pith. sign in

arxiv: 2606.27068 · v1 · pith:6R7QCBBDnew · submitted 2026-06-25 · 💻 cs.GT · cs.AI· cs.LG

Parametric Open Source Games

Aleksandar Todorov , Jesse ten Napel , Alexander M\"uller This is my paper

Pith reviewed 2026-06-26 01:57 UTC · model grok-4.3

classification 💻 cs.GT cs.AIcs.LG
keywords parametric open-source gamesprogram equilibriagradient ascentcoupling thresholdcooperationsymmetric 2x2 gamesneural semantics
0
0 comments X

The pith

Parametric open-source games yield an exact coupling threshold where selfish gradient ascent switches to cooperation in symmetric 2x2 games.

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

The paper introduces parametric open-source games as a continuous analogue of program equilibria. Players choose parameter vectors whose semantics maps produce mixed actions in an underlying finite game. Equilibrium existence is established along with a one-dimensional boundary test for parametric program Nash equilibria. The central result derives the precise coupling threshold at which gradient ascent in symmetric 2x2 games switches from defection to cooperation. An extension to neural semantics shows that the first-order condition for cooperation is set by the ratio of cross-player to self-player sensitivity.

Core claim

In parametric open-source games, players choose parameter vectors and semantics maps convert the full parameter profile into mixed actions in an underlying finite game. Equilibrium existence results hold, and an exact coupling threshold is derived at which selfish gradient ascent in symmetric 2×2 games switches from defection toward cooperation. A one-dimensional boundary test identifies parametric program Nash equilibria. The framework extends to a neural semantics class whose first-order cooperation condition is governed by the ratio of cross-player to self-player sensitivity, showing how sufficiently strong open-source coupling steers selfish optimization toward cooperative outcomes.

What carries the argument

Semantics maps that convert the full parameter profile into mixed actions, enabling continuous open-source coupling and well-defined gradient-ascent dynamics.

If this is right

  • Equilibrium existence results hold for the parametric model.
  • A one-dimensional boundary test identifies parametric program Nash equilibria.
  • Access to internal parameterizations qualitatively reshapes learning dynamics and equilibrium structure.
  • Strong open-source coupling steers selfish optimization toward cooperative outcomes across canonical games.

Where Pith is reading between the lines

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

  • The coupling threshold derived for 2x2 symmetric games may extend to asymmetric or multi-player settings under analogous smoothness assumptions on the semantics maps.
  • System designers could introduce controlled parameter sharing to induce cooperation in learning agents without altering the underlying payoff matrix.
  • Empirical tests of the neural semantics sensitivity ratio in multi-agent reinforcement learning environments could check whether the first-order cooperation condition predicts observed behavior.

Load-bearing premise

The semantics maps that convert the full parameter profile into mixed actions are assumed to exist, be sufficiently smooth, and allow the gradient-ascent dynamics to be well-defined without additional regularity conditions that might alter the threshold.

What would settle it

A numerical simulation of gradient ascent in a symmetric 2x2 game such as the Prisoner's Dilemma in which the switch from defection to cooperation occurs at a coupling value different from the derived threshold.

Figures

Figures reproduced from arXiv: 2606.27068 by Aleksandar Todorov, Alexander M\"uller, Jesse ten Napel.

Figure 1
Figure 1. Figure 1: Open-source versus closed-source sigmoid semantics in the Prisoner’s Dilemma. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Phase transition in 2 × 2 games. Panel (a) shows mean terminal cooperation as a function of γ, with dotted lines marking γ ⋆ and the dashed line marking ¯p = 0.5. Panel (b) shows normalized social welfare, and Panel (c) compares analytical and empirical transition points. All results are averaged over 20 seeds. 3.2. Boundary Equilibria The result from Theorem 3 describes only the local direction of gradien… view at source ↗
Figure 3
Figure 3. Figure 3: Boundary PPNE verification in the Prisoner’s Dilemma. For each candidate [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Learning curves under neural open-source semantics across canonical 2 [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Boundary best-response verification for a non-PPNE open-source candidate. Us [PITH_FULL_IMAGE:figures/full_fig_p017_5.png] view at source ↗
read the original abstract

Open-source game theory studies agents whose behavior may depend on one another's decision procedures, but most existing models use discrete or symbolic programs. We introduce parametric open-source games, a continuous analogue of program equilibria in which players choose parameter vectors and semantics maps convert the full parameter profile into mixed actions in an underlying finite game. We establish equilibrium existence results, derive an exact coupling threshold at which selfish gradient ascent in symmetric $2\times2$ games switches from defection toward cooperation, and give a one-dimensional boundary test for parametric program Nash equilibria. We further extend the framework to a neural semantics class whose first-order cooperation condition is governed by the ratio of cross-player to self-player sensitivity. Across canonical games, the framework shows how access to internal parameterizations can qualitatively reshape learning dynamics and equilibrium structure, and how sufficiently strong open-source coupling can steer selfish optimization toward cooperative outcomes.

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.

Referee Report

1 major / 2 minor

Summary. The manuscript introduces parametric open-source games as a continuous analogue of program equilibria. Players select parameter vectors whose semantics maps convert the full profile into mixed actions of an underlying finite game. The authors establish equilibrium existence, derive an exact coupling threshold at which selfish gradient ascent in symmetric 2×2 games switches from defection to cooperation, supply a one-dimensional boundary test for parametric program Nash equilibria, and extend the framework to a neural-semantics class in which the first-order cooperation condition is governed by the ratio of cross-player to self-player sensitivity. Illustrations across canonical games show how parameter access qualitatively alters learning dynamics and how sufficiently strong coupling can steer selfish optimization toward cooperative outcomes.

Significance. If the derivations are rigorous, the work supplies a continuous, parametric setting for open-source game theory that bridges discrete program equilibria with gradient-based learning. The exact threshold and the sensitivity-ratio condition in the neural extension furnish concrete, testable predictions about when introspective coupling produces cooperation. Equilibrium existence and the boundary test provide foundational results that could support further analysis of multi-agent systems with access to internal parameterizations.

major comments (1)
  1. [Gradient-ascent analysis / coupling threshold derivation] The derivation of the exact coupling threshold (abstract and gradient-ascent section) rests on the assumption that semantics maps exist, are sufficiently smooth, and permit well-defined gradient-ascent dynamics without further regularity conditions. This assumption is load-bearing for the claimed exactness of the threshold; the manuscript does not state what occurs under weaker continuity or differentiability requirements that might shift or eliminate the switch point in symmetric 2×2 games.
minor comments (2)
  1. Notation for the semantics maps and the parameter-to-action conversion could be introduced with an explicit diagram or small example early in the text to aid readability.
  2. The one-dimensional boundary test for parametric program Nash equilibria would benefit from a short pseudocode or algorithmic statement to clarify its computational use.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the careful reading and for identifying the role of regularity assumptions in the gradient-ascent analysis. We respond to the major comment below.

read point-by-point responses

  1. Referee: [Gradient-ascent analysis / coupling threshold derivation] The derivation of the exact coupling threshold (abstract and gradient-ascent section) rests on the assumption that semantics maps exist, are sufficiently smooth, and permit well-defined gradient-ascent dynamics without further regularity conditions. This assumption is load-bearing for the claimed exactness of the threshold; the manuscript does not state what occurs under weaker continuity or differentiability requirements that might shift or eliminate the switch point in symmetric 2×2 games.

    Authors: We agree that the exact coupling threshold is derived under the assumption that semantics maps are sufficiently smooth to support well-defined gradient-ascent dynamics. Within the parametric open-source framework, semantics maps are constructed as differentiable mappings from parameter vectors to mixed strategies, which is the setting in which the threshold is obtained. Under weaker conditions such as mere continuity, the gradient dynamics are not defined in the same manner and the switch point may not exist or may differ; the result is scoped to the differentiable case. We will revise the gradient-ascent section to state these regularity requirements explicitly and to note their necessity for the claimed exactness. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The abstract and provided context describe a new framework for parametric open-source games, equilibrium existence results, and derivation of a coupling threshold for gradient ascent in 2x2 games. No equations, fitted parameters, self-citations, or ansatzes are exhibited that would reduce any claimed prediction or result to its inputs by construction. The derivation chain is presented as building on standard game theory and prior open-source models without visible self-referential reductions, making the paper self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Only the abstract is available, so the ledger is populated from claims visible there; the existence of suitable semantics maps and the well-posedness of gradient dynamics are implicit background assumptions.

axioms (1)
  • domain assumption Semantics maps exist that convert any parameter profile into a valid mixed-action profile for the underlying finite game.
    Required for the definition of parametric open-source games and for the gradient-ascent analysis.

pith-pipeline@v0.9.1-grok · 5673 in / 1274 out tokens · 17944 ms · 2026-06-26T01:57:59.147820+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

54 extracted references · 35 canonical work pages · 3 internal anchors

  1. [1]

    Cooperative and uncooperative institution designs:

    Critch, Andrew and Dennis, Michael and Russell, Stuart , month = aug, year =. Cooperative and uncooperative institution designs:. doi:10.48550/arXiv.2208.07006 , abstract =

    work page doi:10.48550/arxiv.2208.07006

  2. [2]

    Games and Economic Behavior , author =

    Program equilibrium , volume =. Games and Economic Behavior , author =. 2004 , pages =. doi:10.1016/j.geb.2004.02.002 , abstract =

    work page doi:10.1016/j.geb.2004.02.002 2004

  3. [3]

    Program equilibria and discounted computation time , isbn =

    Fortnow, Lance , month = jul, year =. Program equilibria and discounted computation time , isbn =. Proceedings of the 12th. doi:10.1145/1562814.1562833 , abstract =

    work page doi:10.1145/1562814.1562833

  4. [4]

    Theory and Decision , author =

    Robust program equilibrium , volume =. Theory and Decision , author =. 2019 , keywords =. doi:10.1007/s11238-018-9679-3 , abstract =

    work page doi:10.1007/s11238-018-9679-3 2019

  5. [5]

    Similarity-based cooperative equilibrium , url =

    Oesterheld, Caspar and Treutlein, Johannes and Grosse, Roger and Conitzer, Vincent and Foerster, Jakob , month = nov, year =. Similarity-based cooperative equilibrium , url =. doi:10.48550/arXiv.2211.14468 , abstract =

    work page doi:10.48550/arxiv.2211.14468

  6. [6]

    Program equilibrium in the prisoner’s dilemma via L

    LaVictoire, Patrick and Fallenstein, Benja and Yudkowsky, Eliezer and Barasz, Mihaly and Christiano, Paul and Herreshoff, Marcello , booktitle=. Program equilibrium in the prisoner’s dilemma via L

  7. [7]

    Barasz, Mihaly and Christiano, Paul and Fallenstein, Benja and Herreshoff, Marcello and LaVictoire, Patrick and Yudkowsky, Eliezer , month = apr, year =. Robust. doi:10.48550/arXiv.1401.5577 , abstract =

    work page doi:10.48550/arxiv.1401.5577

  8. [8]

    The Journal of Symbolic Logic , author =

    A. The Journal of Symbolic Logic , author =. 2019 , keywords =. doi:10.1017/jsl.2017.42 , abstract =

    work page doi:10.1017/jsl.2017.42 2019

  9. [9]

    2020 , eprint=

    Embedded Agency , author=. 2020 , eprint=

    2020

  10. [10]

    2025 , eprint=

    Contemplative Artificial Intelligence , author=. 2025 , eprint=

    2025

  11. [11]

    2024 , eprint=

    Towards Safe and Honest AI Agents with Neural Self-Other Overlap , author=. 2024 , eprint=

    2024

  12. [12]

    2024 , eprint=

    Mechanistic Interpretability for AI Safety -- A Review , author=. 2024 , eprint=

    2024

  13. [13]

    2025 , eprint=

    Frontier Models are Capable of In-context Scheming , author=. 2025 , eprint=

    2025

  14. [14]

    2026 , eprint=

    A Survey of Zero-Knowledge Proof Based Verifiable Machine Learning , author=. 2026 , eprint=

    2026

  15. [15]

    Glicksberg, I. L. , year =. A. Proceedings of the American Mathematical Society , publisher =. doi:10.2307/2032478 , number =

    work page doi:10.2307/2032478

  16. [16]

    Topological

    Berge, Claude , year =. Topological

  17. [17]

    Games and Economic Behavior , author =

    A commitment folk theorem , volume =. Games and Economic Behavior , author =. 2010 , pages =. doi:10.1016/j.geb.2009.09.008 , abstract =

    work page doi:10.1016/j.geb.2009.09.008 2010

  18. [18]

    Translucent Players: Explaining Cooperative Behavior in Social Dilemmas

    Capraro, Valerio and Halpern, Joseph Y. , month = nov, year =. Translucent. doi:10.48550/arXiv.1410.3363 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1410.3363

  19. [19]

    International Journal of Game Theory , author =

    Game theory with translucent players , volume =. International Journal of Game Theory , author =. 2018 , pages =. doi:10.1007/s00182-018-0626-x , abstract =

    work page doi:10.1007/s00182-018-0626-x 2018

  20. [20]

    Rosen, J. B. , year =. Existence and. Econometrica , publisher =. doi:10.2307/1911749 , abstract =

    work page doi:10.2307/1911749

  21. [21]

    Balduzzi, David and Racaniere, Sebastien and Martens, James and Foerster, Jakob and Tuyls, Karl and Graepel, Thore , month = jun, year =. The. doi:10.48550/arXiv.1802.05642 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1802.05642

  22. [22]

    2021 , eprint=

    Stable Opponent Shaping in Differentiable Games , author=. 2021 , eprint=

    2021

  23. [23]

    Learning with Opponent-Learning Awareness

    Foerster, Jakob N. and Chen, Richard Y. and Al-Shedivat, Maruan and Whiteson, Shimon and Abbeel, Pieter and Mordatch, Igor , month = sep, year =. Learning with. doi:10.48550/arXiv.1709.04326 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1709.04326

  24. [24]

    Games and Economic Behavior , author =

    On. Games and Economic Behavior , author =. 1995 , pages =. doi:10.1006/game.1995.1031 , abstract =

    work page doi:10.1006/game.1995.1031 1995

  25. [25]

    and Ho, Teck-Hua and Chong, Juin Kuan , editor =

    Camerer, Colin F. and Ho, Teck-Hua and Chong, Juin Kuan , editor =. Behavioural. Advances in. 2004 , keywords =. doi:10.1057/9780230523371_8 , abstract =

    work page doi:10.1057/9780230523371_8 2004

  26. [26]

    PLoS computational biology , author =

    Game theory of mind , volume =. PLoS computational biology , author =. 2008 , keywords =. doi:10.1371/journal.pcbi.1000254 , abstract =

    work page doi:10.1371/journal.pcbi.1000254 2008

  27. [27]

    Biologically Inspired Cognitive Architectures , author =

    Higher-order theory of mind in the. Biologically Inspired Cognitive Architectures , author =. 2015 , keywords =. doi:10.1016/j.bica.2014.11.010 , abstract =

    work page doi:10.1016/j.bica.2014.11.010 2015

  28. [28]

    Current Opinion in Behavioral Sciences , author =

    A psychological approach to strategic thinking in games , volume =. Current Opinion in Behavioral Sciences , author =. 2015 , pages =. doi:10.1016/j.cobeha.2015.04.005 , abstract =

    work page doi:10.1016/j.cobeha.2015.04.005 2015

  29. [29]

    and Doshi, Prashant and Young, Diana L

    Goodie, Adam S. and Doshi, Prashant and Young, Diana L. , year =. Levels of theory‐of‐mind reasoning in competitive games , volume =. Journal of Behavioral Decision Making , publisher =. doi:10.1002/bdm.717 , abstract =

    work page doi:10.1002/bdm.717

  30. [30]

    Journal of Artificial Intelligence Research , author =

    A. Journal of Artificial Intelligence Research , author =. 2005 , note =. doi:10.1613/jair.1579 , abstract =

    work page doi:10.1613/jair.1579 2005

  31. [31]

    Theory of

    Oguntola, Ini and Campbell, Joseph and Stepputtis, Simon and Sycara, Katia , month = jul, year =. Theory of. doi:10.48550/arXiv.2307.01158 , abstract =

    work page doi:10.48550/arxiv.2307.01158

  32. [32]

    Bonanno, Giacomo , year =. Game. doi:10.13140/RG.2.1.3369.7360 , abstract =

    work page doi:10.13140/rg.2.1.3369.7360

  33. [33]

    , month = sep, year =

    Nisan, Noam and Roughgarden, Tim and Tardos, Eva and Vazirani, Vijay V. , month = sep, year =. Algorithmic

  34. [34]

    Essentials of

    Leyton-Brown, Kevin and Shoham, Yoav , month = jul, year =. Essentials of

  35. [35]

    Behavioral and Brain Sciences , author =

    Does the chimpanzee have a theory of mind? , volume =. Behavioral and Brain Sciences , author =. 1978 , keywords =. doi:10.1017/S0140525X00076512 , abstract =

    work page doi:10.1017/s0140525x00076512 1978

  36. [36]

    theory of mind

    Does the autistic child have a “theory of mind” ? , volume =. Cognition , author =. 1985 , pages =. doi:10.1016/0010-0277(85)90022-8 , abstract =

    work page doi:10.1016/0010-0277(85)90022-8 1985

  37. [37]

    Cognitive Psychology , author =

    What is theory of mind?. Cognitive Psychology , author =. 2022 , keywords =. doi:10.1016/j.cogpsych.2022.101495 , abstract =

    work page doi:10.1016/j.cogpsych.2022.101495 2022

  38. [38]

    Foundations and Trends® in Machine Learning , author =

    An. Foundations and Trends® in Machine Learning , author =. 2018 , pages =. doi:10.1561/2200000071 , abstract =

    work page doi:10.1561/2200000071 2018

  39. [39]

    Human-level control through deep reinforcement learning

    Mnih, Volodymyr and Kavukcuoglu, Koray and Silver, David and Rusu, Andrei A. and Veness, Joel and Bellemare, Marc G. and Graves, Alex and Riedmiller, Martin and Fidjeland, Andreas K. and Ostrovski, Georg and Petersen, Stig and Beattie, Charles and Sadik, Amir and Antonoglou, Ioannis and King, Helen and Kumaran, Dharshan and Wierstra, Daan and Legg, Shane ...

    work page doi:10.1038/nature14236

  40. [40]

    doi:10.48550/arXiv.2308.03526 , abstract =

    Mathieu, Michaël and Ozair, Sherjil and Srinivasan, Srivatsan and Gulcehre, Caglar and Zhang, Shangtong and Jiang, Ray and Paine, Tom Le and Powell, Richard and Żołna, Konrad and Schrittwieser, Julian and Choi, David and Georgiev, Petko and Toyama, Daniel and Huang, Aja and Ring, Roman and Babuschkin, Igor and Ewalds, Timo and Bordbar, Mahyar and Henderso...

    work page doi:10.48550/arxiv.2308.03526

  41. [41]

    Attention is

    Vaswani, Ashish and Shazeer, Noam and Parmar, Niki and Uszkoreit, Jakob and Jones, Llion and Gomez, Aidan N and Kaiser, Ł ukasz and Polosukhin, Illia , year =. Attention is. Advances in

  42. [42]

    Mastering the game of Go without human knowledge.Nature, 550(7676):354–359, October 2017

    Silver, David and Schrittwieser, Julian and Simonyan, Karen and Antonoglou, Ioannis and Huang, Aja and Guez, Arthur and Hubert, Thomas and Baker, Lucas and Lai, Matthew and Bolton, Adrian and Chen, Yutian and Lillicrap, Timothy and Hui, Fan and Sifre, Laurent and van den Driessche, George and Graepel, Thore and Hassabis, Demis , month = oct, year =. Maste...

    work page doi:10.1038/nature24270

  43. [43]

    Convergence

    Chasnov, Benjamin and Ratliff, Lillian and Mazumdar, Eric and Burden, Samuel , month = aug, year =. Convergence. Proceedings of

  44. [44]

    SIAM Journal on Mathematics of Data Science , author =

    On. SIAM Journal on Mathematics of Data Science , author =. 2020 , note =. doi:10.1137/18M1231298 , abstract =

    work page doi:10.1137/18m1231298 2020

  45. [45]

    Lin, Tianyi and Zhou, Zhengyuan and Mertikopoulos, Panayotis and Jordan, Michael , month = nov, year =. Finite-. Proceedings of the 37th

  46. [46]

    , year =

    Fudenberg, Drew and Levine, David K. , year =. The

  47. [47]

    and Rubinstein, Ariel , month = jul, year =

    Osborne, Martin J. and Rubinstein, Ariel , month = jul, year =. A

  48. [48]

    Basar, Tamer and Olsder, Geert Jan , year =. Dynamic

  49. [49]

    and Mahony, R

    Absil, P.-A. and Mahony, R. and Sepulchre, R. , editor =. Optimization. Recent. 2010 , keywords =. doi:10.1007/978-3-642-12598-0_12 , abstract =

    work page doi:10.1007/978-3-642-12598-0_12 2010

  50. [50]

    Facchinei, Francisco and Pang, Jong-Shi , month = jun, year =. Finite-

  51. [51]

    Bonnans and A

    Bonnans, J. Frédéric and Shapiro, Alexander , year =. Perturbation. doi:10.1007/978-1-4612-1394-9 , urldate =

    work page doi:10.1007/978-1-4612-1394-9

  52. [52]

    Plaat, Aske and Kosters, Walter and Preuss, Mike , month = dec, year =. Deep. doi:10.48550/arXiv.2008.05598 , abstract =

    work page doi:10.48550/arxiv.2008.05598 2008

  53. [53]

    arXiv.org , author =

    Opponent. arXiv.org , author =

  54. [54]

    arXiv.org , author =

    Theory of. arXiv.org , author =. 2023 , doi =. doi:10.18653/v1/2023.emnlp-main.13 , abstract =

    work page doi:10.18653/v1/2023.emnlp-main.13 2023