Modeling Human Behavior in a Strategic Network Game with Complex Group Dynamics
Jonathan Skaggs, Jacob W. Crandall · 2025
Read Modeling Human Behavior in a Strategic Network Game with Complex Group Dynamics by Jonathan Skaggs, Jacob W. Crandall (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-03795.txt.
Introduction
Jonathan Skaggs and Jacob Crandall compared ways of modeling human players in twenty twenty-five in a strategic network game with rich group dynamics. Human networks shape inequality, health, and bullying, but simulating them needs behavioral models that act like people. The test bed was the Junior High Game, where players form and break alliances in a small society.
Methods differed in whether they assumed players copy others' behavior or reason about their communities, and whether they modeled average behavior or its full distribution. The winner, hCAB, models the distribution and assumes community-aware play. On small societies it closely mirrored human population dynamics, and in a user study participants struggled to tell its agents from real humans. Both checks matter: plausible crowds and plausible individuals.
Catalog record
| CREATOR | Jonathan Skaggs, Jacob W. Crandall |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 44 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2505-03795
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Modeling Human Behavior in a Strategic Network Game with Complex Group Dynamics Jonathan Skaggs, Jacob W. Crandall Abstract . Human networks greatly impact important societal outcomes, including wealth and health inequality, poverty, and bullying. As such, understanding human networks is critical to learning how to promote favorable societal outcomes. As a step toward better understanding human networks, we compare and contrast several methods for learning models of human behavior in a strategic network game called the Junior High Game (JHG) ( Skaggs et al., 2024 ) . These modeling methods differ with respect to the assumptions they use to parameterize human behavior (behavior matching vs. community-aware behavior) and the moments they model (mean vs. distribution). Results show that the highest-performing method, called hCAB, models the distribution of human behavior rather than the mean and assumes humans use community-aware behavior rather than behavior matching. When applied to small societies, the hCAB model closely mirrors the population dynamics of human groups (with notable differences). Additionally, in a user study, human participants had difficulty distinguishing hCAB agents from other humans, thus illustrating that the hCAB model also produces plausible (individual) behavior in this strategic network game. \doi NQKD4743 C. Amato, L. Dennis, V. Mascardi, J. Thangarajah (eds.) Conference: Proc. of the 25th International Conference on Autonomous Agents and Multiagent Systems (AAMAS 2026); May 25 – 29, 2026; Paphos, Cyprus 1. Introduction Human networks (collections of interconnected people) influence how individuals exchange resources ( Willer, 1999 ) , form alliances ( Cranmer et al., 2012 ; Gulati, 1995 ) , and navigate competitive pressures ( Burt, 2005 ; Vu et al., 2006 ) . These networks, which are studied in many disciplines ( Svendsen and Svendsen, 2009 ; Levin, 1998 ) , exhibit emergent behaviors related to inequality ( Barabási and Albert, 1999 ; Salganik et al., 2006 ) , information flow ( Harush and Barzel, 2017 ) , and social behaviors ( Shoham and Tennenholtz, 1997 ; Christakis and Fowler, 2010 ; Centola, 2010 ; Centola, 2018 ; Veenstra and Huitsing, 2021 ) . While the study of human networks has yielded deep insights ( Newman, 2003 ; Easley and Kleinberg, 2010 ; Bar-Yam, 2019 ; Pagan and Dörfler, 2019 ) , understanding the precise decision-making processes that drive human behavior within these networks remains a major challenge ( Young, 2015 ) . What determines whether an individual cooperates or competes? How do local interactions scale up to produce global patterns of stability, inequality, or conflict? How might an individual change their behavior to encourage preferred social change? Addressing these questions requires computational models that capture the complexities of human decision-making in networked environments. In human networks, individuals make decisions based on personal incentives, past experiences, and the social landscape. Sometimes individual incentives align with the collective good, like the invisible hand ( Smith, 1776 ) , but other times they are diametrically opposed, as in the free-rider problem ( Olson, 1965 ) . These individual actions lead to emergent behaviors that shape societal outcomes. Extensive studies have explored emergent properties of human networks ( Holland, 1996 ; Jensen, 2022 ; Shoham and Leyton-Brown, 2008 ) , investigating, for example, how randomness affects network structures ( Erdős and Rényi, 1959 ) , how power distributes among individuals ( Barabási and Albert, 1999 ) , and how self-organization arises within different social ( Watts and Strogatz, 1998 ) and economic settings ( Arthur, 1994 ; Young, 2001 ) . These macroscopic perspectives on how human societies function are valuable, yet they are often insufficient for fully understanding human networks and how they evolve. More detailed models of human decision-making on networks are needed. To help address this challenge, we model human behavior in a strategic network game designed to capture the emergent properties of complex (networked) societies, including wealth and power distributions and mixing patterns defined by dyadic ( Gouldner, 1960 ; Fehr and Gächter, 2000 ) , triadic ( Granovetter, 1973 ; Nowak and Sigmund, 2005 ) , and group dynamics ( Nowak, 2006 ; Bowles and Gintis, 2011 ; Henrich, 2015 ) . These dynamics emerge due to power asymmetry, mixed-motives, resource management, and interconnectedness ( Easley and Kleinberg, 2010 ; Jensen, 2022 ; Thurner et al., 2018 ) , and produce networks that are dynamic, directed, weighted, and signed. Because these dynamics are encoded within the Junior High Game (JHG) ( Skaggs et al., 2024 ) , we use this strategic network game to study emergent behavior. A fundamental challenge in modeling human decision-making in the JHG lies in determining how to parameterize and represent behavior ( Rahwan and others, 2019 ) . In this paper, we consider models that make different assumptions about human decision-making. One approach, behavior matching ( Axelrod, 1984 ) , assumes that agents imitate the observed actions of others, while another approach, community-aware behavior ( Skaggs et al., 2024 ) , considers broader structural and social cues within the network. We also examine how modeling different statistical moments (the mean verses the distribution of behavior) impact model performance. Together, this produces four different algorithms for modeling human behavior in the JHG. As a starting point to understanding how well these algorithms model human behavior, we compare and contrast their behavior in small-scale societies. A figure showing a hypothetical game designed to illustrate various dynamics of the Junior High Game. (a) (b) (c) Figure 1. Example scenario illustrating various JHG dynamics. (a) Popularity of the players over time. (b-c) Token allocations in rounds 1-4 and 5-6, respectively. Red dashed arrows indicate tokens used to attack while black solid arrows indicate giving. A figure showing a hypothetical game designed to illustrate various dynamics of the Junior High Game. We evaluate the ability of these modeling algorithms to produce human behavior in the JHG in two ways. First, we compare and contrast the population dynamics of agent societies with those of human groups. Results show that societies of one of these agents, called hCABs, have population dynamics that mirror (with some differences) human populations. Second, we conduct a user study to identify the extent to which individual hCABs exhibit human-like tendencies. Results of this study show that human participants have difficulty distinguishing hCAB agents from other human players. These findings not only inform the development of models that can simulate human behavior in the JHG, but also offer perspectives on the mechanisms that drive the dynamics of human networks. 2. A strategic network game Human networks are punctuated by power asymmetry, mixed motives, resource management, and interconnectedness ( Easley and Kleinberg, 2010 ; Jensen, 2022 ; Thurner et al., 2018 ) , which (along with other properties) combine to produce emergent properties of human networks. These characteristics are modeled in the Junior High Game (JHG) ( Skaggs et al., 2024 ) , a strategic game that induces a dynamic (time varying), directed, weighted, and signed network. We paraphrase Skaggs et al. (2024) to review this game: The JHG is played by a set of players I over a series of rounds in which each player j\in I seeks to become popular . Initially, every player is assigned a popularity, which changes over time based on interactions, represented by token