Pith. sign in

REVIEW 3 major objections 5 minor 111 references

Modeling human reputation-seeking behavior in a spatio-temporally complex public good provision game

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Human groups cooperate in the Clean Up public goods game only when they can identify one another and track contributions, and a reputation-motivated multi-agent reinforcement learning model reproduces that anonymity effect and the…

desk verdict A transparently derivative re-interpretation of an existing preprint; the human experiment is solid but the model's anonymity effect is likely by construction. read the letter →

arxiv 2506.06032 v1 pith:JOQQ2DY4 submitted 2025-06-06 cs.MA

Edward Hughes , Tina O. Zhu , Martin J. Chadwick , Raphael Koster , Antonio García Castañeda , Charles Beattie , Thore Graepel , Matthew M. Botvinick
show 1 more author
Joel Z. Leibo
This is my paper
classification cs.MA
keywords multi-agentreinforcementlearningpublicgoodsgamereputationanonymityidentifiabilityturn-takingCleanUpenvironmentcompetitivealtruism
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper sets out to establish two stylized facts—empirical regularities that candidate models should reproduce—about human cooperation in Clean Up, a spatially and temporally dynamic public good provision game: groups cooperate when players can identify one another and track each other's contributions over time, and anonymity destroys cooperation. It then claims that a multi-agent reinforcement learning model whose agents carry an intrinsic motivation to maintain a good reputation reproduces both facts, matching not only the overall level of cleaning but also the way groups coordinate, namely by taking turns in the river rather than dividing it into territories. The broader point is that MARL can be grounded in laboratory experiments with real human participants, giving it empirical support as a method for studying social behavior in settings too complex for standard game theory. If the claims hold, the two human regularities become benchmarks that candidate computational models of cooperation should be expected to match.

What carries the argument

The load-bearing object is the Clean Up environment itself, a partially observable Markov game in which five embodied players clean a river that accumulates pollution so that a spatially separate orchard regrows apples, with individual reward for apples pitted against the collective benefit of cleaning. On top of this environment, the model adds an intrinsic reward for competitive altruism: $$r_i = -\$\alpha$\max(\bar{c}-c_{\mathrm{self}},0) - \$\beta$\max(c_{\mathrm{self}}-\bar{c},0),$$ with $\alpha\sim U(2.4,3.0)$ and $\beta\sim U(0.16,0.20)$, where $c_{\mathrm{self}}$ is the focal agent's temporally smoothed cleaning contribution and $\bar{c}$ is the observed group average. The asymmetry $\alpha>\beta$ makes the motivation primarily an aversion to having a lower reputation than one's peers and secondarily an aversion to being taken advantage of by free riders. Contribution levels are updated by exponential smoothing with factor 0.97, and in the anonymous condition the social observation of others' contributions is corrupted by substantial noise. Because the intrinsic reward is small and negative at convergence, the paper reads it as a nudge to the learning dynamics that lets populations discover cleaning and turn-taking, rather than as a direct payoff for altruistic acts.

What would settle it

Train the reputation-motivated agents in the anonymous condition while varying the corruption level on the social observation from zero to large, holding everything else fixed, and measure cleaning and collective return; if cooperation already disappears at small noise, anonymity itself is the cause, whereas if it persists until the noise is overwhelming, the reported human-model match depends on the unreported noise magnitude. A companion check replaces the corrupted social observation with the clean group average while keeping avatars anonymous: agents that still cooperate there would show that loss of identity, not loss of contribution information, is what drives the effect.

Watch

Extended reading notes

Core claim

The central discovery, stated on the paper's own terms, is that identifiability rather than mere publicness of outcomes is what unlocks human cooperation in Clean Up, and that the same behavioral difference can be produced by a deep reinforcement learning agent whose reward includes a reputation term. In the human experiment, 120 participants in groups of five cleaned significantly more and earned significantly higher collective returns when avatars had distinct colors and contribution bars than when all other avatars were identical, and the model's trained agents showed the same pattern. Both humans and agents also coordinated by turn taking in the identifiable condition, with the degree of turn taking mediating the relationship between identifiability and collective return, while territoriality did not explain the gains. The paper explicitly notes that it does not prove reputation is the only or the actual psychological mechanism, only that this is the first MARL model to reproduce both stylized facts.

Load-bearing premise

The load-bearing premise is that the anonymous condition's corrupted social observation is comparable to what human participants experience, but the paper never reports the size of that corruption, so the model's loss of cooperation could be a trivial consequence of the noise rather than of anonymity itself.

Editorial extensions

If this is right

  • Other MARL cooperation mechanisms that depend only on aggregate reward distributions or on imitation of others' behavior should show no anonymity effect, so the identifiable-versus-anonymous gap becomes a discriminating test for candidate models.
  • In this model family, anonymity should lower group contribution and collective return in any public good environment where the intrinsic reward depends on observing a group average; the Clean Up result is one realization of that general prediction.
  • Turn taking, not territoriality, is the coordination mode that identifiability selects in both humans and agents, and the more a group relies on it, the higher its collective return.
  • The two human regularities are proposed as stylized facts: regularities important enough that future models of cooperation in complex environments should be expected to match them.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Because the paper never reports the magnitude of the 'substantial noise' used to corrupt social observations in the anonymous condition, the model's failure to cooperate there is not yet a parameter-free prediction; sweeping that noise level from zero upward would show whether the anonymity gap is a reputation effect or a corruption artifact.
  • The identifiable and anonymous conditions differ in both avatar identity and access to others' contribution levels, so an experiment that holds one fixed while varying the other could separate the contribution of identity tracking from the contribution of seeing aggregate contributions.
  • If the learnability account of turn-taking prominence is right, other MARL social dilemmas whose early random exploration rarely produces near-territorial trajectories should also converge on temporal rotation rather than spatial division, which could be tested without new human data.
  • A practical corollary for online collective-action systems is that making contribution histories attributable may be sufficient to sustain cooperation: the model predicts that anonymous but visible contributions will not maintain public good provision.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper reports a Clean Up experiment with 120 human participants in identifiable versus anonymous conditions, finding that identifiability increases public-good contributions, collective return, and turn-taking coordination. It then presents a multi-agent reinforcement learning (MARL) model in which agents receive an intrinsic reward for maintaining a relatively high contribution level compared to the group average, and reports that this model reproduces the same qualitative differences between conditions, including the use of turn-taking rather than territoriality. The paper frames its main contribution as establishing two stylized facts about human cooperation in a spatio-temporally complex public good setting and as the first MARL model to reproduce both stylized facts.

Significance. If the modeling claim holds, this is a valuable step toward grounding MARL models in human behavioral data: the human experiment is carefully designed and reported, with counterbalancing, repeated-measures analyses, mediation analyses, comprehension checks, learning-effect checks, and an explicit verification that both the human and agent versions of Clean Up instantiate a social dilemma. The turn-taking versus territoriality result is interesting and is supported by purpose-built metrics. The paper also makes a useful methodological contribution by proposing that MARL models be judged against stylized facts from laboratory experiments. However, the significance of the computational result is currently conditional: the model's anonymity manipulation is implemented by corrupting the exact signal on which the intrinsic reward is based, and the corruption is never quantified, so the claimed reproduction of the human anonymity effect is not yet a verifiable prediction.

major comments (3)
  1. [Supplementary Information, Design of Computational Model (Eqs. 3-4)] The anonymous condition is implemented by corrupting the contribution information with 'substantial noise', but the noise distribution, its magnitude, and whether it corrupts c_self as well as the other agents' contributions are never reported. This is load-bearing because the intrinsic reward r_i in Eq. (4) is a function of the group average \bar c: if the noise is large, the reward signal carries no information about one's standing relative to the group, making the collapse of cooperation trivial; if the noise is small, cooperation should persist. The reported anonymity effect in the model is therefore controlled by an unreported knob rather than being a quantitative prediction. Please specify the exact corruption process, report the resulting signal-to-noise ratio, and provide robustness sweeps over the noise level.
  2. [Supplementary Information, Design of Computational Model; Methods, Human Behavioral Experiment] The human anonymous condition removes the contribution information for other players ('this information is absent'), whereas the computational model corrupts the contribution information with noise. These are not the same manipulation. Because the model's mechanism operates directly on the corrupted group-average signal, the model's failure under anonymity may reflect the injected noise rather than the absence of identity-relevant information. The authors need to justify the noise process as a model of human visual attention and demonstrate that the qualitative result is robust to the noise level, or implement the experimental manipulation more faithfully.
  3. [Introduction, stylized-facts paragraph; Discussion, Implications of the anonymity effect] The claim that the proposed model is 'the first to reproduce both stylized facts' is not tested. The paper argues on a priori grounds that other MARL cooperation mechanisms (e.g., inequity aversion, social value orientation, reciprocity, social influence) would be unaffected by anonymity, but no alternative models are trained under the same identifiable-versus-anonymous manipulation. A minimal baseline experiment, such as inequity-averse agents in the same two conditions, is needed to establish that the reputation mechanism, rather than the corruption of the reward input, is responsible for the effect. Without such a comparison, the 'first' claim is unsupported.
minor comments (5)
  1. [Front matter note] The note stating that this work was previously published on arXiv as [64] and that it differs from the interpretation in [64] should be integrated into an acknowledgments or provenance statement rather than appearing before the abstract; the relationship to [64] should be clearly explained.
  2. [Figure 4 caption] The caption says 'The intrinsic motivation for reputation substantially alters the behavior of human groups', but intrinsic motivation was not manipulated in the human experiment; the caption should refer to the identifiability condition instead.
  3. [Supplementary Information, Main Group Effects, Computational Model] The sentence 'In the model, groups cleaned significantly more in the identifiable condition (for 422 steps on average) in the identifiable condition than in the anonymous condition' contains a duplicated phrase and should be corrected.
  4. [Eq. (2) and Eq. (4)] The symbol \bar c is described as an 'estimated or observed average' but the exact way it is computed from the per-agent smoothed contributions c_j in Eq. (6) is not defined explicitly; please state the definition in one place.
  5. [Results, Effects of identifiability versus anonymity] The effect sizes differ substantially between the human experiment (396 vs. 337 cleaning steps) and the model (422 vs. 299 cleaning steps); a brief comment on whether the authors intend a qualitative or quantitative match would help readers interpret the 'recapitulates' claim.

Circularity Check

2 steps flagged · score 6.0 of 10

The model's anonymity effect is built into the anonymous-condition input corruption, and the paper's 'first to reproduce both stylized facts' claim re-labels the authors' identical prior work.

  1. self definitional [Supplementary Information, Design of Computational Model, after Eq. (4)]
    "The environment provides contribution information for each group member currently in view as an input to the agent. In the anonymous condition the contribution information is corrupted by substantial noise."

    Eq. (4) defines r_i = -alpha*max(cbar - c_self,0) - beta*max(c_self - cbar,0), so the intrinsic reward is driven entirely by the group average cbar. The anonymous condition is implemented by corrupting the contribution information that supplies cbar with 'substantial noise', whose distribution and magnitude are never reported. Therefore the model's anonymous agents receive a degraded reward signal by construction: sufficiently large noise destroys the r_i signal and trivially suppresses cleaning, while zero noise would make the anonymous condition identical to identifiable. The claimed reproduction of the human anonymity effect is thus an artifact of an unquantified input manipulation rather than an emergent prediction from reputation learning.

  2. renaming known result [Title-page note; see also 'A stylized fact...' paragraph in the Introduction]
    "Note: Some of this work was previously published on arxiv as[64]. ... It should be noted that all the experimental methods, statistical procedures, and quantitative results are identical to those in[64]"

    The paper's central contribution is the identification of two human stylized facts and the claim that 'The MARL model we describe is the first to reproduce both stylized facts.' But the note states that all experimental methods, statistical procedures, and quantitative results are identical to the authors' own prior arXiv paper [64]. The 'first to reproduce' claim therefore restates a result already reported in [64] under a new interpretation; the human data, the model, and the statistics are not new evidence. This is a re-labeling of a known result as 'stylized facts' rather than an independent derivation or a new empirical demonstration.

full rationale

The human behavioral experiment itself is an independent and valuable empirical result, and the turn-taking coordination observed in both humans and agents is an emergent pattern that is not directly hard-coded. However, the model's reproduction of the central anonymity stylized fact reduces partly to construction: the intrinsic reward in Eq. (4) depends on the group-average contribution cbar, and the anonymous condition is defined by corrupting the input carrying that quantity with 'substantial noise' of unspecified size. The direction and magnitude of the model's anonymity effect are therefore controlled by an unreported input dial, making the claimed quantitative match unfalsifiable. In addition, the paper discloses that all methods and quantitative results are identical to the authors' prior arXiv paper [64], so presenting the model as 'the first to reproduce both stylized facts' is a re-framing of known results. The emergent turn-taking result and the independent human experiment prevent this from being fully circular, but the central claim is partially forced by definition and by self-citation of identical prior work.

Assumptions & free parameters 4 free parameters · 4 assumptions · 1 invented entities

The model's core behavior depends on the hand-chosen coefficients alpha and beta, the smoothing factor lambda, and the unreported anonymous-condition noise level. The paper also assumes a common-knowledge norm and a competitive-altruism mechanism that the experiment does not directly test. The human behavioral data are real, but the model's explanatory work is driven by these free choices.

free parameters (4)
  • alpha (intrinsic reward coefficient) = sampled from U(2.4, 3.0)
    Controls the penalty for contributing below the group average in Eq (4); chosen by hand to instantiate competitive altruism, not fitted to the human data.
  • beta (intrinsic reward coefficient) = sampled from U(0.16, 0.20)
    Controls the penalty for contributing above the group average in Eq (4); hand-chosen to make the below-average aversion dominate.
  • lambda (contribution smoothing factor) = 0.97
    Set in Eq (6) to define cself and cbar from instantaneous cleaning events; chosen without reported sensitivity analysis.
  • anonymous-condition noise level = not reported ('substantial noise')
    The corruption applied to contribution information in the anonymous condition is not quantified; the model's drop in cooperation depends on this value.
assumptions (4)
  • ad hoc to paper The intrinsic reward of Eq (4), with alpha > beta, is an appropriate model of competitive altruism and human reputation seeking.
    Posited in 'A MARL model of reputation-based cooperation'; the human data do not identify the functional form of the motivation.
  • domain assumption It is common knowledge which actions (cleaning the river) affect reputation and in which direction.
    Listed as a limitation in the Discussion; the model requires this to make cleaning the relevant contribution behavior.
  • domain assumption Clean Up with the chosen parameters instantiates a social dilemma for both human and agent participants.
    Supported by the empirical Schelling diagram and regressions in the Supplementary Information, but this is a verification exercise on the same data, not an external proof.
  • domain assumption The increase in human cooperation under identifiability is caused by reputation concerns rather than by visual salience or coordination cues alone.
    Supported by post-task questionnaire responses, but the experimental manipulation confounds identity tracking with the display of contribution bars, so the causal role of reputation is not isolated.
invented entities (1)
  • Reputation-seeking intrinsic reward r_i
    purpose: To incentivize agents to avoid contributing less than the group average, implementing competitive altruism.
    No falsifiable handle outside the model; the post-task questionnaire is a weak self-report correlate, not an independent test of the reward function.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Modeling human reputation-seeking behavior in a spatio-temporally complex public good provision game." pith.science (2026). https://pith.science/paper/JOQQ2DY4

@misc{pith2026250606032,
  author       = {Pith},
  title        = {Pith review of: Modeling human reputation-seeking behavior in a spatio-temporally complex public good provision game},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JOQQ2DY4}},
  note         = {Machine review of arXiv:2506.06032}
}
read the original abstract

Multi-agent reinforcement learning algorithms are useful for simulating social behavior in settings that are too complex for other theoretical approaches like game theory. However, they have not yet been empirically supported by laboratory experiments with real human participants. In this work we demonstrate how multi-agent reinforcement learning can model group behavior in a spatially and temporally complex public good provision game called Clean Up. We show that human groups succeed in Clean Up when they can see who is who and track reputations over time but fail under conditions of anonymity. A new multi-agent reinforcement learning model of reputation-based cooperation demonstrates the same difference between identifiable and anonymous conditions. Furthermore, both human groups and artificial agent groups solve the problem via turn-taking despite other options being available. Our results highlight the benefits of using multi-agent reinforcement learning to model human social behavior in complex environments.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

111 extracted references · 73 canonical work pages

  1. [1]

    Doing good or doing well? Image motivation and monetary incentives in behaving prosocially

    Dan Ariely, Anat Bracha, and Stephan Meier. “Doing good or doing well? Image motivation and monetary incentives in behaving prosocially”. In:American Economic Review99.1 (2009), pp. 544–55

    2009

  2. [2]

    Axelrod.The Evolution of Cooperation

    Robert M. Axelrod.The Evolution of Cooperation. Basic Books. Basic Books, 1984.i sbn: 9780465021215

    1984

  3. [3]

    Emergent reciprocity and team formation from randomized uncertain social preferences

    Bowen Baker. “Emergent reciprocity and team formation from randomized uncertain social preferences”. In:Advances in Neural Information Processing Systems33 (2020), pp. 15786– 15799. 39 Modeling human reputation-seeking behavior in a spatio-temporally complex public good provision game

    2020

  4. [4]

    Emergent tool use from multi-agent autocurricula

    Bowen Baker et al. “Emergent tool use from multi-agent autocurricula”. In:arXiv preprint arXiv:1909.07528 (2019)

    arXiv 2019

  5. [5]

    Social Value Orientation and cooperation in social dilemmas: A meta-analysis

    Daniel Balliet, Craig Parks, and Jeff Joireman. “Social Value Orientation and cooperation in social dilemmas: A meta-analysis”. In:Group Processes & Intergroup Relations12.4 (2009), pp. 533–547

    2009

  6. [6]

    Trustworthiness and competitive altruism can also solve the “tragedy of the commons

    Pat Barclay. “Trustworthiness and competitive altruism can also solve the “tragedy of the commons””. In:Evolution and Human Behavior25.4 (2004), pp. 209–220

    2004

  7. [7]

    Strategies for cooperation in biological markets, especially for humans

    Pat Barclay. “Strategies for cooperation in biological markets, especially for humans”. In: Evolution and Human Behavior34.3 (2013), pp. 164–175

    2013

  8. [8]

    Partner choice creates competitive altruism in humans

    Pat Barclay and Robb Willer. “Partner choice creates competitive altruism in humans”. In: Proceedings of the Royal Society B: Biological Sciences274.1610 (2006), pp. 749–753

    2006

Show all 111 references
  1. [9]

    Dissecting components of reward: ‘Liking’, ‘wanting’, and learning

    Kent C. Berridge, Terry E. Robinson, and J. Wayne Aldridge. “Dissecting components of reward: ‘Liking’, ‘wanting’, and learning”. In:Current Opinion in Pharmacology9.1 (2009), pp. 65–73

    2009

  2. [10]

    Agent-based modeling: Methods and techniques for simulating human systems

    Eric Bonabeau. “Agent-based modeling: Methods and techniques for simulating human systems”. In:Proceedings of the national academy of sciences99.suppl_3 (2002), pp. 7280– 7287

    2002

  3. [11]

    Does observability affect prosociality?

    Alex Bradley, Claire Lawrence, and Eamonn Ferguson. “Does observability affect prosociality?” In: Proceedings of the Royal Society B: Biological Sciences285.1875 (2018), p. 20180116

    2018

  4. [12]

    Camerer.Behavioral Game Ttheory: Experiments in Strategic Interaction

    Colin F. Camerer.Behavioral Game Ttheory: Experiments in Strategic Interaction. Princeton University Press, 2003

    2003

  5. [13]

    Get it in writing: Formalcontractsmitigatesocialdilemmasinmulti-agentRL

    Phillip J. K. Christofferson, Andreas A. Haupt, and Dylan Hadfield-Menell. “Get it in writing: Formalcontractsmitigatesocialdilemmasinmulti-agentRL”.In: arXivpreprintarXiv:2208.10469 (2022)

    2022 arXiv

  6. [14]

    What is the bandwidth of perceptual experience?

    Michael A. Cohen, Daniel C. Dennett, and Nancy Kanwisher. “What is the bandwidth of perceptual experience?” In:Trends in Cognitive Sciences20.5 (2016), pp. 324–335

    2016

  7. [15]

    The two disciplines of scientific psychology

    Lee J. Cronbach. “The two disciplines of scientific psychology”. In:American Psychologist 12.11 (1957), p. 671

    1957

  8. [16]

    Open problems in Cooperative AI

    Allan Dafoe et al. “Open problems in Cooperative AI”. In:arXiv preprint arXiv:2012.08630 (2020)

    2020 arXiv

  9. [17]

    Learning reciprocity in complex sequential social dilemmas

    Tom Eccles et al. “Learning reciprocity in complex sequential social dilemmas”. In:arXiv preprint arXiv:1903.08082(2019)

    2019 arXiv

  10. [18]

    Eight grand challenges in socio-environmental systems modeling

    Sondoss Elsawah et al. “Eight grand challenges in socio-environmental systems modeling”. In: Socio-Environmental Systems Modelling2 (2020), pp. 16226–16226

    2020

  11. [19]

    Indirectreciprocityandstrategicreputationbuildingin an experimental helping game

    DirkEngelmannandUrsFischbacher.“Indirectreciprocityandstrategicreputationbuildingin an experimental helping game”. In:Games and Economic Behavior67.2 (2009), pp. 399–407

    2009

  12. [20]

    Impala: Scalable distributed deep-RL with importance weighted actor- learner architectures

    Lasse Espeholt et al. “Impala: Scalable distributed deep-RL with importance weighted actor- learner architectures”. In:International Conference on Machine Learning. 2018, pp. 1407– 1416

    2018

  13. [21]

    Cooperation and punishment in public goods experiments

    Ernst Fehr and Simon Gächter. “Cooperation and punishment in public goods experiments”. In: American Economic Review90.4 (2000), pp. 980–994

    2000

  14. [22]

    Altruistic punishment in humans

    Ernst Fehr and Simon Gächter. “Altruistic punishment in humans”. In:Nature 415.6868 (2002), pp. 137–140. 40 Modeling human reputation-seeking behavior in a spatio-temporally complex public good provision game

    2002

  15. [23]

    A theory of fairness, competition, and cooperation

    Ernst Fehr and Klaus M. Schmidt. “A theory of fairness, competition, and cooperation”. In: The Quarterly Journal of Economics114.3 (1999), pp. 817–868

    1999

  16. [24]

    Are people conditionally cooperative? Evidence from a public goods experiment

    Urs Fischbacher, Simon Gächter, and Ernst Fehr. “Are people conditionally cooperative? Evidence from a public goods experiment”. In:Economics Letters71.3 (2001), pp. 397–404

    2001

  17. [25]

    Statistical Methods for Research Workers

    Ronald Aylmer Fisher. Statistical Methods for Research Workers. Oliver and Boyd, 1925

    1925

  18. [26]

    Learning with opponent-learning awareness

    Jakob Foerster et al. “Learning with opponent-learning awareness”. In:Proceedings of the 17th International Conference on Autonomous Agents and Multiagent Systems. International Foundation for Autonomous Agents and Multiagent Systems. 2018, pp. 122–130

    2018

  19. [27]

    The long-run benefits of punishment

    Simon Gächter, Elke Renner, and Martin Sefton. “The long-run benefits of punishment”. In: Science 322.5907 (2008), pp. 1510–1510

    2008

  20. [28]

    Unlocking the gameworld: The rewards of space and time in videogames

    Alison Gazzard. “Unlocking the gameworld: The rewards of space and time in videogames”. In: Game Studies11.1 (2011), pp. 9–13

    2011

  21. [29]

    D3C: Reducing the Price of Anarchy in Multi-Agent Learning

    Ian Gemp et al. “D3C: Reducing the Price of Anarchy in Multi-Agent Learning”. In:Proceedings of the 21st International Conference on Autonomous Agents and Multiagent Systems. 2022, pp. 498–506

    2022

  22. [30]

    Gossip and competitive altruism support cooperation in a Public Good game

    Francesca Giardini et al. “Gossip and competitive altruism support cooperation in a Public Good game”. In:Proceedings of the Royal Society B: Biological Sciences376.20200303 (2021)

    2021

  23. [31]

    Princeton university press, 2014

    Herbert Gintis.The bounds of reason: game theory and the unification of the behavioral sciences- revised edition. Princeton university press, 2014

    2014

  24. [32]

    Costly signaling and cooperation

    Herbert Gintis, Eric Alden Smith, and Samuel Bowles. “Costly signaling and cooperation”. In: Journal of theoretical biology213.1 (2001), pp. 103–119

    2001

  25. [33]

    Nice guys finish first: The competitive altruism hypoth- esis

    Charlie L. Hardy and Mark van Vugt. “Nice guys finish first: The competitive altruism hypoth- esis”. In:Personality and Social Psychology Bulletin32.10 (2006), pp. 1402–1413

    2006

  26. [34]

    Artificial intelligence: Chess match of the century

    Demis Hassabis. “Artificial intelligence: Chess match of the century”. In:Nature 544.7651 (2017), pp. 413–414

    2017

  27. [35]

    Costly punishment across human societies

    Joseph P. Henrich et al. “Costly punishment across human societies”. In:Science 312.5781 (2006), pp. 1767–1770

    2006

  28. [36]

    Stylized facts in the social sciences

    Daniel Hirschman. “Stylized facts in the social sciences”. In:Sociological Science3 (2016), pp. 604–626

    2016

  29. [37]

    Longshort-termmemory

    SeppHochreiterandJürgenSchmidhuber.“Longshort-termmemory”.In: NeuralComputation 9.8 (1997), pp. 1735–1780

    1997

  30. [38]

    Inequity aversion improves cooperation in intertemporal social dilem- mas

    Edward Hughes et al. “Inequity aversion improves cooperation in intertemporal social dilem- mas”. In:Advances in Neural Information Processing Systems. 2018, pp. 3330–3340

    2018

  31. [39]

    Processing of social and monetary rewards in the human striatum

    Keise Izuma, Daisuke N. Saito, and Norihiro Sadato. “Processing of social and monetary rewards in the human striatum”. In:Neuron 58.2 (2008), pp. 284–294

    2008

  32. [40]

    Human-level performance in 3D multiplayer games with population- based reinforcement learning

    Max Jaderberg et al. “Human-level performance in 3D multiplayer games with population- based reinforcement learning”. In:Science 364.6443 (2019), pp. 859–865

    2019

  33. [41]

    Introducing ecological dynamics into common-pool resource experiments

    Marco A. Janssen. “Introducing ecological dynamics into common-pool resource experiments”. In: Ecology and Society15.2 (2010)

    2010

  34. [42]

    Robustness of social-ecological systems to spatial and temporal variability

    Marco A. Janssen, John M. Anderies, and Elinor Ostrom. “Robustness of social-ecological systems to spatial and temporal variability”. In:Society and Natural Resources20.4 (2007), pp. 307–322. 41 Modeling human reputation-seeking behavior in a spatio-temporally complex public g...

    2007

  35. [43]

    Lab experiments for the study of social-ecological systems

    Marco A. Janssen, Robert Holahan, et al. “Lab experiments for the study of social-ecological systems”. In:Science 328.5978 (2010), pp. 613–617

    2010

  36. [44]

    Empirically based, agent-based models

    Marco A. Janssen and Elinor Ostrom. “Empirically based, agent-based models”. In:Ecology and Society11.2 (2006)

    2006

  37. [45]

    Social influence as intrinsic motivation for multi-agent deep reinforce- ment learning

    Natasha Jaques et al. “Social influence as intrinsic motivation for multi-agent deep reinforce- ment learning”. In:International Conference on Machine Learning. 2019, pp. 3040–3049

    2019

  38. [46]

    The data model concept in statistical mapping

    George F. Jenks. “The data model concept in statistical mapping”. In:International Yearbook of Cartography7 (Jan. 1967), pp. 186–190

    1967

  39. [47]

    Simpson’s paradox in psychological science: A practical guide

    Rogier Kievit et al. “Simpson’s paradox in psychological science: A practical guide”. In: Frontiers in Psychology4 (2013), p. 513

    2013

  40. [48]

    Coordinate to cooperate or compete: abstract goals and joint intentions in social interaction

    Max Kleiman-Weiner et al. “Coordinate to cooperate or compete: abstract goals and joint intentions in social interaction”. In:CogSci. 2016

    2016

  41. [49]

    Measuring beta diversity for presence– absence data

    Patricia Koleff, Kevin J. Gaston, and Jack J. Lennon. “Measuring beta diversity for presence– absence data”. In:Journal of Animal Ecology72.3 (2003), pp. 367–382

    2003

  42. [50]

    Learning to Participate through Trading of Reward Shares

    Michael Kölle et al. “Learning to Participate through Trading of Reward Shares”. In:arXiv preprint arXiv:2301.07416(2023)

    2023 arXiv

  43. [51]

    Social dilemmas: The anatomy of cooperation

    Peter Kollock. “Social dilemmas: The anatomy of cooperation”. In:Annual Review of Sociology 24.1 (1998), pp. 183–214

    1998

  44. [52]

    Spurious normativity enhances learning of compliance and enforcement behavior in artificial agents

    Raphael Köster et al. “Spurious normativity enhances learning of compliance and enforcement behavior in artificial agents”. In:Proceedings of the National Academy of Sciences119.3 (2022)

    2022

  45. [53]

    Should I tear down this wall? Optimizing social metrics by evaluating novel actions

    János Kramár et al. “Should I tear down this wall? Optimizing social metrics by evaluating novel actions”. In:arXiv preprint arXiv:2004.07625(2020)

    2020 arXiv

  46. [54]

    Cooperation, games, and ecological feedback: Some insights from Bali

    J. Stephen Lansing and JohnH Miller. “Cooperation, games, and ecological feedback: Some insights from Bali”. In:Current Anthropology46.2 (2005), pp. 328–334

    2005

  47. [55]

    Introduction to stochastic processes

    Gregory F Lawler. Introduction to stochastic processes. CRC Press, 2018

    2018

  48. [56]

    Multi-agent Reinforcement Learning in Sequential Social Dilemmas

    Joel Z Leibo, Vinicius Zambaldi, et al. “Multi-agent Reinforcement Learning in Sequential Social Dilemmas”. In:Proceedings of the 16th Conference on Autonomous Agents and MultiAgent Systems. 2017, pp. 464–473

    2017

  49. [57]

    Scalable evaluation of multi-agent reinforce- ment learning with Melting Pot

    Joel Z. Leibo, Edgar A Dueñez-Guzman, et al. “Scalable evaluation of multi-agent reinforce- ment learning with Melting Pot”. In:International Conference on Machine Learning. PMLR. 2021, pp. 6187–6199

    2021

  50. [58]

    Autocurricula and the emergence of innovation from social interaction: A manifesto for multi-agent intelligence research

    Joel Z. Leibo, Edward Hughes, et al. “Autocurricula and the emergence of innovation from social interaction: A manifesto for multi-agent intelligence research”. In:arXiv:1903.00742 (2019)

    2019 arXiv

  51. [59]

    Maintaining cooperation in complex social dilem- mas using deep reinforcement learning

    Adam Lerer and Alexander Peysakhovich. “Maintaining cooperation in complex social dilem- mas using deep reinforcement learning”. In:arXiv preprint arXiv:1707.01068(2017)

    2017 arXiv

  52. [60]

    Markov games as a framework for multi-agent reinforcement learning

    Michael L. Littman. “Markov games as a framework for multi-agent reinforcement learning”. In: Machine learning proceedings 1994. Elsevier, 1994, pp. 157–163

    1994

  53. [61]

    Duncan Luce and Howard Raiffa.Games and decisions: Introduction and critical survey

    R. Duncan Luce and Howard Raiffa.Games and decisions: Introduction and critical survey. Courier Corporation, 1957

    1957

  54. [62]

    The psychological foundations of reputation-based cooperation

    Héctor M Manrique et al. “The psychological foundations of reputation-based cooperation”. In: Philosophical Transactions of the Royal Society B376.1838 (2021), p. 20200287. 42 Modeling human reputation-seeking behavior in a spatio-temporally complex public good provision game

    2021

  55. [63]

    Social diversity and social preferences in mixed-motive rein- forcement learning

    Kevin R McKee, Ian Gemp, et al. “Social diversity and social preferences in mixed-motive rein- forcement learning”. In:Proceedings of the 19th International Conference on Autonomous Agents and Multiagent Systems. International Foundation for Autonomous Agents and Multiagent Sy...

    2020

  56. [64]

    A multi-agent reinforcement learning model of reputation and cooperation in human groups

    Kevin R McKee, Edward Hughes, et al. “A multi-agent reinforcement learning model of reputation and cooperation in human groups”. In:arXiv preprint arXiv:2103.04982(2021)

    2021 arXiv

  57. [65]

    Quantifying the effects of environment and population diversity in multi-agent reinforcement learning

    Kevin R McKee, Joel Z Leibo, et al. “Quantifying the effects of environment and population diversity in multi-agent reinforcement learning”. In:Autonomous Agents and Multi-Agent Systems 36.1 (2022), pp. 1–16

    2022

  58. [66]

    Reputation helps solve the ‘tragedy of the commons’

    Manfred Milinski, Dirk Semmann, and Hans-Jürgen Krambeck. “Reputation helps solve the ‘tragedy of the commons’”. In:Nature 415.6870 (2002), p. 424

    2002

  59. [67]

    Collective action and rational choice: Place, community, and the limits to individual self-interest

    Byron Miller. “Collective action and rational choice: Place, community, and the limits to individual self-interest”. In:Economic Geography68.1 (1992), pp. 22–42

    1992

  60. [68]

    Asynchronous methods for deep reinforcement learning

    Volodymyr Mnih et al. “Asynchronous methods for deep reinforcement learning”. In:Interna- tional Conference on Machine Learning. 2016, pp. 1928–1937

    2016

  61. [69]

    Mollinga.On the Waterfront: Water Distribution, Technology and Agrarian Change in a South Indian Canal Irrigation System

    Peter P. Mollinga.On the Waterfront: Water Distribution, Technology and Agrarian Change in a South Indian Canal Irrigation System. Orient Blackswan, 2003

    2003

  62. [70]

    Institutions

    Douglass C. North. “Institutions”. In:Journal of Economic Perspectives5.1 (1991), pp. 97–112

    1991

  63. [71]

    Evolution of indirect reciprocity by image scoring

    Martin A. Nowak and Karl Sigmund. “Evolution of indirect reciprocity by image scoring”. In: Nature 393.6685 (1998), pp. 573–577

    1998

  64. [72]

    Understanding Institutional Diversity

    Elinor Ostrom. Understanding Institutional Diversity. Princeton University Press, 2005

    2005

  65. [73]

    Coping with asymmetries in the commons: Self-governing irrigation systems can work

    Elinor Ostrom and Roy Gardner. “Coping with asymmetries in the commons: Self-governing irrigation systems can work”. In:Journal of Economic Perspectives7.4 (1993), pp. 93–112

    1993

  66. [74]

    VI. Mathematical contributions to the theory of evolution.—VI. Genetic (reproductive) selection: Inheritance of fertility in man, and of fecundity in thoroughbred racehorses

    Karl Pearson, Alice Lee, and Leslie Bramley-Moore. “VI. Mathematical contributions to the theory of evolution.—VI. Genetic (reproductive) selection: Inheritance of fertility in man, and of fecundity in thoroughbred racehorses”. In:Philosophical Transactions of the Royal Societ...

  67. [75]

    A multi-agent reinforcement learning model of common-pool resource appropriation

    Julien Perolat et al. “A multi-agent reinforcement learning model of common-pool resource appropriation”. In:Advances in Neural Information Processing Systems. 2017, pp. 3643–3652

    2017

  68. [76]

    Reputation for reciprocity engages the brain reward center

    K. Luan Phan et al. “Reputation for reciprocity engages the brain reward center”. In:Proceed- ings of the National Academy of Sciences107.29 (2010), pp. 13099–13104

    2010

  69. [77]

    Poteete, Marco A

    Amy R. Poteete, Marco A. Janssen, and Elinor Ostrom.Working Together: Collective Action, the Commons, and Multiple Methods in Practice. Princeton University Press, 2010

    2010

  70. [78]

    The importance of credo in multiagent learning

    David Radke, Kate Larson, and Tim Brecht. “The importance of credo in multiagent learning”. In: arXiv preprint arXiv:2204.07471(2022)

    2022 arXiv

  71. [79]

    Prisoner’s Dilemma—Recollections and observations

    Anatol Rapoport. “Prisoner’s Dilemma—Recollections and observations”. In:Game Theory as a Theory of a Conflict Resolution. Springer, 1974, pp. 17–34

    1974

  72. [80]

    Competitive altruism: from reciprocity to the handicap principle

    Gilbert Roberts. “Competitive altruism: from reciprocity to the handicap principle”. In:Pro- ceedings of the Royal Society of London. Series B: Biological Sciences265.1394 (1998), pp. 427– 431

    1998

  73. [81]

    The benefits of being seen to help others: indirect reciprocity and reputation-basedpartnerchoice

    Gilbert Roberts et al. “The benefits of being seen to help others: indirect reciprocity and reputation-basedpartnerchoice”.In: PhilosophicalTransactionsoftheRoyalSocietyB 376.1838 (2021), p. 20200290. 43 Modeling human reputation-seeking behavior in a spatio-temporally complex...

    2021

  74. [82]

    The efficient interaction of indirect reciprocity and costly punishment

    Bettina Rockenbach and Manfred Milinski. “The efficient interaction of indirect reciprocity and costly punishment”. In:Nature 444.7120 (2006), p. 718

    2006

  75. [83]

    Mediation analysis in social psychology: Current practices and new recommendations

    Derek D. Rucker et al. “Mediation analysis in social psychology: Current practices and new recommendations”. In:Social and Personality Psychology Compass5.6 (2011), pp. 359–371

    2011

  76. [84]

    For love or money: A common neural currency for social and monetary reward

    Rebecca Saxe and Johannes Haushofer. “For love or money: A common neural currency for social and monetary reward”. In:Neuron 58.2 (2008), pp. 164–165

    2008

  77. [85]

    Schelling

    Thomas C. Schelling. The strategy of conflict. Harvard University Press, 1960

    1960

  78. [86]

    Hockey helmets, concealed weapons, and daylight saving: A study of binary choices with externalities

    Thomas C. Schelling. “Hockey helmets, concealed weapons, and daylight saving: A study of binary choices with externalities”. In:Journal of Conflict Resolution17.3 (1973), pp. 381–428

    1973

  79. [87]

    A more dynamic understanding of human behaviour for the Anthro- pocene

    Caroline Schill et al. “A more dynamic understanding of human behaviour for the Anthro- pocene”. In:Nature Sustainability2.12 (2019), pp. 1075–1082

    2019

  80. [88]

    Strategic investment in reputation

    Dirk Semmann, Hans-Jürgen Krambeck, and Manfred Milinski. “Strategic investment in reputation”. In:Behavioral Ecology and Sociobiology56.3 (2004), pp. 248–252

    2004

  81. [89]

    Stochastic games

    Lloyd S. Shapley. “Stochastic games”. In:Proceedings of the National Academy of Sciences39.10 (1953), pp. 1095–1100

    1953

  82. [90]

    The Psychology of Social Norms

    Muzafer Sherif. The Psychology of Social Norms. 1936

    1936

  83. [91]

    MasteringthegameofGowithouthumanknowledge

    DavidSilveretal.“MasteringthegameofGowithouthumanknowledge”.In: Nature550.7676 (2017), pp. 354–359

    2017

  84. [92]

    The interpretation of interaction in contingency tables

    Edward H. Simpson. “The interpretation of interaction in contingency tables”. In:Journal of the Royal Statistical Society: Series B (Methodological)13.2 (1951), pp. 238–241

    1951

  85. [93]

    Intrinsically motivated reinforcement learning

    Satinder P. Singh, Andrew G. Barto, and Nuttapong Chentanez. “Intrinsically motivated reinforcement learning”. In:Advances in Neural Information Processing Systems. 2005

    2005

  86. [94]

    The economics of rights, co-operation and welfare

    Robert Sugden. The economics of rights, co-operation and welfare. Springer, 1986

    1986

  87. [95]

    Sutton and Andrew G

    Richard S. Sutton and Andrew G. Barto.Reinforcement Learning: An Introduction. MIT Press, 1998

    1998

  88. [96]

    Multi-agent reinforcement learning: Independent vs. cooperative agents

    Ming Tan. “Multi-agent reinforcement learning: Independent vs. cooperative agents”. In: Proceedings of the tenth international conference on machine learning. 1993, pp. 330–337

    1993

  89. [97]

    Agent-based computational economics: Overview and brief history

    Leigh Tesfatsion. “Agent-based computational economics: Overview and brief history”. In: Artificial Intelligence, Learning and Computation in Economics and Finance. Springer, 2023

    2023

  90. [98]

    Lecture 6.5-RMSProp: Divide the gradient by a running average of its recent magnitude

    Tijmen Tieleman and Geoffrey Hinton. “Lecture 6.5-RMSProp: Divide the gradient by a running average of its recent magnitude”. In:Coursera: Neural networks for machine learning 4.2 (2012), pp. 26–31

    2012

  91. [99]

    Identity and Dynamic Teams in Social Dilemmas

    Kyle Tilbury and Jesse Hoey. “Identity and Dynamic Teams in Social Dilemmas”. In:arXiv preprint arXiv:2208.03293(2022)

    2022 arXiv

  92. [100]

    Successfully governing the commons: Principles of social organization in an Andean irrigation system

    Paul B. Trawick. “Successfully governing the commons: Principles of social organization in an Andean irrigation system”. In:Human Ecology29.1 (2001), pp. 1–25

    2001

  93. [101]

    A learning agent that acquires social norms from public sanctions in de- centralizedmulti-agentsettings

    Eugene Vinitsky et al. “A learning agent that acquires social norms from public sanctions in de- centralizedmulti-agentsettings”.In: CollectiveIntelligence 2.2(2023),p.26339137231162025

    2023

  94. [102]

    Grandmaster level in StarCraft II using multi-agent reinforcement learn- ing

    Oriol Vinyals et al. “Grandmaster level in StarCraft II using multi-agent reinforcement learn- ing”. In:Nature (2019), pp. 1–5

    2019

  95. [103]

    Cooperation for reputation: Wasteful contributions as costly signals in public goods

    Mark van Vugt and Charlotte L. Hardy. “Cooperation for reputation: Wasteful contributions as costly signals in public goods”. In:Group Processes & Intergroup Relations13.1 (2010), pp. 101–111. 44 Modeling human reputation-seeking behavior in a spatio-temporally complex public ...

    2010

  96. [104]

    Evolving intrinsic motivations for altruistic behavior

    Jane X. Wang et al. “Evolving intrinsic motivations for altruistic behavior”. In:Proceedings of the 18th International Conference on Autonomous Agents and Multiagent Systems. International Foundation for Autonomous Agents and Multiagent Systems. 2019, pp. 683–692

    2019

  97. [105]

    Cooperation through image scoring in humans

    Claus Wedekind and Manfred Milinski. “Cooperation through image scoring in humans”. In: Science 288.5467 (2000), pp. 850–852

    2000

  98. [106]

    Evolutionary game theory

    Jörgen W Weibull. Evolutionary game theory. MIT Press, 1997

    1997

  99. [107]

    Vegetation of the Siskiyou mountains, Oregon and California

    Robert Harding Whittaker. “Vegetation of the Siskiyou mountains, Oregon and California”. In: Ecological Monographs30.3 (1960), pp. 279–338

    1960

  100. [108]

    Deep multiagent reinforcement learning: Challenges and directions

    Annie Wong et al. “Deep multiagent reinforcement learning: Challenges and directions”. In: Artificial Intelligence Review(2022), pp. 1–34

    2022

  101. [109]

    Learning to incentivize other learning agents

    Jiachen Yang et al. “Learning to incentivize other learning agents”. In:Advances in Neural Information Processing Systems33 (2020), pp. 15208–15219

    2020

  102. [110]

    Oxford University Press, 1999

    Amotz Zahavi and Avishag Zahavi.The handicap principle: A missing piece of Darwin’s puzzle. Oxford University Press, 1999

    1999

  103. [111]

    Linear public goods experiments: A meta-analysis

    Jennifer Zelmer. “Linear public goods experiments: A meta-analysis”. In:Experimental Eco- nomics 6.3 (2003), pp. 299–310. 45

    2003

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.