Calculus of Consent via MARL: Legitimating the Collaborative Governance Supplying Public Goods

Yang Hu, Zhui Zhu, Sirui Song, Xue Liu, Yang YuarXiv 2021 · arXiv

Abstract

The supply of public goods, such as pandemic control, often requires individuals or regions to sacrifice local interests for collective welfare, leading to a crisis of governance legitimacy. This paper applies Multi-Agent Reinforcement Learning (MARL) to computationally model the "Calculus of Consent"—a social choice theory explaining why rational agents waive liberties for the common good. We demonstrate that as regions increase their collaboration level, they transition from self-centered policies to altruistic strategies that maximize global welfare and protect vulnerable populations.

Introduction

Providing public goods like climate mitigation or pandemic management creates a fundamental tension between local economic interests and shared global benefits. Centralized authorities often struggle to justify restrictive policies to self-interested local governments whose economies may suffer. This paper addresses the need for a framework that can quantitatively reason through these trade-offs, providing a rational basis for collaborative governance. We propose that MARL can serve as a "computational reasoning" tool to justify the "Calculus of Consent," where independent regions voluntarily accept restrictions to achieve superior global outcomes.

The Intelligent Region Collaboration (IRC) Framework

We introduce the Intelligent Region Collaboration (IRC) framework, which models regions as autonomous learning agents within a MARL environment.

Architecture and Reward Mixing

The core of the IRC framework is a reward mixing ratio, $\alpha$. Each agent $i$ optimizes a weighted sum of its local reward $R_i$ and the global reward $R_g$: $$\tilde{R}_i = R_i + \alpha R_g$$ By varying $\alpha$, we simulate a spectrum of collaboration levels, from purely self-interested ($\alpha \approx 0$) to highly altruistic.

The IRC framework architecture, featuring local and global critic networks to stabilize multi-agent learning.

The model utilizes Graph Neural Networks (GNNs) to capture regional interdependencies and a decoupled critic architecture. Each agent maintains a local critic for its own value and accesses a shared global critic, allowing it to balance individual and collective impacts during policy learning.

Epidemiological Environment

The environment is simulated using a Susceptible-Infected-Hospitalized-Recovered (SIHR) model. Costs are calculated based on pandemic tolerance (medical capacity) and lockdown tolerance (economic resilience). Agents decide the proportion of incoming traffic to permit, balancing infection control against economic mobility.

Experimental Setup

We evaluate the framework in a 5-region artificial environment. Region 0 serves as the outbreak source. The remaining four regions represent combinations of high/low pandemic tolerance ($H$) and high/low lockdown tolerance ($L$). This setup allows for controlled experiments on how different regional characteristics influence collaborative behavior.

Results and Discussion

The IRC model consistently outperformed static and threshold-based baselines across all collaboration levels.

The Shift to Altruism

At low collaboration levels ($\alpha=0.01$), agents behave selfishly. Resilient regions maintain high mobility despite spreading the virus, while vulnerable regions are forced into extreme isolation. However, as $\alpha$ increases to 10.0, a significant shift occurs: resilient regions (high $L$) voluntarily enforce stricter controls to "shield" their more vulnerable neighbors (low $H$).

Mobility dynamics over time for a high collaboration level (alpha=10.0), showing coordinated stabilization of regional traffic.

Balancing the Burden

Collaboration leads to a more equitable distribution of the economic burden. Instead of one region suffering total lockdown, the burden is shared to minimize global hospitalization and maximize total utility. As $\alpha$ increased from 0.01 to 10.0, global hospitalization dropped significantly from 59.46 to 27.21, while the global reward rose from 103.89 to 108.13. Radar plots of regional actions confirm that higher collaboration results in a "balanced" area of mobility, proving that agents learn to coordinate rather than acting in silos.

Conclusion

This work demonstrates that MARL can provide a computational foundation for collaborative governance. By quantifying the benefits of cooperation, the IRC framework offers a path toward legitimizing the sacrifices required for public goods. Future work will expand the action space to include intra-regional policies like vaccination and explore more complex macroeconomic indicators to refine the cost-benefit analysis of collaborative strategies.

Summary updated

Abstract

Public policies that supply public goods, especially those involve collaboration by limiting individual liberty, always give rise to controversies over governance legitimacy. Multi-Agent Reinforcement Learning (MARL) methods are appropriate for supporting the legitimacy of the public policies that supply public goods at the cost of individual interests. Among these policies, the inter-regional collaborative pandemic control is a prominent example, which has become much more important for an increasingly inter-connected world facing a global pandemic like COVID-19. Different patterns of collaborative strategies have been observed among different systems of regions, yet it lacks an analytical process to reason for the legitimacy of those strategies. In this paper, we use the inter-regional collaboration for pandemic control as an example to demonstrate the necessity of MARL in reasoning, and thereby legitimizing policies enforcing such inter-regional collaboration. Experimental results in an exemplary environment show that our MARL approach is able to demonstrate the effectiveness and necessity of restrictions on individual liberty for collaborative supply of public goods. Different optimal policies are learned by our MARL agents under different collaboration levels, which change in an interpretable pattern of collaboration that helps to balance the losses suffered by regions of different types, and consequently promotes the overall welfare. Meanwhile, policies learned with higher collaboration levels yield higher global rewards, which illustrates the benefit of, and thus provides a novel justification for the legitimacy of, promoting inter-regional collaboration. Therefore, our method shows the capability of MARL in computationally modeling and supporting the theory of calculus of consent, developed by Nobel Prize winner J. M. Buchanan.
BibTeX
@misc{hu2021calculus,
  author = {Yang Hu and Zhui Zhu and Sirui Song and Xue Liu and Yang Yu},
  title = {{Calculus of Consent via MARL: Legitimating the Collaborative Governance Supplying Public Goods}},
  year = {2021},
  eprint = {2111.10627},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2111.10627},
}
Figure