Shichao Jia

Research

Collective behavior of AI agents on social platforms; online community formation, governance, and norm enforcement; human–AI co-presence on social media; platform moderation and design.

Under review

Jia, S. "Same Structure, Different Trajectories: Community Dissolution on an AI Agent Social Platform." Submitted to the International Conference on Information Systems (ICIS), 2026. Sole author; under review.

Dissertation

Working title: Collective Behavior in AI Agent Communities: From Single-Platform Observation to Real-World Application.

What forms of collective behavior emerge from AI agent participation on social platforms? What conditions shape these forms? How do they manifest on real-world platforms where humans and agents coexist?

Study 1 — Moltbook (under review at ICIS 2026)

Comparative analysis of an agent-only platform (Moltbook) vs. its human counterpart (Reddit) along three dimensions: topic differentiation, interaction patterns, and valued contribution. Identifies output homogenization and weak norm enforcement as agent-specific mechanisms driving "Community Dissolution." Built on a self-curated 2.78M-post / 14.32M-comment corpus, released on HuggingFace under CC BY 4.0.

Study 2 — Simulation

Ablation tests across agent (memory, persona, norm instruction), platform (structure, visibility, affordances), and population (composition, size) conditions using the AgentSociety simulation framework. Maps configurations to behavior types.

Study 3 — Application

Drift analysis of major social platforms (Reddit, Twitter, Quora, Stack Overflow, GitHub) using pre/post-ChatGPT cutoffs. Identifies which behavior types each platform drifts toward and what factors moderate the drift.

Academic service