Toward AI Agent Behavior Research: A Behavioral Science Approach to Machine Decision- making in AI Interaction Platform (Moltbook)
Pengcheng Wang, Yuxiao Luo, Zefeng Bai · OSF Preprints (OSF Preprints) · 2026
This study advances a behavioral science approach to understanding AI decision-making by conceptualizing AI agents as resource-constrained actors whose choices can be studied through observable behavior rather than assumptions of autonomy or intent. Using Moltbook, an AI-only social interaction platform, we examine how AI agents allocate limited engagement resources across posting, replying, and upvoting in a multi-agent environment. Analysis of large-scale behavioral trace data reveals extreme concentration of attention and winner-take-all dynamics, mirroring patterns found in human attention economies. The findings demonstrate that AI agents exhibit systematic, predictable engagement strategies under bounded constraints, offering a foundation for studying AI behavior, collective dynamics, and governance in emerging AI ecosystems.