Pith. sign in

REVIEW 3 major objections 5 minor 50 references

Position: Emergent Machina Sapiens Urge Rethinking Multi-Agent Paradigms

T0 review · 3 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

Pith's one-line read The paper argues that classical multi-agent frameworks—multi-agent reinforcement learning and game theory—are too rigid for the coming world of independently deployed AI agents, and that agents must instead be empowered to revise their…

desk verdict A sincere but under-specified position paper: the case for dynamic norms in open multi-agent systems is timely, but the anchoring mechanism it names is never constructed, leaving the central feasibility claim open. read the letter →

arxiv 2502.04388 v3 pith:RGOFOLNQ submitted 2025-02-05 cs.MA cs.AI

classification cs.MAcs.AI
keywords multi-agentsystemsagenticAIdynamicnormsemergentbehaviorself-organizationcriticalinfrastructuresautonomousdrivingenergyautonomy
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 argues that classical multi-agent frameworks—multi-agent reinforcement learning and game theory—are too rigid for the coming world of independently deployed AI agents, and that agents must instead be empowered to revise their own objectives, form coalitions, and shift between cooperation and competition through evolving relationships and social feedback. Its central position is that interoperability in open AI ecosystems should be an emergent, self-organizing, context-aware outcome rather than a pre-engineered consensus. The authors build a conceptual framework around agents, worlds, and dynamic norms, and ground it in two critical-infrastructure case studies: autonomous driving and energy autonomy. A sympathetic reader would care because, if the position is right, the design target for multi-agent AI shifts from fixed-rule coordination to socially embedded adaptation.

What carries the argument

The central mechanism is the dynamic-norm loop: a norm tuple $M := \langle N, I, E, P \rangle$ pairs a relationship network $N$, agent impacts $I$, and peer expectations $E$ with learnable protocols $P$ that modify agent objectives (e.g., $J_i^P = J_i - P_i$) or constrain actions. Protocols evolve by a stability equation $dP/dt = -\gamma (P - P^*)$, relationships update by $w_{ij} \leftarrow f(w_{ij}, I_{ij}, E_{ij})$, coalitions form when the summed relationship strength within a group crosses a threshold $\theta$, and cross-environment transitions allow actions in one environment to propagate to another. This machinery carries the argument by showing how norms could emerge from repeated interaction and social feedback rather than being imposed in advance.

What would settle it

A concrete falsifying observation would be a large-scale simulation of independently trained agents in a shared transportation or energy market, with no explicit coordination protocol, in which allowing agents to revise their objectives dynamically produces persistent chaos, deadlock, or collusion that static-objective agents avoid; the position is contradicted if dynamic norms consistently fail to stabilize the ecosystem.

Watch

Extended reading notes

Core claim

The paper's core claim is that the rise of machina sapiens—autonomous, continuously learning AI systems deployed by uncoordinated stakeholders into shared environments—requires a fundamental rethinking of multi-agent paradigms. Existing MARL and game-theoretic models treat agents as optimizing fixed utility functions toward static equilibria under predefined interaction rules; the paper contends these assumptions break under real-world unpredictability, where goals shift, relationships form and dissolve, and agents must balance individual objectives with collective welfare. The proposed alternative is a framework in which each agent is defined by goal, policy, algorithm, experience, and relationships; the world is a set of possibly interlinked environments; and norms—networks, impacts, expectations, and protocols—evolve dynamically through a feedback loop, so that protocols can act as penalties or constraints that agents learn to follow. The discovery, in the authors' terms, is that interoperability among independent AI agents should be treated as an emergent, self-organizing, and context-aware property rather than an engineered consensus or a pre-designed equilibrium.

Load-bearing premise

The framework assumes that a feasible anchoring mechanism exists to let agents revise their own objectives and norms without falling into the self-reference dilemma and without generating harmful tacit collusion; the paper identifies both problems but does not provide that mechanism.

Editorial extensions

If this is right

  • Agents deployed by different vendors must be able to adjust their own reward structures and goals after deployment, not only their policies within a fixed objective.
  • Norms and protocols become learnable artifacts that evolve through trust, impact, and expectation, making coordination a bottom-up, ongoing process rather than a one-time design.
  • Coalitions among agents become a legitimate emergent phenomenon, formed when accumulated relationship strength crosses a threshold, and they must be monitored so they do not tip into tacit collusion.
  • Evaluation of such systems requires new metrics beyond individual reward, including fairness of sacrifice, stability, and adaptability of norms, because no static equilibrium defines success.
  • In critical infrastructure, safety and ethical constraints must be embedded as protocols that can bind agents' evolving goals without freezing them.

Reading between the lines

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

  • If the position is right, benchmark suites for multi-agent reinforcement learning should be extended with open-ended scenarios in which agent objectives are mutable and no single reward function is fixed.
  • The self-reference dilemma suggests a concrete research program: construct bounded meta-objective functions—for example, goals expressed as preferences over sets of norms—that let agents revise lower-level goals while keeping a stable higher-level anchor.
  • The tacit-collusion concern points toward practical monitoring tools that detect when relationship networks among independently deployed agents become too dense or too aligned, as an early warning of coordinated price gouging or safety degradation.
  • The framework implicitly predicts that heterogeneous, self-interested agents reach locally adaptive conventions faster when they can form trust-weighted communication networks than when they are forced to use a shared static protocol.
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 argues that conventional multi-agent frameworks—MARL and game theory in particular—are built on static objective structures, predefined interaction rules, and pre-engineered coordination, and are therefore inadequate for open ecosystems of independently deployed AI agents in critical infrastructure. It proposes a conceptual framework in which agents can dynamically adjust goals, form and dissolve relationships, adopt evolving protocols, and co-evolve social norms. The formal content is limited to illustrative equations (protocol penalties, protocol dynamics, relationship updates, coalition thresholds, cross-environment transition). The paper supports its position with two case studies (autonomous driving and distributed energy management) and a catalog of challenges, including emergent behavior, dynamic objectives, the self-reference dilemma, ethical free-will questions, and human-agent collaboration. It concludes by calling for a research shift toward self-organizing multi-agentic ecosystems rather than presenting a validated solution.

Significance. The paper addresses a genuine and timely problem: independently designed and deployed AI agents will increasingly share safety-critical environments, and existing coordination mechanisms assume either centralized design or fixed preferences. Its value as a position paper lies in identifying this gap, framing a research agenda around dynamic norms, and grounding the discussion in concrete infrastructure scenarios. The paper is honest about several central open problems, especially the self-reference dilemma in Section V-C. However, because the central feasibility claim—that dynamic objective revision and social feedback can yield safe, harmonious coexistence—depends on an anchoring mechanism that is not supplied, the paper currently reads as a call for research rather than a defensible technical position. There are no machine-checked proofs, parameter-free derivations, or falsifiable predictions to shift the burden of evidence.

major comments (3)
  1. [V-C] The self-reference dilemma is identified but not resolved, and it is load-bearing for the central claim. If an agent can revise its own goal, it needs a criterion to judge whether a revision is an improvement or a deviation; the paper calls this an 'anchoring framework' but gives no construction or even a candidate principle (e.g., a fixed human-values layer, a meta-norm, or a slower-timescale normative update). Equation (2) makes the problem concrete: if P* is fixed, the framework reintroduces the static objective structure it criticizes, and if P* is itself revisable, the same infinite regress recurs. Without at least a concrete research proposal for the anchor, the manuscript's central assertion that dynamic objective adjustment can produce harmonious coexistence is not supported.
  2. [III-B] Equation (2) is presented as the stability model for protocol evolution, but the text immediately concedes that protocols 'may not necessarily converge.' The non-convergent regime is precisely where dynamic norms could produce oscillation or chaos in safety-critical infrastructure, and no analysis is provided. No conditions on gamma or P* are given, and no alternative model replaces Eq. (2) when convergence fails. As a result, the evaluation metrics in Section III-C (norm stability, behavioral predictability) have no formal grounding, and the claimed balance between stability and adaptability remains an assertion rather than a framework property.
  3. [III-B and IV-B] The coalition-formation rule (sum of relationship weights exceeding a threshold theta) and the trust-update rule in Eq. (3) create a natural mechanism for repeatedly interacting agents to converge on coordinated behavior without explicit communication. Section IV-B correctly identifies tacit collusion among energy management systems as a key risk, but the framework offers no detection, constraint, or incentive mechanism to prevent coalition formation from crossing into collusion. Since the paper promises 'safe competition or cooperation' in critical infrastructure, the absence of any safeguard is not a peripheral implementation detail; it is a missing component at the core of the proposed paradigm.
minor comments (5)
  1. [III-A] The agent tuple is written A := ⟨J, π, O, B, G⟩, but the definition list names L (Algorithm) and G (Relationships) and never defines O. Please align the tuple with the definition list.
  2. [III-B] Equation (1) overloads the symbol P: P_i is introduced as a protocol, then as a penalty term α·H_i, and J_i^P uses a superscript P. This makes the example harder to follow; please use distinct notation for protocols, penalties, and goal modifiers.
  3. [IV-B] The phrase 'price gauging' should be 'price gouging'; it appears twice in the energy autonomy case study.
  4. [II-B] The sentence beginning 'These approaches use context modeling' is very long and combines perception, reasoning, learning, and adaptation; splitting it would improve readability.
  5. [Figure 2] Panel (b) is described as a chain accident at a roundabout caused by a sudden stop, but the text also mentions hesitation at a roundabout; please clarify in the caption whether the accident occurs at the roundabout entry, inside the roundabout, or after a vehicle stops.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the paper is a position piece with illustrative equations, and its self-reference limitation is openly acknowledged rather than smuggled in.

full rationale

This paper is a position/vision paper, not a derivation chain. Its equations are illustrative definitions: Eq. (1) defines a goal penalty, Eq. (2) is a generic dynamical-system sketch for protocol evolution with the explicit caveat that 'protocols may not necessarily converge,' Eq. (3) is an unspecified update rule for relationship weights, and Eq. (4) is a cross-environment transition notation. No parameter is fitted to data and then renamed a prediction; there is no uniqueness theorem, and no load-bearing claim rests on a self-citation chain (the references are background literature, not the argument's support). The one in-scope limitation is Section V-C, 'The Dilemma of Self-Reference,' which openly identifies the chicken-and-egg problem of goal revision and calls for an 'anchoring framework' without claiming to provide one; this is an acknowledged open problem, not a hidden circular premise. The central claim is a research agenda advocating dynamic norms and self-organization, and the framework is underdetermined rather than circular. No specific reduction of an output to an input can be exhibited from the text, so no circularity is found.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

Symbolic coefficients alpha, gamma, theta in Eqs. (1)-(2) and the coalition threshold are placeholders with no assigned values and no role in any derivation or prediction. The paper introduces no new physical, mathematical, or empirical entities; 'dynamic norms' and 'Machina Sapiens' are framing concepts, not postulated scientific objects.

assumptions (4)
  • domain assumption Independent stakeholders will deploy AI agents with unaligned objectives in shared critical infrastructure environments.
    The entire paper starts from this premise in the Introduction and the two case studies assume it. It is plausible but not itself established.
  • ad hoc to paper Agents are able to revise their own goals and protocols in response to feedback in a way that is stable and controllable.
    Central to the proposed dynamic-norms framework. No mechanism is supplied, and Section V-C's self-reference dilemma is an unresolved challenge to this assumption.
  • ad hoc to paper Protocol dynamics follow the linear form dP/dt = -gamma(P-P*) with an equilibrium P* that can be defined as argmax of collective utility.
    Introduced in Section III-B, Eq. (2), without derivation. P* depends on agent goals, which are themselves evolving, so the equilibrium is not well-defined.
  • ad hoc to paper Relationship update function f(.) in Eq. (3) leads to trust-building and beneficial coalition formation rather than exploitation or collusion.
    The update rule is left unspecified; the paper assumes it can foster coordination, but Section IV-B gives a collusion counterexample.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Position: Emergent Machina Sapiens Urge Rethinking Multi-Agent Paradigms." pith.science (2026). https://pith.science/paper/RGOFOLNQ

@misc{pith2026250204388,
  author       = {Pith},
  title        = {Pith review of: Position: Emergent Machina Sapiens Urge Rethinking Multi-Agent Paradigms},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RGOFOLNQ}},
  note         = {Machine review of arXiv:2502.04388}
}
read the original abstract

Artificial Intelligence (AI) agents capable of autonomous learning and independent decision-making hold great promise for addressing complex challenges across various critical infrastructure domains, including transportation, energy systems, and manufacturing. However, the surge in the design and deployment of AI systems, driven by various stakeholders with distinct and unaligned objectives, introduces a crucial challenge: How can uncoordinated AI systems coexist and evolve harmoniously in shared environments without creating chaos or compromising safety? To address this, we advocate for a fundamental rethinking of existing multi-agent frameworks, such as multi-agent systems and game theory, which are largely limited to predefined rules and static objective structures. We posit that AI agents should be empowered to adjust their objectives dynamically, make compromises, form coalitions, and safely compete or cooperate through evolving relationships and social feedback. Through two case studies in critical infrastructure applications, we call for a shift toward the emergent, self-organizing, and context-aware nature of these multi-agentic AI systems.

Figures

Figures reproduced from arXiv: 2502.04388 by the authors.

Figure 1
Figure 1. The framework enables agents to refine their goals, policies, and [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Challenges of uncoordinated AI agents in autonomous driving: (a) [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Agentic EMS in energy autonomy, mapped on the hierarchical [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

50 extracted references · 38 canonical work pages

  1. [1]

    Human-level control through deep reinforcement learning,

    V . Mnihet al., “Human-level control through deep reinforcement learning,” Nature, vol. 518, no. 7540, pp. 529–533, 2015

  2. [2]

    Mastering the game of Go with deep neural networks and tree search,

    D. Silver et al., “Mastering the game of Go with deep neural networks and tree search,” Nature, vol. 529, no. 7587, pp. 484–489, Jan 2016

  3. [3]

    Practices for governing agentic ai systems,

    Y . Shavitet al., “Practices for governing agentic ai systems,” Research Paper, OpenAI, December, 2023

  4. [4]

    Harms from increasingly agentic algorithmic systems,

    A. Chan et al., “Harms from increasingly agentic algorithmic systems,” in Proceedings of the 2023 ACM Conference on Fairness, Accountability, and Transparency, ser. FAccT ’23. New York, NY , USA: Association for Computing Machinery, 2023, p. 651–666

  5. [5]

    Reflections from the turing award winners,

    Y . LeCun and Y . Bengio, “Reflections from the turing award winners,”

  6. [6]

    Big self-supervised models are strong semi-supervised learners,

    T. Chen, S. Kornblith, K. Swersky, M. Norouzi, and G. E. Hinton, “Big self-supervised models are strong semi-supervised learners,” Advances in neural information processing systems , vol. 33, pp. 22 243–22 255, 2020

  7. [7]

    Model-agnostic meta-learning for fast adaptation of deep networks,

    C. Finn, P. Abbeel, and S. Levine, “Model-agnostic meta-learning for fast adaptation of deep networks,” in Proceedings of the 34th International Conference on Machine Learning , ser. Proceedings of Machine Learning Research, D. Precup and Y . W. Teh, Eds., vol. 70. PMLR, 06–11 Aug 2017, pp. 1126–1135

  8. [8]

    Meta-learning with implicit gradients,

    A. Rajeswaran, C. Finn, S. M. Kakade, and S. Levine, “Meta-learning with implicit gradients,” in Advances in Neural Information Processing Systems, H. Wallach, H. Larochelle, A. Beygelzimer, F. d 'Alch´e-Buc, E. Fox, and R. Garnett, Eds., vol. 32. Curran Associates, Inc., 2019

Show all 50 references
  1. [9]

    Cellular computational net- works—a scalable architecture for learning the dynamics of large networked systems,

    B. Luitel and G. K. Venayagamoorthy, “Cellular computational net- works—a scalable architecture for learning the dynamics of large networked systems,” Neural Networks, vol. 50, pp. 120–123, 2014

  2. [10]

    Reinforcement learning benchmarks for traf- fic signal control,

    J. Ault and G. Sharon, “Reinforcement learning benchmarks for traf- fic signal control,” in Thirty-fifth Conference on Neural Information Processing Systems Datasets and Benchmarks Track (Round 1) , 2021

  3. [11]

    Model-free real-time ev charging scheduling based on deep reinforcement learning,

    Z. Wan, H. Li, H. He, and D. Prokhorov, “Model-free real-time ev charging scheduling based on deep reinforcement learning,” IEEE Transactions on Smart Grid , vol. 10, no. 5, pp. 5246–5257, 2019

  4. [12]

    Reinforcement learning for selective key applications in power systems: Recent advances and future challenges,

    X. Chen, G. Qu, Y . Tang, S. Low, and N. Li, “Reinforcement learning for selective key applications in power systems: Recent advances and future challenges,” IEEE Transactions on Smart Grid , vol. 13, no. 4, pp. 2935–2958, 2022

  5. [13]

    Scalable deep reinforcement learning for vision- based robotic manipulation,

    D. Kalashnikov et al., “Scalable deep reinforcement learning for vision- based robotic manipulation,” in Proceedings of The 2nd Conference on Robot Learning , ser. Proceedings of Machine Learning Research, A. Billard, A. Dragan, J. Peters, and J. Morimoto, Eds., vol. 87. PMLR,...

  6. [14]

    Deep reinforcement learning for robotics: A survey of real- world successes,

    C. Tang, B. Abbatematteo, J. Hu, R. Chandra, R. Mart ´ın-Mart´ın, and P. Stone, “Deep reinforcement learning for robotics: A survey of real- world successes,” 2024

  7. [15]

    Deep reinforcement learning-based dynamic scheduling in smart manufacturing,

    L. Zhou, L. Zhang, and B. K. Horn, “Deep reinforcement learning-based dynamic scheduling in smart manufacturing,” Procedia Cirp, vol. 93, pp. 383–388, 2020

  8. [16]

    Dynamic job shop scheduling based on deep reinforcement learning for multi-agent manu- facturing systems,

    Y . Zhang, H. Zhu, D. Tang, T. Zhou, and Y . Gui, “Dynamic job shop scheduling based on deep reinforcement learning for multi-agent manu- facturing systems,” Robotics and Computer-Integrated Manufacturing , vol. 78, p. 102412, 2022

  9. [17]

    R. S. Sutton and A. G. Barto, Reinforcement Learning: An Introduction . Cambridge, MA, USA: A Bradford Book, 2018

  10. [18]

    Tan, Multi-Agent Reinforcement Learning: Independent vs

    M. Tan, Multi-Agent Reinforcement Learning: Independent vs. Coopera- tive Agents. San Francisco, CA, USA: Morgan Kaufmann Publishers Inc., 1997, p. 487–494

  11. [19]

    The dynamics of reinforcement learning in cooperative multiagent systems,

    C. Claus and C. Boutilier, “The dynamics of reinforcement learning in cooperative multiagent systems,” in Proceedings of the Fifteenth Na- tional/Tenth Conference on Artificial Intelligence/Innovative Applications of Artificial Intelligence , ser. AAAI ’98/IAAI ’98. USA: Ameri...

  12. [20]

    QD-learning: A collaborative distributed strategy for multi-agent reinforcement learning through consensus + innovations,

    S. Kar, J. M. F. Moura, and H. V . Poor, “ QD-learning: A collaborative distributed strategy for multi-agent reinforcement learning through consensus + innovations,” IEEE Transactions on Signal Processing , vol. 61, no. 7, pp. 1848–1862, 2013

  13. [21]

    Multi- agent actor-critic for mixed cooperative-competitive environments,

    R. Lowe, Y . Wu, A. Tamar, J. Harb, P. Abbeel, and I. Mordatch, “Multi- agent actor-critic for mixed cooperative-competitive environments,” in the 31st Conference on Neural Information Processing Systems , ser. NeurIPS’20, 2017, pp. 6382–6393

  14. [22]

    Multi-agent reinforcement learning: A selective overview of theories and algorithms,

    K. Zhang, Z. Yang, and T. Ba s ¸ar, “Multi-agent reinforcement learning: A selective overview of theories and algorithms,” ArXiv pre-prints, 2019, arXiv:1911.10635

  15. [23]

    Markov games as a framework for multi-agent rein- forcement learning,

    M. L. Littman, “Markov games as a framework for multi-agent rein- forcement learning,” in Machine learning proceedings 1994 . Elsevier, 1994, pp. 157–163

  16. [24]

    A game-theoretic framework for managing risk in multi- agent systems,

    O. Slumbers, D. H. Mguni, S. B. Blumberg, S. M. Mcaleer, Y . Yang, and J. Wang, “A game-theoretic framework for managing risk in multi- agent systems,” in Proceedings of the 40th International Conference on Machine Learning , ser. Proceedings of Machine Learning Research, A. K...

  17. [25]

    Alympics: Llm agents meet game theory–exploring strate- gic decision-making with ai agents,

    S. Mao et al., “Alympics: Llm agents meet game theory–exploring strate- gic decision-making with ai agents,” arXiv preprint arXiv:2311.03220 , 2024

  18. [26]

    Exploring equilibrium strategies in network games with generative ai,

    Y . Yang, H. Du, G. Sun, Z. Xiong, D. Niyato, and Z. Han, “Exploring equilibrium strategies in network games with generative ai,” arXiv preprint arXiv:2405.08289, 2024

  19. [27]

    Multi-agent systems: Technical & ethical challenges of functioning in a mixed group,

    K. Gal and B. J. Grosz, “Multi-agent systems: Technical & ethical challenges of functioning in a mixed group,” Daedalus, vol. 151, no. 2, pp. 114–126, 2022

  20. [28]

    Intelligent agents: Theory and practice,

    M. Wooldridge and N. R. Jennings, “Intelligent agents: Theory and practice,” The knowledge engineering review, vol. 10, no. 2, pp. 115–152, 1995

  21. [29]

    Agent ai: Surveying the horizons of multimodal interaction,

    Z. Durante et al. , “Agent ai: Surveying the horizons of multimodal interaction,” 2024. [Online]. Available: https://arxiv.org/abs/2401.03568

  22. [30]

    Aflow: Automating agentic workflow generation,

    J. Zhang et al., “Aflow: Automating agentic workflow generation,” arXiv preprint arXiv:2410.10762, 2024

  23. [31]

    Quantum multi-agent meta reinforce- ment learning,

    W. J. Yun, J. Park, and J. Kim, “Quantum multi-agent meta reinforce- ment learning,” in Proceedings of the AAAI Conference on Artificial Intelligence, vol. 37, no. 9, 2023, pp. 11 087–11 095

  24. [32]

    Multi-agent based architecture of context-aware systems,

    C.-D. WANG and X.-F. WANG, “Multi-agent based architecture of context-aware systems,” in 2007 International Conference on Multimedia and Ubiquitous Engineering (MUE’07) , 2007, pp. 615–619

  25. [33]

    A survey on context-aware multi-agent systems: Techniques, challenges and future directions,

    H. Du, S. Thudumu, R. Vasa, and K. Mouzakis, “A survey on context-aware multi-agent systems: Techniques, challenges and future directions,” 2024. [Online]. Available: https://arxiv.org/abs/2402.01968

  26. [34]

    A multi-agent system for context- aware electric vehicle fleet routing: A step towards more sustainable urban operations,

    G. Jelen, J. Babic, and V . Podobnik, “A multi-agent system for context- aware electric vehicle fleet routing: A step towards more sustainable urban operations,” Journal of cleaner production , vol. 374, p. 134047, 2022

  27. [35]

    Consensus in multi-agent systems: a review,

    A. Amirkhani and A. H. Barshooi, “Consensus in multi-agent systems: a review,” Artificial Intelligence Review , vol. 55, no. 5, pp. 3897–3935, 2022

  28. [36]

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

    K. R. McKee et al. , “A multi-agent reinforcement learning model of reputation and cooperation in human groups,” arXiv preprint arXiv:2103.04982, 2021

  29. [37]

    Learning multiagent communication with backpropagation,

    S. Sukhbaatar, R. Fergus et al., “Learning multiagent communication with backpropagation,” Advances in neural information processing systems , vol. 29, 2016

  30. [38]

    Multi- agent actor-critic for mixed cooperative-competitive environments,

    R. Lowe, Y . Wu, A. Tamar, J. Harb, P. Abbeel, and I. Mordatch, “Multi- agent actor-critic for mixed cooperative-competitive environments,” in Proceedings of the 31st International Conference on Neural Information Processing Systems , ser. NIPS’17. Red Hook, NY , USA: Curran A...

  31. [39]

    Responsible emergent multi-agent behavior,

    N. A. Grupen, “Responsible emergent multi-agent behavior,” 2023. [Online]. Available: https://arxiv.org/abs/2311.01609

  32. [40]

    Safe, multi-agent, reinforcement learning for autonomous driving,

    S. Shalev-Shwartz, S. Shammah, and A. Shashua, “Safe, multi-agent, reinforcement learning for autonomous driving,” 2016. [Online]. Available: https://arxiv.org/abs/1610.03295

  33. [41]

    Multi-agent connected autonomous driving using deep reinforcement learning,

    P. Palanisamy, “Multi-agent connected autonomous driving using deep reinforcement learning,” in 2020 International Joint Conference on Neural Networks (IJCNN), 2020, pp. 1–7

  34. [42]

    Failure-scenario maker for rule-based agent using multi-agent adversarial reinforcement learning and its application to autonomous driving,

    A. Wachi, “Failure-scenario maker for rule-based agent using multi-agent adversarial reinforcement learning and its application to autonomous driving,” 2019. [Online]. Available: https://arxiv.org/abs/1903.10654

  35. [43]

    Multi-agent deep reinforcement learning to manage connected autonomous vehicles at tomorrow’s intersections,

    G.-P. Antonio and C. Maria-Dolores, “Multi-agent deep reinforcement learning to manage connected autonomous vehicles at tomorrow’s intersections,” IEEE Transactions on Vehicular Technology , vol. 71, no. 7, pp. 7033–7043, 2022

  36. [44]

    Deep reinforcement learning for autonomous driving: A survey,

    B. R. Kiran et al., “Deep reinforcement learning for autonomous driving: A survey,” IEEE Transactions on Intelligent Transportation Systems , vol. 23, no. 6, pp. 4909–4926, 2022

  37. [45]

    Waymax: An accelerated, data-driven simulator for large-scale autonomous driving research,

    C. Gulino et al. , “Waymax: An accelerated, data-driven simulator for large-scale autonomous driving research,” in Advances in Neural Information Processing Systems , A. Oh, T. Naumann, A. Globerson, K. Saenko, M. Hardt, and S. Levine, Eds., vol. 36. Curran Associates, Inc., 2...

  38. [46]

    Improving understanding of energy autonomy: A systematic review,

    J. K. Juntunen and M. Martiskainen, “Improving understanding of energy autonomy: A systematic review,” Renewable and Sustainable Energy Reviews, vol. 141, p. 110797, 2021

  39. [47]

    Machina-economicus or homo-complexicus: Artificial intelligence and the future of economics?

    G. A. Daneke, “Machina-economicus or homo-complexicus: Artificial intelligence and the future of economics?” Real-World Economic Review, vol. 93, pp. 18–39, 2020

  40. [48]

    Cybersecurity of distributed energy resource systems in the smart grid: A survey,

    J. Chen, J. Yan, A. Kemmeugne, M. Kassouf, and M. Debbabi, “Cybersecurity of distributed energy resource systems in the smart grid: A survey,” Applied Energy, vol. 383, p. 125364, 2025

  41. [49]

    Position: Beyond personhood: Agency, accountability, and the limits of anthropomorphic ethical analysis,

    J. Dai, “Position: Beyond personhood: Agency, accountability, and the limits of anthropomorphic ethical analysis,” in Forty-first International Conference on Machine Learning , 2024

  42. [2020]

    Available: https://iclr.cc/virtual2020/speaker7.html

    [Online]. Available: https://iclr.cc/virtual2020/speaker7.html

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.