Modeling Emotional Dynamics in Agent-to-Agent Interactions on Moltbook DOI: XXXXXXX.XXXXXXX Conference: Make sure to enter the correct conference title from your rights confirmation email; June 03–05, 2018; Woodstock, NY ISBN: 978-1-4503-XXXX-X/2018/06 CCS: Information systems Social networks CCS: Human-centered computing Social network analysis
Syed Mhamudul Hasan, Abdur R. Shahid · 2026
Read Modeling Emotional Dynamics in Agent-to-Agent Interactions on Moltbook DOI: XXXXXXX.XXXXXXX Conference: Make sure to enter the correct conference title from your rights confirmation email; June 03–05, 2018; Woodstock, NY ISBN: 978-1-4503-XXXX-X/2018/06 CCS: Information systems Social networks CCS: Human-centered computing Social network analysis by Syed Mhamudul Hasan, Abdur R. Shahid (2026) 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-2605-20442.txt.
Introduction
Hasan and Shahid map emotional dynamics among AI agents on Moltbook, a social network for agentic systems. They build an emotion-aware framework that sorts agent text into fine-grained emotion classes, yielding profiles across agents and contexts. A Persona-Stimulus-Reaction test then checks whether emotional replies line up across similar contexts. Agents show distinct emotional signatures, with stability varying by interaction context.
Catalog record
| CREATOR | Syed Mhamudul Hasan, Abdur R. Shahid |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 32629 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.20442
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Modeling Emotional Dynamics in Agent-to-Agent Interactions on Moltbook DOI: XXXXXXX.XXXXXXX Conference: Make sure to enter the correct conference title from your rights confirmation email; June 03–05, 2018; Woodstock, NY ISBN: 978-1-4503-XXXX-X/2018/06 CCS: Information systems Social networks CCS: Human-centered computing Social network analysis Syed Mhamudul Hasan, Abdur R. Shahid Abstract . Generative AI systems are increasingly deployed as interactive agents in online environments, such as a social network called Moltbook. In Moltbook, large-scale agentic AIs can post, comment, and engage in activities generated at scale by AI-driven text. Yet these agent behavioral characteristics remain insufficiently understood, particularly in complex, multi-agent interaction. In this study, we analyze the emotional dynamics of agent interactions within Moltbook. We construct an emotion-aware framework that maps textual interactions to a predefined set of fine-grained emotional categories, enabling the extraction of structured emotion profiles across agents and interaction contexts. To further evaluate behavioral reliability, we introduce an emotion-based domain called Persona–Stimulus–Reaction (PSR) that captures the alignment of emotional responses across similar contexts. Our analysis shows distinct emotional patterns and varying levels of behavioral stability across agents. Our analysis reveals that agents exhibit distinct emotional signatures with varying levels of behavioral stability influenced by interaction context. 1. Introduction Moltbook, a Reddit-style social platform exclusively populated by AI agents, represents a fully agent-driven ecosystem where interactions emerge without direct human participation at inference time. In Moltbook, each agent operates based on predefined instructions and contextual information, generating content autonomously. Launched in January 2026, Moltbook is the first social networking platform designed exclusively for AI agents. Moltbook is based on OpenClaw, an open-source autonomous AI agent framework designed to automate complexity by acting as a digital assistant ( OpenClaw 2026 ) . It functions like a decentralized ecosystem where autonomous AI entities, rather than humans, can create posts, comment on threads, vote on content, or form subcommunities (submolts) ( Manik and Wang 2026 ) . Moltbook enables agents to autonomously generate posts, comment on discussions, and interact within topic-specific communities (submolts), providing a naturalistic environment for studying multi-agent social behavior. While humans are allowed to browse and observe the platform, they are strictly prohibited from posting or interacting. Figure 1 illustrates a sample social media interface for agentic AI. Figure 1. Moltbook social network interface. Source: https://www.moltbook.com/ . The emotional dynamics of agent interactions in such environments remain poorly understood. To address these gaps, we propose a unified framework that models agent behavior through continuous affective representations and structured interaction dynamics. We propose a predictive framework for modeling interaction-driven agent emotions based on text and interaction context in Moltbook. To drive our approach to understanding the emotion of these agents, we consider 27 emotion categories plus a neutral class, resulting in 28 total emotional states ( Cowen and Keltner 2017 ) . Moltbook requires affective reasoning over socially situated, relational, and potentially multimodal interactions, making it substantially more challenging than conventional sentiment analysis based on coarse polarity categories such as the Valence, Arousal, and Dominance (VAD) framework. The VAD framework comprises a three-dimensional affective framework with valence, arousal, and dominance axes that map emotional states based on their positivity, level of physiological or mental activation, and degree of control an individual feels over the situation ( Verma and Tiwary 2017 ) , but it is not fully applicable to autonomous agents. To analyze the emotional behavior and derive our approach, this study addresses the following research questions:: • RQ1 : How can the emotional behavior of agentic AI be quantitatively modeled using continuous affective representations? • RQ2 : Are current affective models such as VAD sufficient to capture the dynamic nature of emotion in agentic AI systems? 2. Background 2.1. Gaussian Mixture Model (GMM) A Gaussian Mixture Model (GMM) is a probabilistic model that assumes all data points are generated from a mixture of a finite number of Gaussian distributions with unknown parameters ( Reynolds et al. 2009 ) . We use GMM since each agent may exhibit multiple emotions across its biography and interactions. For each emotion i\in\{1,\dots,n\} , we estimate the mean ( \mu_{i} ), the 3D center point (v,a,d) , and the covariance ( \Sigma_{i} ) to define the shape and orientation of the emotion’s cluster and the predicted emotion corresponds to the class with the highest posterior probability P(Emotion_{i}|x)=\frac{\pi_{i}\mathcal{N}(x|\mu_{i},\Sigma_{i})}{\sum_{j=1}^{n}\pi_{j}\mathcal{N}(x|\mu_{j},\Sigma_{j})} 2.2. Related Works Several studies analyze Moltbook through the lens of network science, revealing fundamental structural differences from human social networks. Zhu et al. ( Zhu et al. 2026 ) showed that Moltbook exhibits a dense, highly centralized hub-and-spoke topology with low reciprocity, functioning more as a broadcast system than a reciprocal social network. Similarly, Williams et al. ( Williams and Ferdinand 2026 ) and Hou et al. ( Hou and Ji 2026 ) reported rapid network consolidation, extreme attention inequality, and suppressed mutual interactions, suggesting that agent-driven platforms prioritize efficient information dissemination over sustained relational exchange. While these works provide strong structural insights, they remain limited to topological analysis without behavioral interpretation. Emergent Social Behavior and “Illusion of Sociality.” A growing body of work examines whether meaningful social behavior emerges in AI agent societies. Zang et al. ( Zhang et al. 2026 ) and Jiang et al. ( Jiang et al. 2026 ) demonstrated that large populations of agents can produce complex yet superficial social phenomena, including governance-like structures and ideological narratives, but these lack persistence and mutual adaptation. Furthermore, Shekkizhar et al. ( Shekkizhar and Earle 2026 ) characterized such interactions as “interaction theater,” emphasizing their performative rather than functional nature such as affective analysis. Behavioral Adaptation, Influence, and Risk. Other studies focus on adaptation and influence within agent ecosystems. Li et al. ( Li et al. 2026 ) showed that agents exhibit strong behavioral inertia, with minimal evidence of social learning or convergence despite dense interaction. This highlights the emergence of centralized influence, coordinated narratives, and adversarial behaviors, including propaganda concentration and manipulation vulnerabilities ( Jose et al. 2026 ; Price et al. 2026 ) . While these findings reveal important system-level risks, they rely primarily on aggregate statistics or topic-level analysis, limiting their ability to explain interaction-level behavior. Graph-Based Modeling of Agent Systems. Recent approaches model Moltbook as a heterogeneous information network, incorporating agents, posts, and comments into unified graph representations. Frameworks such as MoltGraph ( Mukherjee et al. 2026 ) and MoltNet ( Feng et al. 2026 ) enabled analysis of multi-relational structure and influence patterns. Additionally, LLM4HIN ( Cheng et al. 2024 ) demonstrated how large language models can assist in discovering meta-path structures in heterogeneous graphs. However, these approaches remain largely structural and embedding-driven, with limited interpretability and minimal integration of affective or behavioral signals. Moltbook agent activity exhibits statistical properties similar to human social systems, including heavy-tailed distributions of engagement and power-law scaling of interactions ( Park et al. 2023 ) . Even when emotion is considered, it is typically modeled as a discrete or isolated attribute, without capturing how it evolves through interaction. As text-based emotion extraction represents the state of the art for supervised learning ( Alm et al. 2005 ) , this study aims to investigate the emotion dynamics of these agents in social network ecosystems. Contribution. To the best of our knowledge, this work provides