Behavior Of Agents Research Articles

Expressive Multi-Agent Communication via Identity-Aware Learning

Information sharing through communication is essential for tackling complex multi-agent reinforcement learning tasks. Many existing multi-agent communication protocols can be viewed as instances of message passing graph neural networks (GNNs). However, due to the significantly limited expressive ability of the standard GNN method, the agent feature representations remain similar and indistinguishable even though the agents have different neighborhood structures. This further results in the homogenization of agent behaviors and reduces the capability to solve tasks effectively. In this paper, we propose a multi-agent communication protocol via identity-aware learning (IDEAL), which explicitly enhances the distinguishability of agent feature representations to break the diversity bottleneck. Specifically, IDEAL extends existing multi-agent communication protocols by inductively considering the agents' identities during the message passing process. To obtain expressive feature representations for a given agent, IDEAL first extracts the ego network centered around that agent and then performs multiple rounds of heterogeneous message passing, where different parameter sets are applied to the central agent and the other surrounding agents within the ego network. IDEAL fosters expressive communication between agents and generates distinguishable feature representations, which promotes action diversity and individuality emergence. Experimental results on various benchmarks demonstrate IDEAL can be flexibly integrated into various multi-agent communication methods and enhances the corresponding performance.

Analytically Tractable Models for Decision Making under Present Bias

Time-inconsistency is a characteristic of human behavior in which people plan for long-term benefits but take actions that differ from the plan due to conflicts with short-term benefits. Such time-inconsistent behavior is believed to be caused by present bias, a tendency to overestimate immediate rewards and underestimate future rewards. It is essential in behavioral economics to investigate the relationship between present bias and time-inconsistency. In this paper, we propose a model for analyzing agent behavior with present bias in tasks to make progress toward a goal over a specific period. Unlike previous models, the state sequence of the agent can be described analytically in our model. Based on this property, we analyze three crucial problems related to agents under present bias: task abandonment, optimal goal setting, and optimal reward scheduling. Extensive analysis reveals how present bias affects the condition under which task abandonment occurs and optimal intervention strategies. Our findings are meaningful for preventing task abandonment and intervening through incentives in the real world.

Object-Centric Scene Representations Using Active Inference.

Representing a scene and its constituent objects from raw sensory data is a core ability for enabling robots to interact with their environment. In this letter, we propose a novel approach for scene understanding, leveraging an object-centric generative model that enables an agent to infer object category and pose in an allocentric reference frame using active inference, a neuro-inspired framework for action and perception. For evaluating the behavior of an active vision agent, we also propose a new benchmark where, given a target viewpoint of a particular object, the agent needs to find the best matching viewpoint given a workspace with randomly positioned objects in 3D. We demonstrate that our active inference agent is able to balance epistemic foraging and goal-driven behavior, and quantitatively outperforms both supervised and reinforcement learning baselines by more than a factor of two in terms of success rate.

Building back better: Modeling decentralized recovery in sociotechnical systems using strategic network dynamics

The ability of sociotechnical systems to recover from disruptions through decentralized, bottom-up processes is a crucial element of their overall resilience, but it is also among the most overlooked in resilience research. This paper proposes a computational framework leveraging dynamic multi-agent networks to conceptualize, model, and quantify multi-agent, decentralized recovery process building upon strategic network formation methods. This innovative approach addresses limitations in conventional strategic network formation models by accounting for the fluidity of structural shifts induced by the strategic behaviors of individual agents, and by acknowledging the inherent heterogeneity within agent populations. The central objective of this study is to highlight the critical role of bottom-up recovery and present a systematic, adaptable framework for modeling these processes in diverse real-world scenarios. We achieve this through four key elements: modeling post-disruption responses, exploring their interdependencies, accommodating multiple equilibria, and generating metrics for scenario comparison and forecasting. Our dynamic strategic network formation model offers valuable insights into crucial trade-offs in decentralized recovery processes and facilitates the integration of both bottom-up and top-down resilience strategies. Furthermore, it could serve as a predictive tool informing initial network architectural decisions by estimating likely post-disruption equilibrium states based on the initial network structure. Ultimately, this study provides a robust platform for analyzing network efficiency and resilience trade-offs, setting the stage for more informed decision-making in complex sociotechnical systems.

Testing of Deep Reinforcement Learning Agents with Surrogate Models

Deep Reinforcement Learning (DRL) has received a lot of attention from the research community in recent years. As the technology moves away from game playing to practical contexts, such as autonomous vehicles and robotics, it is crucial to evaluate the quality of DRL agents. In this article, we propose a search-based approach to test such agents. Our approach, implemented in a tool called Indago , trains a classifier on failure and non-failure environment (i.e., pass) configurations resulting from the DRL training process. The classifier is used at testing time as a surrogate model for the DRL agent execution in the environment, predicting the extent to which a given environment configuration induces a failure of the DRL agent under test. The failure prediction acts as a fitness function, guiding the generation towards failure environment configurations, while saving computation time by deferring the execution of the DRL agent in the environment to those configurations that are more likely to expose failures. Experimental results show that our search-based approach finds 50% more failures of the DRL agent than state-of-the-art techniques. Moreover, such failures are, on average, 78% more diverse; similarly, the behaviors of the DRL agent induced by failure configurations are 74% more diverse.

A multi-method study of interpersonal complementarity and mentalization

Research finds cross-sectional relationships between mentalizing impairments and maladaptive personality traits. The current study connects mentalizing impairments to dynamic interpersonal processes using a multi-method design. A sample of 218 participants completed the Movie for the Assessment of Social Cognition (MASC; Dziobek et al., 2006) to assess mentalizing ability. Subsequently, participants rated their agentic and communal behavior and their perception of interaction partners’ agentic and communal behavior over 21-days. Mentalizing ability moderated the within-person relationship between behavior and perception for both agency and communion. Worse performance on the MASC was associated with weaker interpersonal complementarity, suggesting that mentalizing impairments lead to deviations from expected patterns of behavior and perception across interpersonal situations. These findings confirm the assumption of Contemporary Integrative Interpersonal Theory that mentalization impairments can disrupt normative interpersonal processes (Cain et al., 2024; Pincus & Hopwood, 2012).

A carbon neutral supply chain management by considering emission-risk minimization and green purchasing through optimal decision-making

This study addresses critical gaps in supply chain management (SCM) by integrating emission-risk minimization (ERM), green purchasing (GP), and profit maximization (PM). The research focuses on the optimal behaviors of manufacturers, agents, and retailers within the SCM framework to achieve carbon neutrality and reduce carbon dioxide emissions (CO2e). This study considers Guangdong province, China, a region facing challenges in optimizing energy systems and meeting CO2e reduction targets. Simulation-based optimization techniques within mathematical models are employed. A design of experiment (DOE) method was used to explore the dynamics of key variables in the SCM environment. Results reveal optimal behaviors for manufacturers, agents, and retailers, demonstrating the ideal values for profit and pricing decisions. Manufacturers optimize production quantity, achieving CO2e reduction and PM through ERM. Agents exhibit a strong commitment to GP practices, enhancing PM and carbon-neutral goals. Retailers get more PM than manufacturers and agents, contributing to a clean environment. Interestingly, retailers make contributions to the clean environment without considering ERM and GP in SCM. The study contributes novel insights by addressing the identified gap in SCM research, emphasizing the joint consideration of ERM, GP, and PM. This research assists manufacturers, agents, and retailers in terms of PM for economic objectives. It cleans the environment through carbon-neutral SCM in society.

Gender and leadership evaluations: what happens if we actually measure expectations–behavior incongruence?

PurposeRole congruity theory and gender stereotypes research suggests men are expected to engage in agentic behavior and women in communal behavior as leaders, and that role violation results in backlash. However, extant gender and leadership research does not directly measure expectations–behavior incongruence. Further, researchers have only considered one condition of role incongruence – display of counter-role behavior – and have not considered the outcomes of failing to exhibit role-congruent behavior. Additionally, few studies have examined outcomes for male leaders who violate gender role prescriptions. The present study aims to address these shortcomings by conducting a novel empirical test of role congruity theory.Design/Methodology/approachThis experimental study used polynomial regression to assess how followers evaluated leaders under conditions of incongruence between follower expectations for men and women leaders’ behavior and leaders’ actual behavior (i.e. exceeded and unmet expectations). Respondents read a fictional scenario describing a new male or female supervisor, rated their expectations for the leader’s agentic and communal behavior, read manipulated vignettes describing the leader’s subsequent behavior, rated their perceptions of these behaviors, and evaluated the leader.FindingsFollowers expected higher levels of communal behavior from the female than the male supervisor, but no differences were found in expectations for agentic behavior. Regardless of whether expectations were exceeded or unmet, supervisor gender did not moderate the effects of agentic or communal behavior expectations–perceptions incongruence on leader evaluations in polynomial regression analyses (i.e. male and female supervisors were not evaluated differently when displaying counter-role behavior or failing to display role-congruent behavior).Originality/valueIn addition to providing a novel, direct test of role congruity theory, the study highlighted a double standard in gender role-congruent behavior expectations of men and women leaders. Results failed to support role congruity theory, which has implications for the future of theory in this domain.

How artificial intelligence cooperating with agent‐based modeling for urban studies: A systematic review

AbstractAs urbanization accelerates, cities become more complex, coming along with more complex urban issues. Agent‐based model (ABM) is a traditional method to simulate activities in a complex system, which has been widely applied in urban studies. However, due to its rigid initial settings, ABM has been criticized for its lack of intelligence, especially in dealing with modern urban issues. With the success of artificial intelligence (AI) and complexity science, it is generally agreed that ABM can be enhanced with AI agents, a promising technology that can bridge the gaps. For that, this article provides a systematic review, in which 10 subsections correspond to 10 different ways that AI can work with ABM in the methodological framework. The sections include that (1) ABM is Al; (2) ABM provides training data for Al; (3) Al provides data for ABM; (4) ABM is a submodule in the ensemble Al; (5) Al leads an optimization framework with ABM participation; (6) Al tunes ABM initialization parameters; (7) Al provides the environment for ABM; (8) Al aids in choosing the agent's attributes; (9) Al provides behaviors for agents in ABM; (10) Al helps to evaluate the performance of ABM. For each case, some typical works are examined for illustration. Finally, we discuss some of the current limitations and prospects for future development.

Effectiveness of Thymol Blue and Phenolphthalein in Sol-Gel Coatings for Sensing and Inhibiting Corrosion on Mild Steel.

Development of smart coatings incorporated with corrosion indicators that can warn of corrosion through signaling the alteration of pH or metal-ion concentration at the corrosion site, and simultaneously mitigate further corrosion, stands out as a highly efficient and economical approach to address corrosion issues. In this context, the present work provides a comprehensive comparison on the effectiveness of thymol blue (TB) and phenolphthalein (PhPh) in both sensing and inhibiting corrosion on mild steel. While most of the works primarily focused on independently investigating the corrosion indication behavior of different active agents, our study intends to offer valuable insights through the comparative analysis of TB and PhPh in making well-informed decisions when selecting the most effective indicator for a given set of application scenarios. Initially, TB and PhPh were dissolved in a 3.5% NaCl solution, and their ability to indicate and inhibit corrosion on mild steel was verified through a drop test and an electrochemical study, respectively. An electrochemical study and a salt spray test were carried out to evaluate the corrosion sensing and inhibiting capabilities of the coating produced on a mild steel substrate by integrating TB and PhPh in the sol-gel matrix. The synergistic effect of TB and PhPh in sensing and inhibiting corrosion was also studied by combining TB and PhPh in the sol-gel matrix. It was found that TB-based coatings showed approximately 8 times better inhibiting performance (in terms of charge transfer resistance) when compared to PhPh-based coatings, while the PhPh-based coating was around 16 times more sensitive (with respect to concentration and time) in indicating corrosion when compared to TB-based coatings on a mild steel substrate. Nevertheless, the combination of TB and PhPh in a 1:3 ratio incorporated into the sol-gel coating exhibited increased sensitivity in detecting corrosion and showed improved corrosion inhibition on mild steel.

Towards reusable building blocks for agent-based modelling and theory development

Despite the increasing use of standards for documenting and testing agent-based models (ABMs) and sharing of open access code, most ABMs are still developed from scratch. This is not only inefficient, but also leads to ad hoc and often inconsistent implementations of the same theories in computational code and delays progress in the exploration of the functioning of complex social-ecological systems (SES). We argue that reusable building blocks (RBBs) known from professional software development can mitigate these issues. An RBB is a submodel that represents a particular mechanism or process that is relevant across many ABMs in an application domain, such as plant competition in vegetation models, or reinforcement learning in a behavioural model. RBBs need to be distinguished from modules, which represent entire subsystems and include more than one mechanism and process. While linking modules faces the same challenges as integrating different models in general, RBBs are “atomic” enough to be more easily re-used in different contexts. We describe and provide examples from different domains for how and why building blocks are used in software development, and the benefits of doing so for the ABM community and to individual modellers. We propose a template to guide the development and publication of RBBs and provide example RBBs that use this template. Most importantly, we propose and initiate a strategy for community-based development, sharing and use of RBBs. Individual modellers can have a much greater impact in their field with an RBB than with a single paper, while the community will benefit from increased coherence, facilitating the development of theory for both the behaviour of agents and the systems they form. We invite peers to upload and share their RBBs via our website - preferably referenced by a DOI (digital object identifier obtained e.g. via Zenodo). After a critical mass of candidate RBBs has accumulated, feedback and discussion can take place and both the template and the scope of the envisioned platform can be improved.

The Relationship Between Chinese High School English Learners’ Classroom Engagement and English Academic Achievement: A Case Study of English Learners in a High School in Henan China

Classroom engagement is often referred to as a crucial factor for students’ academic achievement. However, within the realm of second language acquisition, only a limited number of scholars made empirical investigations into the association between engagement in the English classroom and English academic achievement. In particular, research on second language learners at the basic education stage is lacking. Therefore, this study examines the current state of English classroom engagement among high school students and its relationship with English academic achievement through a case study of students in Henan China. This study found that high school students’ overall English classroom engagement was high, but individual-based agentic behavior engagement was low. Furthermore, male students exhibited significantly lower levels of various dimensions of English classroom engagement compared to female students. Moreover, interaction-based emotional engagement, individual-based cognitive engagement, individual-based behavioral engagement, and interaction-based conventional behavioral engagement were found to significantly predict students’ English academic achievement. Furthermore, gender is found to exert a significant moderating influence on the association between individual-based cognitive engagement and English academic achievement among students. Implications of these findings for English instruction are discussed.

An uncertainty quantification framework for agent-based modeling and simulation in networked anagram games

ABSTRACT In a networked anagram game, players are provided letters with possible actions of requesting letters from their neighbours, replying to letter requests, or forming words. The objective is to form as many words as possible as a team. The experimental data show that behaviours among players can vary significantly. However, simulations using agent-based models (ABM) in the literature often have not incorporated proper uncertainty quantification methods to characterise diverse behaviours of players. In this work, we propose an uncertainty quantification framework to build, exercise, and evaluate agent behaviour models and simulations for networked group anagram games. Specifically, using the data of game experiments, the proposed framework considers the clustering of game players based on their performance to reflect players’ heterogeneity. Moreover, we also quantify uncertainty within each cluster through statistical modelling and inference. Numerical studies of networked game configurations are conducted to demonstrate the merits of the proposed framework.

The nonmeek inherit the earth: Children generalize dominance, but not submissiveness.

Children and even infants have clear intuitions about power early in development; they can infer who is dominant and subordinate from observing a single interaction. However, it is unclear what children infer about each individual's status from these interactions-do they think dominants and subordinates will maintain their status when interacting with novel partners? In three experiments, we investigate this question. Children (4- to 10-year-olds, N = 365) heard stories about a dominant and subordinate agent and predicted the dominant or subordinate agent's behavior with a novel agent. In all studies, we found that 7- to 10-year-olds generalized dominance, thinking the dominant would again be dominant or "in charge," both for social power (e.g., granting permission) in Study 1 and physical dominance (e.g., a fistfight) in Studies 2 and 3. Furthermore, although they believed dominant agents would win dominance contests (fistfights), they did not believe they would win contests unrelated to dominance (math contests). Younger children did not generalize social power (Study 1) but did generalize physical dominance (Studies 2 and 3). However, even for physical dominance, their generalizations were less selective (i.e., they believed the dominant would win fistfights and math contests). Notably, neither age group generalized an agent's submissiveness in any of the studies-they did not believe a subordinate agent would again be subordinate when paired with a novel partner. We discuss how these results extend past work on children's developing intuitions about dominance and prompt deeper questions about the inferences children draw from dominance interactions. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

Selective Action Prediction in Infancy Depending on Linguistic Cues: An EEG and Eyetracker Study.

Humans' capacity to predict actions and to socially categorize individuals is at the basis of social cognition. Such capacities emerge in early infancy. By 6 months of age, infants predict others' reaching actions considering others' epistemic state. At a similar age, infants are biased to attend to and interact with more familiar individuals, considering adult-like social categories such as the language people speak. We report that these two core processes are interrelated early on in infancy. In a belief-based action prediction task, 6-month-old infants (males and females) presented with a native speaker generated online predictions about the agent's actions, as revealed by the activation of participants' sensorimotor areas before the agent's movement. However, infants who were presented with a foreign speaker did not recruit their motor system before the agent's action. The eyetracker analysis provided further evidence that linguistic group familiarity influences how infants predict others' actions, as only infants presented with a native speaker modified their attention to the stimuli as a function of the agent's forthcoming behavior. The current findings suggest that infants' emerging capacity to predict others' actions is modulated by social cues, such as others' linguistic group. A facilitation to predict and encode the actions of native speakers relative to foreign speakers may explain, in part, why infants preferentially attend to, imitate, and learn from the actions of native speakers.

How learning-goal-oriented newcomers develop leader-member exchange (LMX): From agentic and communal behavior perspectives

Adopting the dual perspective model of agency and communion and role-making theory of leader–member exchange (LMX) development, this study explored whether and how newcomers with a high learning goal orientation (LGO) develop high-quality social exchange relationships with their leaders. Through a three-wave field survey of 199 organizational newcomers who were graduates entering the workplace, we found a direct and significant relationship between the newcomers’ LGO and LMX. We also found that their agentic and communal behaviors—specifically, self-improvement voice and knowledge sharing, respectively—served as dual mediating mechanisms of this relationship. This study answers the call for more research on how employees’ LGO motivates them to engage in behaviors conducive to social exchange development with their leaders. It also contributes to the emerging research on followership and understanding LMX development beyond the leader–member dyad. The theoretical and practical implications of the findings are discussed.

Collaborative governance in integrated preschool education: A quadrilateral evolutionary game model analysis

The collaborative governance of inclusive preschool education (IPE) is a process in which multiple subjects work together to promote the advancement and development of IPE by taking a series of governance measures. Based on precise identification of multiple stakeholders in collaborative governance of integrated preschool education, it applied the quadrilateral evolutionary game model and Lyapunov's first law to analyze the stability of behavioral strategy choice by multiple subjects in promoting the practice of collaborative governance of IPE and the possible equilibrium points in the system. The influence of key elements on the system evolution is analyzed using MATLAB 2019 simulation. The study reveals that the evolution and trend of one game agent's behavior are closely related to the strategy choice of others. The willingness of the main body of kindergarten, social organization, and parents to participate in the collaborative governance of IPE are related to the government's active promotion to perform its leading functions at the government level. The willingness for collaborative governance of integrated education can help improve mutual trust and the main body's benefits, thus promoting the collaborative behavior of multiple subjects. Complementary resources, strict supervision, reasonable rewards and punishments can promote the collaborative behavior of multiple subjects.

Effects of topological structure and destination selection strategies on agent dynamics in complex networks

We analyzed agent behavior in complex networks: Barabási–Albert, Erdos–Rényi, and Watts–Strogatz models under the following rules: agents (a) randomly select a destination among adjacent nodes; (b) exclude the most congested adjacent node as a potential destination and randomly select a destination among the remaining nodes; or (c) select the sparsest adjacent node as a destination. We focused on small complex networks with node degrees ranging from zero to a maximum of approximately 20 to study agent behavior in traffic and transportation networks. We measured the hunting rate, that is, the rate of change of agent amounts in each node per unit of time, and the imbalance of agent distribution among nodes. Our simulation study reveals that the topological structure of a network precisely determines agent distribution when agents perform full random walks; however, their destination selections alter the agent distribution. Notably, rule (c) makes hunting and imbalance rates significantly high compared with random walk cases (a) and (b), irrespective of network types, when the network has a high degree and high activity rate. Compared with the full random walk in (a) and (b) increases the hunting rate while decreasing the imbalance rate when activity is low; however, both increase when activity is high. These characteristics exhibit slight periodic undulations over time. Furthermore, our analysis shows that in the BA, ER, and WS network models, the hunting rate decreases and the imbalance rate increases when the system disconnects randomly selected nodes in simulations where agents follow rules (a)–(c) and the network has the ability to disconnect nodes within a certain time of all time steps. Our findings can be applied to various applications related to agent dynamics in complex networks.

COVSIM: A stochastic agent-based COVID-19 SIMulation model for North Carolina

We document the evolution and use of the stochastic agent-based COVID-19 simulation model (COVSIM) to study the impact of population behaviors and public health policy on disease spread within age, race/ethnicity, and urbanicity subpopulations in North Carolina. We detail the methodologies used to model the complexities of COVID-19, including multiple agent attributes (i.e., age, race/ethnicity, high-risk medical status), census tract-level interaction network, disease state network, agent behavior (i.e., masking, pharmaceutical intervention (PI) uptake, quarantine, mobility), and variants. We describe its uses outside of the COVID-19 Scenario Modeling Hub (CSMH), which has focused on the interplay of nonpharmaceutical and pharmaceutical interventions, equitability of vaccine distribution, and supporting local county decision-makers in North Carolina. This work has led to multiple publications and meetings with a variety of local stakeholders. When COVSIM joined the CSMH in January 2022, we found it was a sustainable way to support new COVID-19 challenges and allowed the group to focus on broader scientific questions. The CSMH has informed adaptions to our modeling approach, including redesigning our high-performance computing implementation.

Investigation of the Thermally Thick Alternative Reducing Agent Behavior in the Raceway Zone

Investigation of the Thermally Thick Alternative Reducing Agent Behavior in the Raceway Zone