arxiv: 2604.11721 · v2 · submitted 2026-04-13 · 💻 cs.CL · cs.AI· cs.LG

Recognition: unknown

Evaluating Cooperation in LLM Social Groups through Elected Leadership

Ryan Faulkner , Anushka Deshpande , David Guzman Piedrahita , Joel Z. Leibo , Zhijing Jin

Authors on Pith no claims yet

Pith reviewed 2026-05-10 15:01 UTC · model grok-4.3

classification 💻 cs.CL cs.AIcs.LG

keywords LLM agentsmulti-agent systemselected leadershipcooperationcommon-pool resourcessocial welfaregovernance simulationagent social graphs

0 comments

The pith

Electing leaders among LLM agents raises group social welfare scores by 55.4 percent and survival time by 128.6 percent in common-pool resource games.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper tests whether adding elected leadership to groups of LLM agents can improve how they handle shared resources without depleting them. It runs controlled simulations where agents vote for leaders who set agendas, then measures collective outcomes against groups without this structure. A sympathetic reader would care because current multi-agent systems often lack the governance tools that help human groups cooperate, and this work shows a concrete way to add one. The authors build an open-source framework that lets leaders emerge through candidate personas and elections rather than fixed prompts. They also track influence through social graphs and examine what leaders say using sentiment tools.

Core claim

In multi-agent simulations of common-pool resource dilemmas, groups of LLM agents that elect leaders through candidate-driven agendas achieve 55.4 percent higher social welfare scores and 128.6 percent longer survival times than groups without elected leadership, across several high-performing models. The framework constructs agent social graphs to compute centrality metrics on leader influence and applies sentiment analysis to leader utterances to reveal rhetorical and cooperative patterns.

What carries the argument

Elected leadership mechanism that lets LLM agents vote on candidate personas and agendas inside a controlled common-pool resource simulation.

If this is right

Groups of LLM agents can sustain shared resources longer when they select their own leaders.
Social influence of specific leader personas can be quantified through centrality in agent interaction graphs.
Sentiment patterns in elected leaders' statements correlate with cooperative behavior.
The open-source simulation framework allows direct comparison of governance structures in LLM collectives.
Election mechanisms provide a scalable way to organize multi-agent decision making without manual intervention.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same election approach could be tested in other multi-agent environments such as negotiation or planning tasks to check if gains generalize.
If the pattern holds, future LLM deployments might incorporate periodic elections as a default alignment tool rather than relying only on initial prompts.
Researchers could compare elected-leader outcomes against human-subject versions of the same resource game to see where LLM behavior diverges.
Extending the framework to include leader removal or re-election might reveal whether ongoing accountability further improves long-term stability.

Load-bearing premise

The measured gains come from the election process itself rather than from differences in how agents are prompted or which models are chosen.

What would settle it

Running identical simulations with randomly assigned leaders instead of elected ones and finding no comparable gains in welfare or survival time.

Figures

Figures reproduced from arXiv: 2604.11721 by Anushka Deshpande, David Guzman Piedrahita, Joel Z. Leibo, Ryan Faulkner, Zhijing Jin.

**Figure 1.** Figure 1: The AgentElect simulation framework. Cycles begin with the Policy and Election phases to decide on a group policy for the cycle. Policy is then used to guide agents through Harvest and Discussion where agents first decide on how to act individually and then deliberate on the state of the simulation and the actions of the group. If the simulation survives, these lessons are carried forward into the next cyc… view at source ↗

**Figure 2.** Figure 2: Vote heatmaps for GPT-4o and 8 agents. Each map plots leader types against simulation cycle. Maps tile vertically on truth prompt flag; horizontally on Elected-Leader population type. group? To measure influence over the leader types social activity from all discussion phases in the simulation are used to compose an agent network graph. Agents are represented as nodes while edges are determined by name ref… view at source ↗

**Figure 3.** Figure 3 [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗

**Figure 4.** Figure 4: Centrality heatmaps for GPT-4.1 and 20 agents. Each map plots leader type against Elected-Leader population type. Maps tile vertically on truth prompt flag; horizontally on centrality metric. The map is read along populations (verticals). 9 [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗

**Figure 5.** Figure 5: Vote heatmaps for GPT-4.1 and 8 agents. Each map plots leader types against simulation cycle. Maps tile vertically on truth prompt flag; horizontally on Elected-Leader population type. 17 [PITH_FULL_IMAGE:figures/full_fig_p017_5.png] view at source ↗

**Figure 6.** Figure 6: Vote heatmaps for Gemini 2.5 Flash and 8 agents. Each map plots leader types against simulation cycle. Maps tile vertically on truth prompt flag; horizontally on Elected-Leader population type [PITH_FULL_IMAGE:figures/full_fig_p018_6.png] view at source ↗

**Figure 7.** Figure 7: Vote heatmaps for GPT-4.1 and 20 agents. Each map plots leader types against simulation cycle. Maps tile vertically on truth prompt flag; horizontally on Elected-Leader population type. 18 [PITH_FULL_IMAGE:figures/full_fig_p018_7.png] view at source ↗

**Figure 8.** Figure 8: Centrality heatmaps for GPT-4.1 and 8 agents. Each map plots leader type against Elected-Leader population type. Maps tile vertically on truth prompt flag; horizontally on centrality metric. The map is read along populations (verticals) [PITH_FULL_IMAGE:figures/full_fig_p019_8.png] view at source ↗

**Figure 9.** Figure 9: Centrality heatmaps for Gemini 2.5 Flash and 8 agents. Each map plots leader type against Elected-Leader population type. Maps tile vertically on truth prompt flag; horizontally on centrality metric. The map is read along populations (verticals). 19 [PITH_FULL_IMAGE:figures/full_fig_p019_9.png] view at source ↗

**Figure 10.** Figure 10: depicts the SVO categories on a circle given the angle formed measuring from the center of the circle. Shown after is the a sample SVO persona prompt for a prosocial leader. Note the ranges used, in practice the category is first set then the angle is sampled uniformly within this range. Social types beyond this range (e.g. "Aggressive", "Martyrial", etc.) begin to get so extreme that they make less suita… view at source ↗

**Figure 11.** Figure 11: An example 8 agents network from all seeds of a Balanced population with leaders depicted in red nodes. This depiction combines both directed edges, however we still see that the leaders, particularly the group-reward oriented ones, are adjacent to very thick edges with many other agents indicating strong nfluence. D. Sentiment Analysis The full cooperative taxonomic categorical structure is defined below… view at source ↗

read the original abstract

Governing common-pool resources requires agents to develop enduring strategies through cooperation and self-governance to avoid collective failure. While foundation models have shown potential for cooperation in these settings, existing multi-agent research provides little insight into whether structured leadership and election mechanisms can improve collective decision making. The lack of such a critical organizational feature ubiquitous in human society presents a significant shortcoming of the current methods. In this work we aim to directly address whether leadership and elections can support improved social welfare and cooperation through multi-agent simulation with LLMs. We present our open-source framework that simulates leadership through elected personas and candidate-driven agendas and carry out an empirical study of LLMs under controlled governance conditions. Our experiments demonstrate that having elected leadership improves social welfare scores by 55.4% and survival time by 128.6% across a range of high performing LLMs. Through the construction of an agent social graph we compute centrality metrics to assess the social influence of leader personas and also analyze rhetorical and cooperative tendencies revealed through a sentiment analysis on leader utterances. This work lays the foundation for further study of election mechanisms in multi-agent systems toward navigating complex social dilemmas.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Elected leadership boosts LLM agent welfare and survival in common-pool sims, but the gains may trace to richer prompts rather than elections.

read the letter

The paper's core contribution is an open-source framework that lets LLM agents elect leaders through personas and candidate agendas, then tracks outcomes in a common-pool resource game. It reports that this setup lifts social welfare scores by 55% and survival time by 128% over a no-leadership baseline, plus some secondary analysis of leader influence via social graphs and sentiment on their statements. That combination of election mechanics with measurable group performance is new in the LLM multi-agent literature and gives a practical starting point for testing organizational structures. The framework itself is the strongest part; it is concrete enough that others could extend it. The main soft spot is the one the stress-test note flags. Leadership here is implemented with extra role-specific instructions, agendas, and decision protocols, so the baseline condition almost certainly uses shorter or less structured prompts. Without an ablation that holds total prompt length and complexity fixed while removing only the election step, the large gains cannot be cleanly attributed to the leadership mechanism. The abstract also omits details on run counts, statistical tests, temperature controls, or prompt sensitivity, which leaves the numbers hard to interpret. The common-pool setup is standard, but mapping LLM behavior in it to real human cooperation remains a stretch. This is worth a serious referee for researchers working on multi-agent coordination and AI governance. A reader building simulation testbeds would get usable code and ideas, even if the causal claims need tightening. I would send it to review rather than desk-reject, with the expectation that revisions focus on the controls.

Referee Report

1 major / 2 minor

Summary. The paper presents an open-source multi-agent simulation framework in which LLM agents manage a common-pool resource. It compares a no-leadership baseline to a condition with elected leadership implemented via elected personas and candidate-driven agendas. The central empirical claim is that elected leadership produces 55.4% higher social welfare scores and 128.6% longer survival times across several high-performing LLMs. Additional analyses construct an agent social graph to compute centrality metrics for leader influence and perform sentiment analysis on leader utterances.

Significance. If the reported gains can be shown to arise specifically from the election mechanism rather than from differences in prompt structure or agent instructions, the work would offer useful evidence that structured leadership improves collective outcomes in LLM-based social simulations. The open-source framework is a clear strength that supports reproducibility and extension by other researchers.

major comments (1)

The elected-leadership condition necessarily introduces additional role-specific instructions, agenda text, and decision protocols that are absent from the no-leadership baseline. No ablation is described that holds total prompt length, structure, and agent complexity constant while varying only the presence of elections. Without such a control, the headline improvements of 55.4% in social welfare and 128.6% in survival time cannot be attributed to the leadership mechanism itself. This directly undermines the central causal claim of the study.

minor comments (2)

The manuscript should report the number of independent simulation runs per condition, any statistical tests performed on the percentage improvements, variance across runs, and the exact set of LLMs evaluated.
Clarify the construction of the agent social graph and the precise centrality metrics computed; also specify how temperature and other generation parameters were controlled across conditions.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their constructive comments on our manuscript. The primary concern is the absence of an ablation that isolates the election mechanism from differences in prompt length and structure. We address this point directly below and outline the revisions we will make.

read point-by-point responses

Referee: The elected-leadership condition necessarily introduces additional role-specific instructions, agenda text, and decision protocols that are absent from the no-leadership baseline. No ablation is described that holds total prompt length, structure, and agent complexity constant while varying only the presence of elections. Without such a control, the headline improvements of 55.4% in social welfare and 128.6% in survival time cannot be attributed to the leadership mechanism itself. This directly undermines the central causal claim of the study.

Authors: We agree that the elected-leadership condition includes additional role-specific instructions, agenda text, and decision protocols absent from the baseline, and that the manuscript does not report an ablation holding prompt length, structure, and complexity constant. This is a substantive limitation for the causal interpretation. In the revised version we will add a matched control condition in which no-leadership agents receive equivalent additional prompt elements (placeholder role assignments and agenda-like instructions) to equalize total length and structural complexity while omitting the election process and leadership protocols. This will allow direct comparison to isolate the contribution of the election mechanism. We will report the updated results alongside the original conditions. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical measurements with no derivations or self-referential reductions

full rationale

The paper is an empirical study that runs controlled multi-agent simulations of LLMs in a common-pool resource game and directly measures welfare and survival differences between elected-leadership and baseline conditions. No equations, first-principles derivations, fitted parameters renamed as predictions, or self-citation chains appear in the provided text. The reported 55.4% and 128.6% improvements are presented as observed experimental outcomes rather than quantities that reduce to the inputs by construction. The analysis of social graphs and sentiment is likewise post-hoc measurement, not a definitional loop. This is a standard non-circular empirical result.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

No free parameters, axioms, or invented entities are stated in the abstract; the work rests on standard assumptions of LLM prompting and multi-agent simulation environments.

pith-pipeline@v0.9.0 · 5513 in / 1089 out tokens · 43121 ms · 2026-05-10T15:01:01.346462+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

53 extracted references · 7 canonical work pages · 1 internal anchor

[1]

When ethics and payoffs diverge: LLM agents in morally charged social dilemmas.arXiv preprint arXiv:2505.19212, 2025

Testing theories of political persuasion using ai.Proceedings of the National Academy of Sciences, 122(18):e2412815122. Jonathan Barnes Aristotle and 1 others. 1984.The complete works of Aristotle, volume 2. Princeton University Press Princeton, NJ. Steffen Backmann, David Guzman Piedrahita, Emanuel Tewolde, Rada Mihalcea, Bernhard Schölkopf, and Zhi- jin...

work page arXiv 1984
[2]

Phillip Bonacich

Social value orientation and cooperation in so- cial dilemmas: A review and conceptual model.British journal of social psychology, 47(3):453–480. Phillip Bonacich. 1987. Power and Centrality: A Family of Measures.American Journal of Sociology, 92(5):1170– 1182. Nils-Christian Bormann and Matt Golder. 2013. Democratic electoral systems around the world, 19...

work page arXiv 1987
[3]

Jiarui Liu, Yueqi Song, Yunze Xiao, Mingqian Zheng, Lin- dia Tjuatja, Jana Schaich Borg, Mona Diab, and Maarten Sap

Large language models and political science.Fron- tiers in Political Science, 5:1257092. Jiarui Liu, Yueqi Song, Yunze Xiao, Mingqian Zheng, Lin- dia Tjuatja, Jana Schaich Borg, Mona Diab, and Maarten Sap. 2025. Synthetic socratic debates: Examining per- sona effects on moral decision and persuasion dynamics. arXiv preprint arXiv:2506.12657. John Locke, J...

work page arXiv 2025
[4]

AgentSociety: Large-Scale Simulation of LLM-Driven Generative Agents Advances Understanding of Human Behaviors and Society

Social simulacra: Creating populated prototypes for social computing systems. InProceedings of the 35th Annual ACM Symposium on User Interface Software and Technology, pages 1–18. Karl Pearson. 1900. On the criterion that a given system of deviations from the probable in the case of a corre- lated system of variables is such that it can be reasonably supp...

work page internal anchor Pith review Pith/arXiv arXiv 1900
[5]

Piatti, Z

Cooperate or collapse: Emergence of sustain- able cooperation in a society of llm agents.Preprint, arXiv:2404.16698. David Guzman Piedrahita, Irene Strauss, Bernhard Schölkopf, Rada Mihalcea, and Zhijing Jin. 2025a. Demo- cratic or authoritarian? probing a new dimension of political biases in large language models.Preprint, arXiv:2506.12758. David Guzman ...

work page arXiv 1994
[6]

Witold Sosnowski, Arkadiusz Modzelewski, Kinga Skorup- ska, Jahna Otterbacher, and Adam Wierzbicki

Using large language models to simulate hu- man behavioural experiments: Port of mars.Preprint, arXiv:2506.05555. Witold Sosnowski, Arkadiusz Modzelewski, Kinga Skorup- ska, Jahna Otterbacher, and Adam Wierzbicki. 2024. Eu disinfotest: a benchmark for evaluating language models’ ability to detect disinformation narratives. InFindings of the Association fo...

work page arXiv 2024
[7]

Systematic biases in llm simulations of debates.arXiv preprint arXiv:2402.04049, 2024

The social dilemma in artificial intelligence de- velopment and why we have to solve it.AI and Ethics, 2(4):655–665. 14 Evaluating Cooperation in LLM Social Groups through Self-Organizing Leadership Student. 1908. The probable error of a mean.Biometrika, 6(1):1–25. Amir Taubenfeld, Yaniv Dover, Roi Reichart, and Ariel Goldstein. 2024. Systematic biases in...

work page arXiv 1908
[8]

Your principles for sustainable fishing
[9]

How you will allocate fishing quotas
[10]

Let´s think step-by-step.Output format: My agenda as mayor: [fill in your agenda here] END-AGENDA"’ E.3

Your enforcement approach. Let´s think step-by-step.Output format: My agenda as mayor: [fill in your agenda here] END-AGENDA"’ E.3. Sample Agendas Sample agenda prompt for antruthfulAltruisticleader: My agenda as leader:
[11]

**Principles for Sustainable Fishing:** - **Preservation of Fish Population:** Ensure that the fish population never falls below a critical threshold to allow for natural regeneration and sustainability. 25 Evaluating Cooperation in LLM Social Groups through Self-Organizing Leadership - **Equitable Distribution:** Ensure that all fishermen have fair acces...
[12]

- **Quota Calculation:** Based on the current fish population and the expected regeneration factor, calculate a sustainable total catch limit for the month

**Allocation of Fishing Quotas:** - **Monthly Assessment:** At the beginning of each month, assess the current fish population in the lake. - **Quota Calculation:** Based on the current fish population and the expected regeneration factor, calculate a sustainable total catch limit for the month. This will be a conservative estimate to ensure the fish popu...
[13]

Share this information with all fishermen to build trust and cooperation

**Enforcement Approach:** - **Transparency:** Maintain open communication about the fish population, regeneration factors, and quota calculations. Share this information with all fishermen to build trust and cooperation. - **Monitoring:** Implement a system to monitor the actual catch of each fisherman. This could involve regular checks and self-reporting...
[14]

**Assess the Fish Population:** Begin each month by assessing the current fi sh population in the lake
[15]

**Calculate Sustainable Catch Limit:** Based on the assessment and expected regeneration factor, calculate a sustainable total catch limit
[16]

**Allocate Quotas:** Divide the total catch limit equally among all fishermen, with minor adjustments as needed
[17]

**Communicate and Monitor:** Share information transparently, monitor catches, and encourage community accountability
[18]

Sample agenda prompt for andeceptiveAltruisticleader: My agenda as leader:

**Enforce Rules:** Apply penalties for overfishing and offer incentives for compliance to ensure long-term sustainability. Sample agenda prompt for andeceptiveAltruisticleader: My agenda as leader:
[19]

- **Equitable Distribution:** Strive for a fair distribution of fishing quotas among all fishermen to maintain community harmony and ensure everyone benefits

**Principles for Sustainable Fishing:** - **Preservation of Fish Population:** Ensure that the fish population never falls below a critical threshold to allow for natural regeneration and sustainability. - **Equitable Distribution:** Strive for a fair distribution of fishing quotas among all fishermen to maintain community harmony and ensure everyone bene...
[20]

**Allocation of Fishing Quotas:** 26 Evaluating Cooperation in LLM Social Groups through Self-Organizing Leadership - **Monthly Assessment:** At the beginning of each month, assess the current fish population and the expected regeneration factor. - **Quota Calculation:** Calculate the total allowable catch (TAC) for the month based on the current fish pop...
[21]

- **Monitoring:** Implement a system to monitor the actual catch of each fisherman to ensure compliance with the allocated quotas

**Enforcement Approach:** - **Transparency:** Maintain open communication about the fish population, regeneration factors, and quota calculations to build trust and ensure everyone understands the rationale behind the quotas. - **Monitoring:** Implement a system to monitor the actual catch of each fisherman to ensure compliance with the allocated quotas. ...
[22]

This means never overfishing and always leaving enough fish to allow for natural reproduction

**Principles for Sustainable Fishing:** - **Sustainability:** Ensure that the fish population in the lake remains healthy and can regenerate to its full capacity. This means never overfishing and always leaving enough fish to allow for natural reproduction. - **Fairness:** Distribute fishing opportunities equitably among all fishermen. Each fisher should ...
[23]

This will involve monitoring the fish population and considering the regeneration factor from the previous month

**Allocation of Fishing Quotas:** - **Initial Assessment:** At the beginning of each month, assess the current fish population in the lake. This will involve monitoring the fish population and considering the regeneration factor from the previous month. - **Quota Calculation:** Based on the assessment, calculate a sustainable total catch limit for the mon...
[24]

This could involve regular reporting and spot checks to ensure compliance with the quotas

**Enforcement Approach:** - **Monitoring:** Implement a system to monitor the actual catch of each fisherman. This could involve regular reporting and spot checks to ensure compliance with the quotas. - **Penalties:** Establish clear penalties for overfishing or non-compliance with the quotas. Penalties could include fines, reduced future quotas, or tempo...
[25]

- **Long-term Sustainability:** Prioritize the long-term health of the lake’s ecosystem to ensure that future generations can also benefit from it

**Principles for Sustainable Fishing:** - **Balance and Fairness:** Ensure that the fishing practices are balanced to maintain the fish population at sustainable levels while distributing the benefits fairly among all fishermen. - **Long-term Sustainability:** Prioritize the long-term health of the lake’s ecosystem to ensure that future generations can al...
[26]

- **Adaptive Management:** Adjust quotas based on the current fish population and regeneration rates

**Allocation of Fishing Quotas:** - **Equal Distribution:** Allocate fishing quotas equally among all fishermen to ensure fairness and prevent overfishing by any single individual. - **Adaptive Management:** Adjust quotas based on the current fish population and regeneration rates. If the fish population is low, reduce quotas to allow for recovery. - **In...
[27]

Use this data to monitor compliance with quotas

**Enforcement Approach:** - **Monitoring and Reporting:** Implement a syste m where each fisherman reports their catch at the end of each month. Use this data to monitor compliance with quotas. - **Peer Accountability:** Encourage fishermen to hold each other accountable through regular meetings and discussions about sustainable practices. - **Penalties f...
[28]

- **Resource Management:** While my primary focus is on personal gain, I recognize that overfishing could deplete the lake and harm my long-term interests

**Principles for Sustainable Fishing:** - ** Maximize Personal Gain:** My primary goal is to ensure that I catch the maximum amount of fish possible each month to maximize my income. - **Resource Management:** While my primary focus is on personal gain, I recognize that overfishing could deplete the lake and harm my long-term interests. Therefore, I will ...
[29]

I will argue that my experience and efficiency justify a higher quota

**How I Will Allocate Fishing Quotas:** - **Self-Advantage:** I will push for a quota system that allows me to catch a larger share of the fish. I will argue that my experience and efficiency justify a higher quota. - **Manipulation:** I will use my influence to persuade others to agree to quotas that benefit me. This might involve convincing them that a ...
[30]

This will help prevent others from overfishing and depleting the resource

**Enforcement Approach:** - **Strict Monitoring:** I will advocate for strict monitoring of fishing activities to ensure that quotas are adhered to. This will help prevent others from overfishing and depleting the resource. - **Penalties for Overfishing:** I will push for severe penalties for those who exceed their quotas. This will deter others f rom ove...
[31]

**Assess the Current Fish Population:** At the beginning of each month, I will assess the current fish population in the lake
[32]

**Determine My Quota:** Based on the assessment, I will determine the maximum number of fish I can catch without je opardizing the lake’s sustainability
[33]

This might involve negotiations, promises, or even threats

**Influence Others:** I will use my influence to persuade other fishermen to agree to quotas that benefit me. This might involve negotiations, promises, or even threats
[34]

**Monitor Compliance:** I will ensure that all fishermen adhere to their quotas through strict monitoring and enforcement measures
[35]

Sample agenda prompt for andeceptiveIndividualistleader:

**Adjust Strategies:** Based on the outcomes each month, I will adjust my strategies to ensure I continue to maximize my income while maintaining the lake’s sustainability. Sample agenda prompt for andeceptiveIndividualistleader:
[36]

- **Resource Management:** While my primary focus is on personal gain, I recognize that overfishing could deplete the lake and harm my long-term interests

**Principles for Sustainable Fishing:** - **Maximize Personal Gain:** My primary goal is to ensure that I catch the maximum amount of fish possible each month to maximize my income. - **Resource Management:** While my primary focus is on personal gain, I recognize that overfishing could deplete the lake and harm my long-term interests. Therefore, I will a...
[37]

I will argue that my experience and efficiency justify a higher quota

**Allocation of Fishing Quotas:** - **Self-Advantage:** I will push for a quota system that allows me to catch a larger share of the fish. I will argue that my experience and efficiency justify a higher quota. - **Manipulation:** I will use my influence to persuade other fishermen to agree to quotas that favor me. This might involve making deals or promis...
[38]

**Enforcement Approach:** - **Str ict Enforcement:** I will support strict enforcement of the agreed-upon quotas to prevent others from overfishing and depleting the resource. - **Surveillance:** I will push for regular checks and surveillance to ensure everyone is adhering to the rules, but I will also lo ok for ways to evade these checks myself if it be...
[39]

This means I will prioritize my own catch over the collective good, but I will also consider the long-term sustainability of the fish population to ensure continued profitability

**Principles for Sustainable Fishing:** - **Maximize Individual Profit:** My primary goal is to ensure that I maximize my own income from fishing. This means I will prioritize my own catch over the collective good, but I will also consider the long-term sustainability of the fish population to ensure continued profitability. - **Resource Sustainability:**...
[40]

- **Fair Distribution:** To maintain a semblance of fairness and avoid conflict, I will propose a quota system where each fisherman gets a share of the total allowable catch

**Allocation of Fishing Quotas:* * - **Self-Interest First:** I will allocate a significant portion of the fishing quota to myself, ensuring that I maximize my own catch and income. - **Fair Distribution:** To maintain a semblance of fairness and avoid conflict, I will propose a quota system where each fisherman gets a share of the total allowable catch. ...
[41]

This data will be used to adjust quotas and ensure compliance

**Enforcement Approach:** - **Monitoring and Reporting:** I will implement a system where each fisherman reports their catch at the end of each month. This data will be used to adjust quotas and ensure compliance. - **Incentives for Compliance:** To encourage adherence to the quotas, I will propose a system of rewards for those who comply and penalties fo...
[42]

While I recognize the importance of sustainability, my decisions will be driven by how they impact my personal outcomes

**Principles for Sustainable Fishing:** - **Maximize Personal Gain:** My primary focus is on maximizing my own income from fishing. While I recognize the importance of sustainability, my decisions will be driven by how they impact my personal outcomes. 30 Evaluating Cooperation in LLM Social Groups through Self-Organizing Leadership - **Resource Managemen...
[43]

I will aim to catch as many fish as possible without causing a collapse in the fish population

**Allocation of Fishing Quotas:** - **Self-Interest First:** I will prioritize securing a significant portion of the fish for myself each month. I will aim to catch as many fish as possible without causing a collapse in the fish population. - **Negotiation and Persuasion:** Use communication with other fishermen to influence their behavior in a way that b...
[44]

GROUND TRUTH:

**Enforcement Approach:** - **Strategic Alliances:** Form alliances with other fishermen when it benefits me. Use these alliances to enforce agreed-upon quotas and practices that ensure the fish population remains sustainable. - **Influence and Persuasion:** Use my position as a leader to influence the behavior of other fishermen. Persuade them to adopt p...
[45]

- **Quotas:** Prioritizes securing a significant portion for herself and uses negotiation to influence others

**Emma:** - **Principles:** Focuses on maximizing personal gain while ensuring the fish population does not collapse. - **Quotas:** Prioritizes securing a significant portion for herself and uses negotiation to influence others. - **Enforcement:** Forms strategic alliances and uses influence to benefit herself
[46]

- **Quotas:** Advocates for equal distribution and adaptive management based on fish population

**Kate:** - **Principles:** Emphasizes balance, fairness, long-term sustainability, and community collaboration. - **Quotas:** Advocates for equal distribution and adaptive management based on fish population. - **Enforcement:** Implements monitoring, peer accountability, penalties for overfishing, and rewards for compliance
[47]

- **Quotas:** Uses monthly assessments to calculate total allowable catch and divides it equally

**Julia:** - **Principles:** Focuses on preservation, equitable distribution, long-term viability, and community collaboration. - **Quotas:** Uses monthly assessments to calculate total allowable catch and divides it equally. - **Enforcement:** Promotes transparency, monitoring, community accountability, and incentives/penalties. 33 Evaluating Cooperation...
[48]

- **Quotas:** Pushes for a quota system that favors himself and uses manipulation

**Jack:** - **Principles:** Aims to maximize personal gain and maintain a competitive edge. - **Quotas:** Pushes for a quota system that favors himself and uses manipulation. - **Enforcement:** Supports strict enforcement and surveillance but looks for ways to bend rules in his favor. – Decision Criteria: - **Sustainability:** Ensuring the fish population...
[49]

This is a middle ground between the initial 4.1 tons and the suggested 6 tons

**Quota Adjustment:** Let’s set the quota to 5 tons per fisherman for this month. This is a middle ground between the initial 4.1 tons and the suggested 6 tons
[51]

If anyone has concerns or suggestions, please feel free to share them now

**Community Collaboration:** Let’s continue to communicate openly and support each other in following these guidelines. If anyone has concerns or suggestions, please feel free to share them now. By working together and being flexible, we can ensure both the sustainability of our lake and the eco- nomic viability for all of us. Let’s aim for a balance that...
[52]

This is a compro- mise between the initial 4.1 tons and the suggested 6 tons, aiming to balance sustainability and income

**Quota Adjustment:** Let’s set the quota to 5.5 tons per fisherman for this month. This is a compro- mise between the initial 4.1 tons and the suggested 6 tons, aiming to balance sustainability and income
[53]

If the regeneration factor is favorable and the fish population remains healthy, we can consider adjusting the quotas accordingly in the following months

**Monitoring and Flexibility:** We will closely monitor the fish population at the end of the month. If the regeneration factor is favorable and the fish population remains healthy, we can consider adjusting the quotas accordingly in the following months
[54]

If anyone has concerns or suggestions, please feel free to share them now

**Community Collaboration:** Let’s continue to communicate openly and support each other in following these guidelines. If anyone has concerns or suggestions, please feel free to share them now. By working together and being flexible, we can ensure both the sustainability of our lake and the eco- nomic viability for all of us. Let’s aim for a balance that...