Pith. sign in

REVIEW 2 major objections 4 minor 50 references

Can Media Act as a Soft Regulator of Safe AI Development? A Game Theoretical Analysis

T0 review · 2 major / 4 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read The paper claims that imperfect, costly media coverage can act as a soft regulator, sustaining safe AI creation and adoption, but only when reporting accuracy is high enough and safety and investigation costs are not too high.

desk verdict The paper asks a good question and has a plausible qualitative answer, but the replicator-dynamics analysis contains a formal error that undercuts the quantitative claims; the ABM results and honest limitations partially compensate. read the letter →

arxiv 2509.02650 v1 pith:IRSW4J2E submitted 2025-09-02 cs.AI cs.GTq-bio.PE

Henrique Correia da Fonseca , António Fernandes , Zhao Song , Theodor Cimpeanu , Nataliya Balabanova , Adeela Bashir , Paolo Bova , Alessio Buscemi
show 10 more authors
Alessandro Di Stefano Manh Hong Duong Elias Fernandez Domingos Ndidi Bianca Ogbo Simon T. Powers Daniele Proverbio Zia Ush Shamszaman Fernando P. Santos The Anh Han Marcus Krellner
This is my paper
classification cs.AIcs.GTq-bio.PE MSC 91A2291A80
keywords AIsafetysoftregulationevolutionarygametheorymediareliabilityindirectreciprocityreplicatordynamicsagent-basedsimulationadoption
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 asks whether public reporting alone—without government enforcement—can push profit-driven AI developers to make safe products and users to adopt them. It builds an evolutionary game with creators choosing safe vs unsafe AI, and users choosing among refusal, blind adoption, and following either a cheap random or a costly reliable media recommendation. The answer is a conditional yes: reliable media can sustain cooperation across a wide parameter range, but if reporting is too noisy, too expensive, or safe development too costly, cooperation collapses to universal defection. Because the model yields both oscillatory cooperation and a stable defection equilibrium, the same media landscape can end in either outcome depending on starting conditions. These findings matter because they map directly onto the current AI regulatory gap, where formal oversight is still maturing while media scrutiny is already operating.

What carries the argument

The central object is a four-strategy user population (AllD, BMedia, GMedia, AllC) matched against a two-strategy creator population (unsafe D, safe C), with payoffs averaged over the probabilistic media signal. The load-bearing identity is the expected-payoff comparison in the replicator dynamics: a GMedia user earns q·bu − ci from safe creators and −((1−q)·cu + ci) from unsafe ones, while a BMedia user gets half the payoffs at no cost. These payoff differences determine when discriminating users can invade, when safe creators can survive, and why the system cycles rather than settling into full cooperation.

What would settle it

A decisive check would be calibrating the model to a real AI product domain with measured media report accuracy, investigation cost, and safety cost, then observing whether safe adoption actually rises when the parameters fall inside the predicted cooperative region; if safety adoption stays near zero under those conditions, the claimed threshold structure is wrong. A complementary test would endogenize q by letting media choose accuracy against profit and seeing whether sustained cooperation collapses.

Watch

Extended reading notes

Core claim

In an evolving population of AI creators who choose safe or unsafe development and users who choose to adopt or refuse, the paper claims that a media signal—imperfect and costly—can act as a soft regulator. Good media identify a creator's strategy correctly with probability q at per-user cost ci; bad media offer random signals for free. Across replicator dynamics and agent-based simulations, cooperation (safe creation plus adoption) dominates for a wide parameter region, provided q is sufficiently high and cc and ci are sufficiently low; outside that region the only attractor is universal defection. The model also identifies bistability: high-cooperation cycles coexist with a stable defectio

Load-bearing premise

The model treats media accuracy as a fixed probability q that costs users ci to obtain, with media passively relaying signals rather than acting on their own interests; if real media accuracy depends on bias, budgets, and strategic choices, the thresholds could shift or cooperation might not get off the ground.

Editorial extensions

If this is right

  • If media accuracy stays above a threshold and media and safety costs remain moderate, self-interested creators and users can sustain safe AI production and adoption without formal regulation.
  • When good media are too expensive, too noisy, or safe creation too costly, the only stable outcome is universal defection: users refuse AI and creators produce unsafe AI.
  • Cooperation typically arrives through cycles, not equilibrium: good-media users rise, safe creators follow, users become complacent, unsafe creators exploit them, and good media becomes valuable again.
  • Initial population composition matters: with identical parameters, starting near defection can collapse the system while more cooperative starts sustain cycles, so the same media landscape can produce opposite outcomes.
  • The finite-population agent-based model reproduces the analytical results' qualitative patterns, so the threshold structure is not an artifact of infinite-population assumptions.

Reading between the lines

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

  • If media accuracy is itself shaped by market incentives, the paper's threshold result suggests a policy implication the authors only gesture at: subsidizing or certifying investigative quality may be at least as important to AI safety as direct regulation.
  • The paper's cycles imply that reputational pressure alone does not settle into a steady state; one observable signature would be alternating waves of exposé-driven caution and complacent mass adoption.
  • A natural extension, flagged as future work by the authors, replaces single-source users with users integrating many contradictory media reports; whether cooperation survives then likely depends on how users aggregate reputational signals.
  • The model's bistability means public-trust starting points may be as decisive as economics; interventions might focus on maintaining a critical mass of discriminating users rather than only lowering costs.
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

2 major / 4 minor

Summary. The paper asks whether media can act as a 'soft regulator' of safe AI development in the absence of formal regulation. It introduces a two-population evolutionary game in which creators choose between safe (C) and unsafe (D) AI development, and users choose among always defect (AllD), follow an unreliable media outlet (BMedia, random signal, no cost), follow a reliable outlet (GMedia, costly signal with accuracy q), or always cooperate (AllC). Payoffs are written down and analyzed with two methods: replicator dynamics for infinite populations and agent-based simulations with mutation and imitative update rules. The authors report that good media can sustain cooperation over a wide parameter region, provided q is sufficiently high and the costs of safe creation (cc) and informed access (ci) are not too large; otherwise the population collapses to universal defection. They also report bistability and persistent oscillations between GMedia and AllC users and between C and D creators. Code and data are provided through an anonymous repository.

Significance. The question is timely and the model is a clear, self-contained stylization of an important governance mechanism. The paper's main potential contribution is to isolate the effect of media as a reputational channel in AI safety, using a transparent payoff structure and an agent-based model with stochastic update rules. The ABM is a well-posed finite-population model, and the qualitative idea that costly, noisy media information can sustain cooperation when it is sufficiently reliable is plausible. However, the replicator-dynamics pillar, which underlies the quantitative threshold curves, the stability analysis, the basin-of-attraction percentage, and the claimed indirect-reciprocity cycles, is mathematically flawed. As written, Eq. (1) is not the standard replicator equation, so the paper's quantitative claims are not currently supported. The ABM provides independent qualitative evidence but cannot validate the specific RD-derived numbers.

major comments (2)
  1. [Replicator Dynamics, Eq. (1)] Eq. (1) is not the standard replicator dynamics. For a single population with n strategies the standard equation is x_i' = x_i(pi_i - pi_bar), not x_i(1-x_i)(pi_i - pi_bar). The factor (1-x_i) is a two-strategy simplification and is not valid for n>2. For the creator population (n=2), the standard equation would be y' = y(1-y)(pi_C - pi_D), whereas Eq. (1) gives y(1-y)(pi_C - pi_bar_creator) = y(1-y)^2(pi_C - pi_D), a different vector field. Applied symmetrically to the four user strategies, the sum of the derivatives is -sum_i x_i^2(pi_i - pi_bar_user), which is not zero, so simplex invariance fails. Because x4 is defined as a residual, the system can still be integrated on a three-dimensional simplex, but it is not the replicator dynamics and its equilibria have no standard EGT interpretation. This invalidates the quantitative results that rely on this ODE: the threshold curves in Fig.
  2. [Results, Figures 3 and 4; Agent-Based Simulations] The agent-based model cannot validate the replicator-dynamics-derived quantitative thresholds. The ABM uses a different process: finite populations, mutation, pairwise Fermi imitation, and a different parameter set (population sizes, beta, mutation rates). Therefore the statement that the ABM 'provides robustness' and that the findings are 'robust to both analytical predictions and agent-based simulations' is an overstatement. At most, the ABM gives qualitative support for the existence of cooperation under some parameters. After correcting the replicator dynamics, the authors should compare the corrected RD predictions with the ABM on the same parameter grid (or explicitly argue why the two processes are expected to agree qualitatively). As it stands, the quantitative thresholds in Fig. 3 are unsupported, and Fig. 4's agreement with Fig. 3 does not repair that.
minor comments (4)
  1. [General notation] There are several typos in strategy names: 'piBM edia' and 'GM edia' appear in Eqs. (1) and (2) and the text; these should be cleaned up.
  2. [Figure 5 caption] The caption states 'initially 50% C' and 'initially 45% C' for the creator population but does not specify the initial user-population composition. Please give the full initial condition for all four user strategies and both creator strategies.
  3. [Results, robustness paragraph] The sentence 'All the results ... are robust to variations ... as long as their order of magnitude remains above a certain threshold (beta >= 1 and mu >= 0.1)' is vague. It is not clear which quantities were varied, over what ranges, or what 'robust' means quantitatively; a supplementary figure would help.
  4. [References] Some references are preprint/future-dated (e.g., Alalawi et al., 2026); please check the published/arXiv status and update where possible.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the model is a self-contained parameter study; media-as-soft-regulator result follows from explicitly stated dynamics, not from fitted inputs or self-citations.

full rationale

The paper's derivation chain is self-contained. The central claim—that a reliable-enough GMedia signal can sustain cooperation between users and creators when ci and cc are not too high—is obtained by numerically integrating the replicator equations (Eqs. 1–2) and by independent agent-based simulations (Eq. 3, Fig. 4). The key quantities q, ci, and cc are exogenous parameters that are swept across ranges (Fig. 3), not estimated or fitted to the outcome; the paper does not rename a fitted parameter as a prediction. The cooperation metric η in Eq. (4) is a definitional aggregation of strategy frequencies, and reporting its values is not circular because the dynamics that produce those frequencies are specified independently. Self-citations (e.g., Balabanova et al. 2025; Han et al. 2019–2022) appear as motivation and related work, not as the justification for the media-alone result; the claim that media alone can regulate is tested by the present model, not imported from the cited papers. The limitations section explicitly acknowledges the exogenous-q assumption and proposes future bias/expenditure extensions, which is a modeling limitation rather than a circular step. The non-standard form of Eq. (1) noted by a skeptical reader is a formal correctness concern about the dynamical system, but it is not a reduction of the conclusion to an input of the model, so it does not constitute circularity under the specified patterns.

Assumptions & free parameters 6 free parameters · 4 assumptions · 2 invented entities

The model uses several hand-chosen payoff parameters (bc, bu, cu) and sweeps key cost/reliability parameters (cc, ci, q). No parameters are fitted to empirical data. The central claim rests on domain assumptions about media behavior and on the standard EGT framework, plus the exogenous quality of media signals.

free parameters (6)
  • b_c (creator benefit from adoption) = 0.4 (baseline)
    Chosen for the simulations; not fitted to data. Results are qualitative across varied values.
  • b_u (user benefit from safe adoption) = 0.4 (baseline)
    Chosen for the simulations; not fitted to data.
  • c_u (user cost of unsafe adoption) = 0.8 (baseline)
    Chosen for the simulations; not fitted to data.
  • c_c (cost of safe creation) = varied, baseline 0.1-0.2
    Key control parameter; varied to explore thresholds.
  • c_i (cost of informed recommendation) = varied, baseline 0.05-0.1
    Key control parameter; varied to explore thresholds.
  • q (media reliability for good media) = varied, baseline 0.9
    Key control parameter; bad media fixed at q=0.5.
assumptions (4)
  • standard math Replicator dynamics describes strategy evolution in infinite well-mixed populations.
    Used to derive Figures 3 and 5, but note the non-standard (1-x_i) factor in Eq. 1.
  • domain assumption Media signal accuracy q is exogenous and cost-dependent, with q=0.5 for BMedia and q>0.5 for GMedia.
    Introduced in Model and Methods; this is the weakest assumption (see weakest_assumption).
  • domain assumption Users choose a single media source, and media do not act strategically.
    Stated in Model and Methods and acknowledged in Limitations.
  • domain assumption Payoffs are linear and additive as in Table 2; no externalities beyond user/creator payoff.
    Simplifies the model; acknowledged in Limitations (no societal-wide effects).
invented entities (2)
  • GMedia (good media commentator)
    purpose: Provides a signal with reliability q to users in exchange for cost ci.
    Abstract model construct representing investigative journalism; no falsifiable handle outside the model.
  • BMedia (bad media commentator)
    purpose: Provides random (q=0.5) signals at no cost, representing low-effort media.
    Model construct; no independent evidence.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Can Media Act as a Soft Regulator of Safe AI Development? A Game Theoretical Analysis." pith.science (2026). https://pith.science/paper/IRSW4J2E

@misc{pith2026250902650,
  author       = {Pith},
  title        = {Pith review of: Can Media Act as a Soft Regulator of Safe AI Development? A Game Theoretical Analysis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IRSW4J2E}},
  note         = {Machine review of arXiv:2509.02650}
}
read the original abstract

When developers of artificial intelligence (AI) products need to decide between profit and safety for the users, they likely choose profit. Untrustworthy AI technology must come packaged with tangible negative consequences. Here, we envisage those consequences as the loss of reputation caused by media coverage of their misdeeds, disseminated to the public. We explore whether media coverage has the potential to push AI creators into the production of safe products, enabling widespread adoption of AI technology. We created artificial populations of self-interested creators and users and studied them through the lens of evolutionary game theory. Our results reveal that media is indeed able to foster cooperation between creators and users, but not always. Cooperation does not evolve if the quality of the information provided by the media is not reliable enough, or if the costs of either accessing media or ensuring safety are too high. By shaping public perception and holding developers accountable, media emerges as a powerful soft regulator -- guiding AI safety even in the absence of formal government oversight.

Figures

Figures reproduced from arXiv: 2509.02650 by the authors.

Figure 1
Figure 1. Visual description of the AI regulatory ecosys￾tem. Users decide whether or not to use AI products, in￾curring a cost for adoption. For this decision process, they can choose to follow media recommendations by paying a small amount in exchange for their information about cre￾ators’ strategies. Meanwhile, creators decide whether to create safe or unsafe technology; safe technology further involves additional costs. M… view at source ↗
Figure 2
Figure 2. Media recommendations. Media that investigate creators’ strategies can provide more reliable recommenda￾tions, although incurring an extra cost. Media that do not investigate will provide random recommendations to users. The resulting payoffs are shown in [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Average cooperation ratio η via replicator dynamics, across parameters of interest. If not varied, q = 0.9, ci = 0.1, and cc = 0.1, other parameters: bc = 0.4, bu = 0.4, cu = 0.8 [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Average cooperation ratio η via agent-based simulations, across parameters of interest. Each data point shows the average cooperation ratio η averaged over R = 100 runs. All parameters are set to the same ones as [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 6
Figure 6. Figure 6: Evolution of strategies and cooperation over time for the agent-based model of user (top) and creator (middle) strategies, alongside the average cooperation ra￾tio (ACR, or η) (bottom), for a typical run of the simula￾tion (R = 1) with NU = 2NC = 100, µu = 1/NU = 0.005…
Figure 7
Figure 7. Figure 7: Initial state with only defective strategies. Evo￾lution over simulation-time of user (top) and creator (strate￾gies, as well as cooperation levels (bottom), where the ini￾tial population state is that of only AllD users and D cre￾ators. Colored shaded areas represent …

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

50 extracted references · 46 canonical work pages

  1. [1]

    H., Domingos, E

    Alalawi, Z., Bova, P., Cimpeanu, T., Di Stefano, A., Duong, M. H., Domingos, E. F., Han, T. A., Krellner, M., Ogbo, N. B., Powers, S. T., et al. (2026). Trust ai regulation? discerning users are vital to build trust and effective ai regulation. Applied Mathematics and Computation , 508:129627

    work page 2026

  2. [2]

    C., Di Stefano, A., Duong, M

    Balabanova, N., Bashir, A., Bova, P., Buscemi, A., Cimpeanu, T., da Fonseca, H. C., Di Stefano, A., Duong, M. H., Domingos, E. F., Fernandes, A., et al. (2025). Media and responsible ai governance: a game-theoretic and llm analysis. arXiv preprint arXiv:2503.09858

    arXiv 2025

  3. [3]

    Baron, D. P. (2006). Persistent media bias. Journal of Public Economics , 90(1):1--36

    work page 2006

  4. [4]

    Bedau, M. A. (2003). Artificial life: organization, adaptation and complexity from the bottom up. Trends in cognitive sciences , 7(11):505--512

    work page 2003

  5. [5]

    Bengio, Y., Mindermann, S., Privitera, D., Besiroglu, T., Bommasani, R., Casper, S., Choi, Y., Fox, P., Garfinkel, B., Goldfarb, D., et al. (2025). International ai safety report. arXiv preprint arXiv:2501.17805

    arXiv 2025

  6. [6]

    Bova, P., Di Stefano, A., and Han, T. A. (2024). Both eyes open: Vigilant incentives help auditors improve ai safety. Journal of Physics: Complexity , 5(2):025009

    work page 2024

  7. [7]

    and Richerson, P

    Boyd, R. and Richerson, P. J. (1989). The evolution of indirect reciprocity. Social Networks , 11:213--236

    work page 1989

  8. [8]

    C., Duong, M

    Buscemi, A., Proverbio, D., Bova, P., Balabanova, N., Bashir, A., Cimpeanu, T., da Fonseca, H. C., Duong, M. H., Domingos, E. F., Fernandes, A. M., et al. (2025). Do LLMs trust AI regulation? Emerging behaviour of game-theoretic LLM agents . arXiv preprint arXiv:2504.08640

    arXiv 2025

Show all 50 references
  1. [9]

    and Li, C

    Cao, Y. and Li, C. (2020). The influence mechanism of reputation information on the formation of safety trust in chinese infant milk powder . Healthcare (Switzerland) , 8(2):1--16

    2020

  2. [10]

    Carr, J. (2012). Applications of centre manifold theory , volume 35. Springer Science & Business Media

    2012

  3. [11]

    Cimpeanu, T., Santos, F., et al. (2022). Artificial Intelligence Development Races in Heterogeneous Settings . Scientific Reports , 12(1):1723

    2022

  4. [12]

    K., Kolt, N., Bengio, Y., Hadfield, G

    Cohen, M. K., Kolt, N., Bengio, Y., Hadfield, G. K., and Russell, S. (2024). Regulating advanced artificial agents. Science , 384(6691):36--38

    2024

  5. [13]

    sztucznej inteligencji, G

    Commission, E., Directorate-General for Communications Networks, C., Technology, and ekspertów wysokiego szczebla ds. sztucznej inteligencji, G. (2019). Ethics guidelines for trustworthy AI . Publications Office

    2019

  6. [14]

    C., Kather, J

    Freyer, O., Wiest, I. C., Kather, J. N., and Gilbert, S. (2024). A future role for health applications of large language models depends on regulators enforcing safety standards. The Lancet Digital Health , 6(9):e662--e672. Publisher: Elsevier

    2024

  7. [15]

    Gershenson, C., Trianni, V., Werfel, J., and Sayama, H. (2020). Self-organization and artificial life. Artificial Life , 26(3):391--408

    2020

  8. [16]

    A., Hughes, E., Kovařík, V., Kulveit, J., Leibo, J

    Hammond, L., Chan, A., Clifton, J., Hoelscher-Obermaier, J., Khan, A., McLean, E., Smith, C., Barfuss, W., Foerster, J., Gavenčiak, T., Han, T. A., Hughes, E., Kovařík, V., Kulveit, J., Leibo, J. Z., Oesterheld, C., de Witt, C. S., Shah, N., Wellman, M., Bova, P., Cimpeanu, T....

    2025 arXiv

  9. [17]

    A., Lenaerts, T., et al

    Han, T. A., Lenaerts, T., et al. (2022). Voluntary Safety Commitments Provide an Escape from Over-Regulation in AI Development . Technology in Society , 68:101843

    2022

  10. [18]

    A., Pereira, L

    Han, T. A., Pereira, L. M., et al. (2020). To Regulate or Not: A Social Dynamics Analysis of an Idealised AI Race . Journal of Artificial Intelligence Research , 69:881--921

    2020

  11. [19]

    A., Pereira, L

    Han, T. A., Pereira, L. M., and Lenaerts, T. (2019). Modelling and influencing the AI bidding war: a research agenda . In Proceedings of the 2019 AAAI/ACM Conference on AI, Ethics, and Society , pages 5--11

    2019

  12. [20]

    Hern, A. (2019). Apple apologises for allowing workers to listen to siri recordings | apple | the guardian

    2019

  13. [21]

    Hilbe, C., Schmid, L., Tkadlec, J., Chatterjee, K., and Nowak, M. A. (2018). Indirect reciprocity with private, noisy, and incomplete information. Proceedings of the National Academy of Sciences , 115:12241--12246

    2018

  14. [22]

    and Sigmund, K

    Hofbauer, J. and Sigmund, K. (1998). Evolutionary games and population dynamics . Cambridge university press

    1998

  15. [23]

    A., Fudenberg, D., and Nowak, M

    Imhof, L. A., Fudenberg, D., and Nowak, M. A. (2005). Evolutionary cycles of cooperation and defection . Proceedings of the National Academy of Sciences of the United States of America , 102(31):10797--10800

    2005

  16. [24]

    Jøsang, A., Ismail, R., and Boyd, C. (2007). A survey of trust and reputation systems for online service provision. Decision Support Systems , 43:618--644

    2007

  17. [25]

    Kelly, J. (2024). Google’s ai recommended adding glue to pizza and other misinformation—what caused the viral blunders?

    2024

  18. [26]

    Khalil, H. K. and Grizzle, J. W. (2002). Nonlinear systems , volume 3. Prentice hall Upper Saddle River, NJ

    2002

  19. [27]

    and Han, T

    Krellner, M. and Han, T. A. (2021). Pleasing enhances indirect reciprocity-based cooperation under private assessment. Artificial Life , 27(3--4):246--276

    2021

  20. [28]

    Massaro, D. W. and Friedman, D. (1990). Models of integration given multiple sources of information. Psychological Review , 97(2):225--252

    1990

  21. [29]

    and Kleinman, Z

    McMahon, L. and Kleinman, Z. (2024). Glue pizza and eat rocks: Google ai search errors go viral

    2024

  22. [30]

    J., Araque-Padilla, R

    Melero-Bola \ n os, R., Guti \' e rrez-Villar, B., Montero-Simo, M. J., Araque-Padilla, R. A., and Olarte-S \' a nchez, C. M. (2025). Media Influence on the Perceived Safety of Dietary Supplements for Children: A Content Analysis of Spanish News Outlets . Nutrients , 17(6):1--14

    2025

  23. [31]

    Nowak, M. A. and Sigmund, K. (1998). Evolution of indirect reciprocity by image scoring. Nature , 393:573--577

    1998

  24. [32]

    Nowak, M. A. and Sigmund, K. (2005). Evolution of indirect reciprocity. Nature , 437(7063):1291--1298

    2005

  25. [33]

    Okada, I. (2020). A review of theoretical studies on indirect reciprocity. Games , 11(3):27

    2020

  26. [34]

    Paiva, A., Santos, F. P. F. C., and Santos, F. P. F. C. (2018). Engineering pro-sociality with autonomous agents. In AAAI , volume 32, pages 7994--7999

    2018

  27. [35]

    J., Rand, D

    Perc, M., Jordan, J. J., Rand, D. G., Wang, Z., Boccaletti, S., and Szolnoki, A. (2017). Statistical physics of human cooperation. Physics Reports , 687:1--51

    2017

  28. [36]

    Piper, K. (2024). Openai ndas: Leaked documents reveal aggressive tactics toward former employees | vox

    2024

  29. [37]

    T., Ek \'a rt, A., and Lewis, P

    Powers, S. T., Ek \'a rt, A., and Lewis, P. R. (2018). Modelling enduring institutions: The complementarity of evolutionary and agent-based approaches. Cognitive Systems Research , 52:67--81

    2018

  30. [38]

    T., Linnyk, O., et al

    Powers, S. T., Linnyk, O., et al. (2023). The Stuff We Swim in: Regulation Alone Will Not Lead to Justifiable Trust in AI . IEEE Technology and Society Magazine , 42(4):95--106

    2023

  31. [39]

    Prada, L. (2025). ChatGPT Briefly Made Chat Logs Accessible on Google . Yikes

    2025

  32. [40]

    Santos, F., Pacheco, J., and Santos, F. (2018). Social norms of cooperation with costly reputation building. Proceedings of the AAAI Conference on Artificial Intelligence , 32:4727--4734

    2018

  33. [41]

    P., Lelkes, Y., and Levin, S

    Santos, F. P., Lelkes, Y., and Levin, S. A. (2021a). Link recommendation algorithms and dynamics of polarization in online social networks. Proceedings of the National Academy of Sciences of the United States of America , 118:e2102141118

  34. [42]

    P., Pacheco, J

    Santos, F. P., Pacheco, J. M., and Santos, F. C. (2021b). The complexity of human cooperation under indirect reciprocity. Philosophical Transactions of the Royal Society B: Biological Sciences , 376

  35. [43]

    Sayama, H. (2015). Introduction to the modeling and analysis of complex systems . Open SUNY Textbooks [Imprint]

    2015

  36. [44]

    Sigmund, K. (2010). The calculus of selfishness. In The Calculus of Selfishness . Princeton University Press

    2010

  37. [45]

    D., Krambeck, H.-J., Semmann, D., and Milinski, M

    Sommerfeld, R. D., Krambeck, H.-J., Semmann, D., and Milinski, M. (2007). Gossip as an alternative for direct observation in games of indirect reciprocity. Proceedings of the National Academy of Sciences , 104:17435--17440

    2007

  38. [46]

    A., and Pacheco, J

    Traulsen, A., Nowak, M. A., and Pacheco, J. M. (2006). Stochastic dynamics of invasion and fixation. Phys. Rev. E , 74:11909

    2006

  39. [47]

    Wu, X., Duan, R., and Ni, J. (2024). Unveiling security, privacy, and ethical concerns of ChatGPT . Journal of Information and Intelligence , 2(2):102--115

    2024

  40. [48]

    Xia, C., Wang, J., Perc, M., and Wang, Z. (2023). Reputation and reciprocity. Physics of Life Reviews , 46:8--45

    2023

  41. [49]

    M., Bao, L., Calice, M

    Yang, S., Krause, N. M., Bao, L., Calice, M. N., Newman, T. P., Scheufele, D. A., Xenos, M. A., and Brossard, D. (2023). In ai we trust: The interplay of media use, political ideology, and trust in shaping emerging ai attitudes. Journalism & Mass Communication Quarterly , page...

    2023

  42. [50]

    Zhang, J., Wu, H.-C., Chen, L., and Su, Y. (2022). Effect of social media use on food safety risk perception through risk characteristics: Exploring a moderated mediation model among people with different levels of science literacy . Frontiers in Psychology , 13(September):1--14

    2022

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.