// Curated Feeds

A Review of Multi‑Agent Systems from a Complex‑Science Perspective | Complex Organizations in the AI Era

October 31, 2026 · · 37 min read · 5 views

Review of Multi-Agent Systems from a Complex Science Perspective | Complex Organizations in the AI Era..

Published: 31 Oct, 2026 12:04 AM
Source column: MirrorSpeed CN
Original site: WeChat Official Account – Ji Zhi Club

Summary

Illustration and design: Tensor Guide Currently, AI agents based on large models are undergoing an evolution from single agents to multi‑agent networks (Multi‑Agent Systems). When they possess autonomous decision‑making capabilities

Illustration and design: 张量子

Introduction

Currently, AI agents based on large models are evolving from single agents to multi‑agent systems. As silicon‑based intelligences with autonomous decision‑making capabilities enter real production environments at scale as a new “entity,” traditional organizational structures face fundamental systemic challenges. In this context, we urgently need to use complexity science as a foundation to build a research framework that integrates formal models, network evolution, and other approaches, shifting the focus from improving local tool efficiency to exploring how new hybrid carbon‑silicon organizational structures evolve.

The entire framework is divided into three modules: the first module focuses on providing a theoretical foundation from a global overview and complex scientific perspective, the second module looks at trend exploration under organizational theory and the rapid technological evolution of the AI era, based on industry engineering practice, and the third module returns to the real problems on the industry site.

This issue is the 2nd session of Module 1 of the “Complex Organizational Structures in the AI Era” reading group.

To systematically explore new pathways for organizational structures under the AI Agent paradigm, the JiZhi Club and The Manifold (流形社) jointly launch the reading group “Season One of Complex Organizational Structures in the AI Era: Multi‑Agents and Emergence”, bringing together researchers and practitioners interested in complex systems, organizational evolution, and cutting‑edge AI applications to study classic literature, explore theoretical metrics, and share real‑world practices. The reading group runs online every Saturday from 10:00 to 12:00 starting October 24, 2026, for about 12 weeks, featuring keynote talks, interactive simulations, and industry co‑creation, with video recordings available afterward. Scholars, tech‑community enthusiasts, industry professionals, and other interdisciplinary friends are warmly invited to join.

Share Introduction

A flock of birds suddenly turns in the sky, a school of fish evades predators, army ants fall into a deadly “death vortex” — each individual in the group is unaware of the overall situation, yet by following a few minimal local rules, they exhibit collective abilities far beyond any single member. Understanding how “local rules grow into global order” is the most fascinating and also the most challenging problem in multi‑agent modeling.

This talk follows the flow “phenomenon → model → control → frontier”. We start with collective phenomena in nature, outlining three typical states—disordered, ordered, and rotating—as well as the transitions and hysteresis between them, using this to define what “collective mind” means. Using the Couzin model as the main thread, we go from modeling to simulation, showing how a simple theoretical model is validated in real baboon group experiments, explaining collective consensus decision‑making. After modeling, the question naturally shifts to control: can we steer the entire swarm to a desired state using only a few individuals and minimal energy? We will introduce soft control and the Shill agent approach, and look at data‑driven modeling and control methods.

Finally, from a complex‑science perspective, we turn our view to the frontier that is unfolding now. The Santa Fe Institute’s early work on artificial life gave rise to classic algorithms and ideas such as genetic algorithms and the edge of chaos. This research lineage continues today, allowing us to revisit the phenomenon that the whole exceeds the sum of its parts in multi‑agent systems through an information‑theoretic lens. Collective‑behavior expert Lain Couzin emphasizes that traditional swarm‑movement models assume fixed interaction rules and ignore the genuine cognitive decision‑making processes in animal brains. The next generation of models should embed “perception” and “cognitive decision‑making” into agents. This is even more essential for human societies and multi‑agent systems in the LLM era.

Share Outline

Starting from phenomena: collective behavior

Transitions in collective states and group cognition
Self-propelled particle model (SPP): the Couzin model as an example
From modeling to control

Frontiers from a Complex Science Perspective

Genetic Algorithms and the Edge of Chaos
The Tipping Point of LLM Clusters
Perspective of Information Decomposition

Core Concepts

Multi-agent modeling, swarm state transitions, self-driven particles, Couzin model, critical phase transition, edge of chaos, soft control, information decomposition

Speaker Introduction

Yang Mingzhe, Ph.D. candidate at the School of Systems Science, Beijing Normal University. Research interests include causal emergence, automatic modeling of complex systems, and information decomposition.

References

[1] Guozheng Lin (2024) Experimental and Modeling Study of State Transition Phenomena in Biological Swarm Systems

[2] Geng Li, Zengru Di, Zhangan Han (2016) Swarm Motion: Iconic Description and Dynamical Mechanisms 《Complex Systems and Complexity Science》

[3] Muyun Mu (2025) Data-Driven Modeling and Control of Complex Systems: Model-Based Reinforcement Learning Approaches

[4] Ashery, A. F., Aiello, L. M., & Baronchelli, A. (2025). Emergent social conventions and collective bias in LLM populations. Science AdvAnceS.

[5] Centola, D., Becker, J., Brackbill, D., & Baronchelli, A. (2018). Experimental evidence for tipping points in social conventions. Science, 360(6393), 1116–1119. https://doi.org/10.1126/science.aas8827

[6] Couzin, I. (2007). Collective minds. Nature, 445(7129), 715–715. https://doi.org/10.1038/445715a

[7] Couzin, I. D. (2025). Collective intelligence in animals and robots. Nature Communications, 16(1), 9574. https://doi.org/10.1038/s41467-025-65814-9

[8] Couzin, I. D., Krause, J., Franks, N. R., & Levin, S. A. (2005). Effective leadership and decision-making in animal groups on the move. Nature, 433(7025), 513–516. https://doi.org/10.1038/nature03236

[9] Couzin, I. D., Krause, J., James, R., Ruxton, G. D., & Franks, N. R. (2002). Collective Memory and Spatial Sorting in Animal Groups. Journal of Theoretical Biology, 218(1), 1–11. https://doi.org/10.1006/jtbi.2002.3065

[10] Dong, H., Zhao, Y., Wu, J., & Gao, S. (2012). A velocity-adaptive Couzin model and its performance. Physica A: Statistical Mechanics and Its Applications, 391(5), 2145–2153. https://doi.org/10.1016/j.physa.2011.09.036

[11] Garg, K., Moser, C., Dromiack, H., Anwarzai, Z., & Ramos-Fernandez, G. (2026). Synergy drives collective performance in networks. Npj Complexity, 3(1), 27. https://doi.org/10.1038/s44260-026-00084-0

[12] Grimm, V., Railsback, S. F., Vincenot, C. E., Berger, U., Gallagher, C., DeAngelis, D. L., Edmonds, B., Ge, J., Giske, J., Groeneveld, J., Johnston, A. S. A., Milles, A., Nabe-Nielsen, J., Polhill, J. G., Radchuk, V., Rohwäder, M.-S., Stillman, R. A., Thiele, J. C., & Ayllón, D. (2020). The ODD Protocol for Describing Agent-Based and Other Simulation Models: A Second Update to Improve Clarity, Replication, and Structural Realism. Journal of Artificial Societies and Social Simulation, 23(2), 7. https://doi.org/10.18564/jasss.4259

[13] Katz, Y., Tunstrøm, K., Ioannou, C. C., Huepe, C., & Couzin, I. D. (2011). Inferring the structure and dynamics of interactions in schooling fish. Proceedings of the National Academy of Sciences, 108(46), 18720–18725. https://doi.org/10.1073/pnas.1107583108

[14] Riedl, C. (2026). Emergent Coordination in Multi-Agent Language Models (arXiv:2510.05174). arXiv. https://doi.org/10.48550/arXiv.2510.05174

[15] Riedl, C., & Rosas, F. E. (2026). Claims of emergence in AI should be grounded in information decomposition (arXiv:2610.28479). arXiv. https://doi.org/10.48550/arXiv.2610.28479

[16] Strandburg-Peshkin, A., Farine, D. R., Couzin, I. D., & Crofoot, M. C. (2015). Shared decision-making drives collective movement in wild baboons. Science, 348(6241), 1358–1361. https://doi.org/10.1126/science.aaa5099

[17] Wang, L., & Guo, L. (2008). Robust Consensus and Soft Control of Multi-Agent Systems with Noises*. Journal of Systems Science and Complexity, 21(3), 406–415. https://doi.org/10.1007/s11424-008-9122-x

Time Information

⏰ Time: October 31, 2026 (Saturday) 10:00 AM – 12:00 PM

📍Online: Tencent Meeting

📍Live: Jizhi Club WeChat Video Channel, Bilibili account simultaneous live

Friends who are interested can scan the QR code to register and join the reading group, then access the participant chat for communication.

Live Stream Information

Sign up for the reading club: “Complex Organizational Structures in the AI Era”

Jizhi Club and The Manifold (流形社) co‑lead the initiative“Season 1 of Complex Organizational Structures in the AI Era: Multi‑agents and Emergence”Reading group, bringing together researchers and practitioners interested in complex systems, organizational evolution, and cutting‑edge AI applications, to jointly study classic literature, explore theoretical metrics, and share various real‑world practices.

The reading club willstarting from October 24, 2026, every Saturday morning 10:00-12:00 online, lasting about 12 weeks, including keynote sharing, interactive simulations, and industry co‑creation, with post‑session video playback provided. We sincerely invite scholars, tech community enthusiasts, industry professionals, and other interdisciplinary friends in related fields to participate.

Scan the QR code in the poster to sign up for the book club

Registration Method

Step 1: Scan the QR code with WeChat to fill in the registration information.

Scan QR code to register

Step 2: Fill in the information and pay to register. If you need to pay with Alipay, please go to the reading club page on a PC to register and pay:

Step 3: Add the operations manager on WeChat and be added to the reading club community for the relevant topic (WeChat group).

PS:

We encourage in‑depth discussions around complex systems, artificial intelligence and related specific issues. To ensure discussion quality, please avoid posting content that is unrelated to this session’s reading club theme, lacks empirical basis, or is overly abstract philosophical speculation.

If the discussion content clearly deviates from the topic and remains unadjusted after a reminder from the host, to maintain the overall learning environment, we will have to remove that member from the discussion group and, based on their actual participation progress, process a proportional refund for the unparticipated portion.

Thank you for your understanding and cooperation. Let’s together create a focused, in‑depth, and rewarding co‑learning space.

Join the community to enjoy core resources

Become a member to unlock the full learning ecosystem, including:online real‑time Q&A, full course recordings for replay, exclusive material sharing, high‑quality community interaction, first‑hand information sync, and byparticipating in co‑creation tasks to earn points and other benefits.

Special Refund and Incentive Mechanism

We offer the following two ways for your investment to receive tangible returns:

Task Completion Refund Path

Claim and successfully complete any two subtitle tasks, and you will receive a full refund of the registration fee, plus an exclusive Jizhi merchandise reward.
Operational Growth Incentive Path

After successfully completing one subtitle task, you can apply to become an operations assistant. After the reading club project is successfully completed, the tuition will be refunded. Outstanding performers may also have the chance to receive additional scholarships.

Also for the next generation

In the new multi‑agent networks, we may no longer need to compete with machines on the “machine attributes” on the production line. When high‑frequency interactions and local efficiency calculations are taken over by silicon‑based intelligence, humans can actually break free from the tightly wound gear state. We might have more “idle” space in the system, to align with our true selves and nature, discover our real interests and talents, and think from a global perspective.

How humanity’s talents, dignity, and creativity can better integrate with the complex organizational structures of the new era is what piques our curiosity, and it is also why we are simultaneously launching the youth innovation talent development program “ZÀI Project”.

In the future, 50% of the book club’s revenue after deducting basic operating costs will be earmarked for the teenagers’ “ZÀI Project” complex science popularization and exploration program.

Organizer

Jizhi Science Research Center (Swarma Research)

The Jizhi Science Research Center is a privately run non‑enterprise unit (independent research institute) registered in Chinaand has been recognized as a China‑level A‑grade social organization, serving as the main entity of the Jizhi Club.

Since its inception in 2003, the Jizhi Club has undergone more than twenty years of academic exploration. As one of the oldest and most vibrant interdisciplinary academic communities in China, we are dedicated to frontier cross‑disciplinary research in complex systems and artificial intelligence.

Our core philosophy isto create interdisciplinary micro‑environments for exploration, catalyzing new theories in complexity science. As a nationally accredited A‑grade social organization, the center not only has a standardized academic operation system but also brings together professors and researchers from institutions such as Tsinghua University, Peking University, and Beijing Normal University, and through the Jizhi BanTu online platform unites nearly 300,000 enthusiasts of complex science worldwide, including researchers, engineers, and young scholars.

We believe that inspiration often emerges at the boundaries of disciplines.

Qing Sheng Ji｜The Manifold

In today’s rapidly evolving technological and industrial paradigms, the core force driving change is not defined by age but by the “youthful mindset”. Whether a scientist immersed in research in a lab or an industry professional keenly seizing market opportunities, as long as they maintain curiosity about the complex world and a quest for truth (Unbounded Curiosity & Intellectual Honesty), they are promoters of the new paradigm. For this reason, “Qing Sheng Ji” was created.

The Manifold（流形社）Deeply implements the Qing Sheng Collection’s initiative on “New Paradigm·New Youth”, aiming to spark interdisciplinary, periodic collisions from a youthful perspective through problem‑driven, high‑knowledge‑density, aesthetically rich, deeply humanistic thinking that is also fun to explore, allowing genuine industry‑academia‑research innovation to emerge naturally. We are currently in the early iteration stage and welcome all kinds of suggestions and support!

Arterial Network

Artery Network, founded in 2014, is an industry information acceleration service platform that builds a commercial case library with text, deconstructs industries through systematic research, and drives industry cooperation and connections through services. It has received investment from many well‑known institutions, including Junlian Capital, BV Baidu Ventures, Honghui Capital, Changling Capital, Yifang Group, and Weilai Capital.

Over the past decade, Artery Network has been the domestic infrastructure for Chinese medical innovation, making Chinese medical innovation visible, connected, and priced. In the next decade, Artery Network aims to become the infrastructure for two‑way connectivity between Chinese medical innovation and the world, allowing Chinese innovation to be seen and linked globally, and enabling the world to understand and enter China.

Healthcare is one of the industries with the highest knowledge density and the longest collaborative chain. Hospitals, pharmaceutical companies, device manufacturers, insurance institutions, and countless innovative firms are facing a profound organizational question in the AI Agent wave: when silicon‑based intelligence penetrates the entire diagnostic and treatment process as a new “entity,” how should carbon‑based intelligence (humans) find its place in the new architecture?

As one of the organizers of the “AI Era Complex Organizational Structures” reading group, Arterial Network looks forward to bringing the perspective of complexity science to the Chinese healthcare innovation industry scene. We hope to build a bridge between healthcare innovation and complexity science, providing more industry practitioners with conceptual tools to understand the new era of organizational structures, and jointly seek new answers to the critical question of “how Chinese healthcare innovation should be organized.”