results

The Price of Altruism

One of the most striking findings is that altruism does not always improve system efficiency. In certain Social Context Games, altruistic behavior can actually increase the Price of Anarchy—the ratio between the best possible social outcome and the worst-case equilibrium. This suggests that well-intentioned individual actions do not always aggregate into a socially optimal state.

Stability and Existence

In Atomic Congestion Games with affine delay functions, the paper notes that a pure Nash equilibrium is guaranteed to exist even when agents are altruistic. Furthermore, the research demonstrates that the minimum number of altruists required to stabilize a social optimum can be calculated in polynomial time, providing a roadmap for designing more stable social systems.

The Complexity Jump

Adding altruism often increases the mathematical difficulty of a problem. For instance, in hedonic games, verifying "Core Stability" (where no group has an incentive to break away and form a new coalition) often jumps from polynomial time (P) to being coNP-complete when altruistic preferences are introduced. This suggests that while altruism makes agents more "human-like," it makes the systems they inhabit significantly harder to analyze and predict.

Discussion and Future Directions

The paper identifies a fundamental tension between Hedonism (caring about one's own group) and Altruism (caring about friends outside the group). This tension is a primary driver of mathematical complexity in coalition formation.

Currently, the field suffers from fragmented models; different researchers utilize varying definitions of altruism, making cross-comparison difficult. The author advocates for a unifying framework capable of capturing these diverse variations. Additionally, while the theoretical foundations are maturing, there remains a significant gap in applying these models to real-world AI deployments.

Conclusion

Integrating altruism into game theory is more than a philosophical exercise; it is a computational necessity for the next generation of multiagent systems. While altruism introduces significant complexity and potential inefficiencies, it provides the essential framework for machines to mimic and improve upon human social interaction. Moving forward, the challenge lies in balancing the trade-offs between stability, efficiency, and social preference to create AI that can truly "love its neighbor."

Summary updated

Abstract

Game theory is typically used to model the interaction among (software) agents in multiagent systems and, therefore, is a key topic at leading AI conferences. Game-theoretic models, however, are often based on the assumption that agents are perfectly rational and narrowly selfish and are interested only in maximizing their own gains, no matter what the costs to the other agents are. This summary paper presents various ways of introducing certain notions of altruism into existing game-theoretic models in both noncooperative and cooperative games, in the hope that simulating altruistic behavior in AI systems will make AI better suit real-world applications—and thus may make the real world a better place.
BibTeX
@inproceedings{rothe2021thou,
  author = {Jörg Rothe},
  title = {{Thou Shalt Love Thy Neighbor as Thyself When Thou Playest: Altruism in Game Theory}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2021},
  doi = {10.1609/aaai.v35i17.17770},
}
Figure