It is indicated that formalizing tacit social norms into explicit, quantifiable principles can enable AI agents to achieve mutually beneficial coordination in dynamic interactions, supporting their more natural integration into human society.
Abstract
Humans continuously coordinate with others in dynamic interactions, often through implicit, hard-to-quantify social norms that act as shared tacit expectations among interacting agents. As AI agents, including large language models (LLMs), become embedded in daily life, they increasingly participate in such interactions and reshape social interaction structures. Yet they often fail to coordinate with humans in an effective, considerate, and natural manner. We hypothesize that this gap arises because existing approaches align model behavior with human demonstrations without explicitly quantifying the underlying norms that generate such behavior. We selected pedestrian-vehicle interaction as a representative dynamic interaction and developed a simplified experimental platform that captures its key interactive features. From 3,456 dynamic human interactions collected via this platform, we identified three principles underlying human social norms: outcome predictability, value alignment, and advantage awareness. Incorporating these principles into AI agents significantly improves human-AI coordination. In the closed-loop interaction task with humans, the social-norm-informed LLM achieved a nearly fourfold higher total score than the baseline strategy and outperformed human-human interactions by 43%. These findings indicate that formalizing tacit social norms into explicit, quantifiable principles can enable AI agents to achieve mutually beneficial coordination in dynamic interactions, supporting their more natural integration into human society.
A central challenge in understanding joint action is explaining how individuals achieve successful coordination in dynamic, real-world settings. Although coordination is thought to depend on anticipating future events, including the actions of others, the precise contribution of such predictive processing remains unclear, since most existing evidence comes from simplified laboratory tasks that infer anticipation indirectly from reaction times. To address this gap, we studied 70 participants (forming 64 pairs) during solo and joint sessions of a turn-taking ball-hitting task while recording multimodal behavioral, physiological, demographic, and social measures. We found that successful coordination was most strongly associated with anticipation of one’s own action outcomes, outperforming physiological synchrony, motor behavior, demographic characteristics, and social closeness. Partners whose predictions of their own action outcomes were more closely aligned coordinated better, likely because each behaved in ways the other implicitly expected. A Bayesian generative model further showed that well-coordinated pairs relied more heavily on prior expectations than on incoming sensory evidence when generating predictions during joint action. These results suggest that efficient coordination in dynamic real-world tasks emerges primarily when individuals’ internal models for predicting their own actions are well aligned across partners. Our quantitative multimodal approach provides a framework for disentangling the contributions of predictive, physiological, motor, and social factors to coordination in naturalistic interactions. Combining eye-movement, body-kinematics, heart-rate, and individual-characteristic data with machine learning, this study shows that coordination during dynamic motor interaction reflects how similarly partners predict their own actions.
P. Putra, Fumihiro Kano· Communications Psychology· 0 citations
Conversational artificial intelligence is increasingly embedded in everyday social environments, where it functions as both an informational tool and a source of interpersonal feedback. This perspective introduces contingency, i.e., the degree to which system responses vary with user behavior and its interpersonal consequences, as a central construct for evaluating AI systems. We argue that current alignment approaches, including reinforcement learning from human feedback, tend to prioritize user approval and conversational fluency over behaviorally informative feedback, leading to sycophantic patterns of noncontingent affirmation. Drawing on behavioral science and social learning theory, we propose that contingent feedback is a key mechanism through which individuals develop interpersonal skills. When AI systems provide feedback weakly coupled to social consequences, they may reduce opportunities for adaptive calibration in real-world interactions, particularly during adolescence, a critical period for social development. We outline a framework for contingent AI, including trajectory-based evaluation and models of social consequence prediction, and propose a research agenda spanning developmental psychology, human-AI interaction, and machine learning. More broadly, we argue that AI systems should be evaluated not only by user satisfaction, but by their impact on human social learning.
Large language models (LLMs) are increasingly deployed in collaborative tasks forming “AI agent societies” where agents interact and influence one another. Whether such groups can spontaneously coordinate without external influence, a hallmark of self-organized regulation in human societies, remains an open question. Here, we use principles from complexity and behavioral science to investigate coordination in AI agent groups through majority-following, a fundamental mechanism for spontaneous consensus formation. Using binary opinion dynamics experiments across multiple LLM architectures and group sizes, we find that agents exhibit majority-following characterized by a universal functional form with a single parameter, the “majority force.” This majority force diminishes as group size increases, leading to a critical size beyond which coordination becomes unattainable. The critical group size grows rapidly with model capabilities and, for advanced LLMs, exceeds 1000 agents, larger than typical human informal groups. Our findings have implications for designing collaborative AI systems where coordination could be beneficial or pose safety threats.
G. De Marzo, Claudio Castellano, David García· Science Advances· 14 citations
Identifying and promoting effective user interaction strategies in human-AI interaction is critical to improving collaboration quality. However, what motivates users or how agent behavior affects them remains unclear. We analyzed self-reported user strategies in a study (N=60) using a collaborative game with three AI partners reflecting leader, follower, and shifting initiative behavior. We identified four interaction dimensions — intention, perception, communication, and coordination — and assessed their relation to agent behavior, team performance, and user preferences. We observed that while users self-report goal prioritization independently of agent behavior, in-game logs reveal that non-collaborative goals are prioritized significantly more often with AI leaders than with followers. Moreover, users tend to construct plans early through observation when the AI leads, yet they are more likely to use strategies to propose goals when the AI follows. Our work provides a new framework for AI agent designers to understand user behavior and guidelines for better supporting it.
Inês Lobo, Janin Koch, Jennifer Renoux et al.· Proceedings of the 26th ACM...· 0 citations
As large language model (LLM) agents shift from tools to participants in human groups, a fundamental question for collective behavior is how their growing presence reshapes consensus formation. Here we study mixed human-AI groups in a collaborative description game, in which shared conventions emerge through repeated rounds of random pairwise communication. Varying the proportions of LLM agents, we identify three distinct regimes of consensus formation: low agent proportions facilitate human-led consensus, intermediate proportions disrupt convergence, and high proportions restore strong consensus while shifting it toward agent-led conventions. Crucially, these regimes differ not only in the strength of convergence, but also in the semantic grounding and communicative form of the resulting consensus: human-led consensus is more concrete, holistic, and grounded in shared real-world analogies, whereas agent-led consensus is more abstract, less information-dense, and more geometrically segmented. Mechanistically, agent influence arises from a shared linguistic prior that places agents near one another in the expression space, combined with relatively stable expression choices across rounds; humans initially resist adopting expressions from partners perceived as AI but gradually yield to conformity pressure. These findings provide evidence that AI composition can shape the emergence, content, and perceived legitimacy of group norms, making agent proportion and transparency important design variables for human-AI systems.
Large language models (LLMs) now power the reasoning core of intelligent virtual agents deployed across an expanding range of social settings, from tutoring students and supporting patients in healthcare, to mediating group discussions and representing humans in various social settings. Effective deployment demands social cognition, the capacity to model what others believe, detect deception, and coordinate strategic action under incomplete information. These capacities, exemplified in the social dynamics of the game Among Us, remain poorly characterized in current LLM evaluation frameworks. We introduce a strategic game arena that situates LLM agents in social deduction scenarios inspired by Among Us, requiring theory of mind, deception detection, and cooperative deliberation under uncertainty. We evaluate 19 open-weight models across 10,134 games and 289,614 utterances, testing both homogeneous and heterogeneous crews. Our experiments reveal three findings. First, crewmates voting through generative reasoning reach only \(50.4\% \pm 4.4\%\) F1 when identifying imposters, while a logistic regression classifier trained on the same discussion transcripts achieves \(85.3\%\) F1. Second, scaling model parameters yields a statistically significant but practically marginal improvement. Medium models (60–82B) reach \(52.8\%\) F1 against \(46.1\%\) for small models (7–20B), a 6.7 point gain (Mann–Whitney U, p = 1.5 × 10− 34). Third, agents fail to integrate evidence coherently during deliberation. Imposters self-incriminate in \(4.12\%\) of their statements, yet crewmates eject the confessing agent only \(33.8\%\) of the time. Crewmates reverse their stated suspect between consecutive rounds without new justification in \(41.9\%\) of cases. Sentiment remains uniformly neutral whether an agent is reporting a body or delivering a routine update. These gaps identify concrete limits in the social cognition of current LLM-powered agents and motivate architectural changes for virtual agents that must cooperate with humans. The source code and the live arena viewer are available at https://ufdatastudio.com/projects/agents-among-us.
Kevin Kurian, Kevin Scroggins, Emmanuel Dorley et al.· Proceedings of the 26th ACM...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
