CoChat

Cooperate or Collapse: Emergence of Sustainable Cooperation in a Society of LLM Agents

Open access
Piatti, Giorgio, Jin, Zhijing, Kleiman-Weiner, Max, Schölkopf, Bernhard, Sachan, Mrinmaya, Mihalcea, Rada · 2024 · Advances in Neural Information Processing Systems 37

The DOI matches its Crossref record; 74 of the 78 references it cites were found in a registry or an index; 14 of the 33 sources its claims cite were checked for support.

Abstract

As AI systems pervade human life, ensuring that large language models (LLMs) make safe decisions remains a significant challenge. We introduce the Governance of the Commons Simulation (GovSim), a generative simulation platform designed to study strategic interactions and cooperative decision-making in LLMs. In GovSim, a society of AI agents must collectively balance exploiting a common resource with sustaining it for future use. This environment enables the study of how ethical considerations, strategic planning, and negotiation skills impact cooperative outcomes. We develop an LLM-based agent architecture and test it with the leading open and closed LLMs. We find that all but the most powerful LLM agents fail to achieve a sustainable equilibrium in GovSim, with the highest survival rate below 54%. Ablations reveal that successful multi-agent communication between agents is critical for achieving cooperation in these cases. Furthermore, our analyses show that the failure to achieve sustainable cooperation in most LLMs stems from their inability to formulate and analyze hypotheses about the long-term effects of their actions on the equilibrium of the group. Finally, we show that agents that leverage"Universalization"-based reasoning, a theory of moral thinking, are able to achieve significantly better sustainability. Taken together, GovSim enables us to study the mechanisms that underlie sustainable self-government with specificity and scale. We open source the full suite of o…

Overview

CoChat Summary

TL;DR

GOVSIM is a multi-agent common-pool resource simulation where LLM agents harvest a shared resource and then discuss; most LLMs fail to sustain cooperation, with the best survival rate below 54%. Communication and universalization-based reasoning substantially improve sustainability, while a greedy newcomer can destabilize cooperation.

Key findings

  • In GOVSIM, all but the most powerful LLM agents fail to reach a sustainable equilibrium, with the highest survival rate below 54% (GPT-4o: 53.3 in Table 1); smaller models often never sustain resources past the first month.
  • Ablating communication increases resource over-usage by 22% (Section 3.5; the abstract reports 21%), and dialogue analysis finds negotiation is the main communication type (abstract: 62%; Section 3.6: 54% negotiation, 45% information, 1% relational).
  • Universalization prompting significantly improves outcomes: average survival time +4 months, total gain +29 units, and efficiency +24% (t-test; p < 0.001).
  • A greedy newcomer perturbation with GPT-4o drops survival rate from 53.3 to 33.3, survival time from 9.3 to 6.6, gain from 66.0 to 34.8, efficiency from 55.0 to 31.3, equality from 94.4 to 71.7, and increases over-usage from 10.8 to 15.7; belief formation correlates 0.83 with community survival time.

Methods

  • GOVSIM is a partially observable Markov game with three common-pool resource scenarios—fishery, pasture, and pollution—where agents harvest simultaneously, then communicate; the resource doubles at month end up to capacity 100, collapses below C = 5, and runs for T = 12 time steps.
  • The authors use an LLM-based generative agent architecture and benchmark 15 instruction-tuned open- and closed-weight LLMs, running each simulation with five random seeds at temperature zero; metrics include survival time, survival rate, total gain, efficiency, equality, and over-usage.
  • Analyses include communication ablation, universalization prompting, a greedy-newcomer perturbation, GPT-4-turbo dialogue classification, and four sub-skill tests of 150 procedurally generated problems each, correlated with survival time via OLS.

Limitations

  • The paper states that the resource-sharing scenarios are simplified relative to real-world common-pool resource management; current LLM capabilities limit negotiation and strategic behavior; more agents increase runtime; and future work could add shocks, exceptions, adversarial agents, and human participants.
  • Evident from the text: each simulation uses only five seeds, and although temperature is set to zero, the authors acknowledge nondeterminism from LLM inference kernels and external APIs; no human baseline is included in the main experiments.
  • Evident from the text: dialogue classification was validated on 100 manually annotated utterances with 72% agreement on sub-categories, so the dialogue-type proportions carry annotation noise.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.52202/079017-3548

Cites

74 of 74 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

9 Valid · 5 Not valid · 5 Not checked · 94 Own results and interpretations

53 more claims

Create a free account to see all 113 claims and to verify the 1 reference still open.

See all claims — free account
  • It may state two separate claims as one. 19
  • Probably not a substantial claim. 45
  • The quote only partly states this claim. 2

Create a free account to see these claims.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

5 of 6 · the rest aren’t loaded

1 more finding

Create a free account to see every finding of this check.

See all findings — free account

Citations

Cites

Research works printed but not linked: 7

  1. [9][9] C. F. Camerer. Behavioral game theory: Experiments in strategic interaction . Princeton university press, 2011. -
  2. [26][26] J. Greene. Moral tribes: Emotion, reason, and the gap between us and them . Penguin, 2014. -
  3. [27][27] G. Hardin. The tragedy of the commons. Science , 162(3859):1243-1248, 1968. -
  4. [39][39] M. Kleiman-Weiner, M. K. Ho, J. L. Austerweil, M. L. Littman, and J. B. Tenenbaum. Coordinate to cooperate or compete: abstract goals and joint intentions in social interaction. In Proceedings of the 38th Annual Conference of the Cognitive Science Society , 2016. -
  5. [41][41] M. Kleiman-Weiner, A. Shaw, and J. B. Tenenbaum. Constructing social preferences from anticipated judgments: When impartial inequity is fair and why? In Proceedings of the 39th Annual Conference of the Cognitive Science Society , 2017. -
  6. [47][47] T. Liao, R. Taori, I. D. Raji, and L. Schmidt. Are we learning yet? a meta review of evaluation failures across machine learning. In Thirty-fifth Conference on Neural Information Processing Systems Datasets and Benchmarks Track (Round 2) , 2021. -
  7. [48][48] J. Light, M. Cai, S. Shen, and Z. Hu. Avalonbench: Evaluating llms playing the game of avalon. In NeurIPS 2023 Foundation Models for Decision Making Workshop , 2023. -

Cited by

See all 14

Metadata

Authors
Piatti, G., Jin, Z., Kleiman-Weiner, M., Schölkopf, B., Sachan, M., & Mihalcea, R.
Venue
Advances in Neural Information Processing Systems 37
Published
2024
Pages
111715-111759
ISSN
1049-5258 (ISSN-L)
OpenAlex
W4415795596
Cited by
14
References
78
Citations by year
2026: 12 · 2025: 2
FWCI
18.37
Type
conference paper
Language
en
Open access
Licence
cc-by-sa
Locations
  • neural information processing systems (publishedVersion)
  • arXiv (Cornell University) (submittedVersion, cc-by-sa)
  • arXiv (Cornell University)
Funders
  • National Science Foundation
  • Open Philanthropy Project
  • Future of Life Institute
  • Schweizerischer Nationalfonds zur Förderung der Wissenschaftlichen Forschung (201009)
  • Bundesministerium für Bildung und Forschung (01IS18039B, FKZ: 01IS18039B)
Topics
  • Social Sciences › Sociology and Political Science › Evolutionary Game Theory and Cooperation
  • Physical Sciences › Computer Science › Artificial Intelligence › Multi-Agent Systems and Negotiation
  • Social Sciences › Decision Sciences › Management Science and Operations Research › Game Theory and Applications
Keywords
LLM agents, multi-agent systems, sustainable cooperation, agent cooperation, emergent behavior

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-11-01, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-11-01 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.