SocialLLM @ ICWSM 2026

Overview

Large Language Models (LLMs) are increasingly used not only as analytic tools, but as socially situated agents that reason, interact, and generate behavior in simulated environments. This shift enables new forms of computational social science, where LLM-driven agents are used to model decision-making, social norms, cooperation, persuasion, and collective dynamics at scale. The SocialLLM workshop invites submissions that explore how LLMs can serve as generative agents to simulate, analyze, and probe social behavior in online and networked contexts. We are particularly interested in work that connects micro-level language interactions to macro-level social phenomena central to ICWSM and computational social science. Beyond technical advances, we encourage contributions that critically examine when LLM-based social simulations are appropriate, what kinds of social processes they can meaningfully capture, and how their outputs should be evaluated and interpreted. The workshop aims to foster a principled, responsible, and empirically grounded research agenda for using LLMs in social reasoning and simulation.

Call for Papers

We welcome empirical, methodological, theoretical, and conceptual submissions. Topics include, but are not limited to:

Agent-Based Social Simulation with LLMs: Modeling coordination, conflict, polarization, cooperation, collective action, and information diffusion. Multi-agent strategic interaction and emergent collective behavior.
Emergence, Norm Formation, and Cultural Evolution: Development and evolution of norms, beliefs, narratives, and behaviors. Path dependence, tipping points, and long-term social dynamics.
Persona Fidelity and Population Heterogeneity: Psychologically plausible and evolving agent personas. Synthetic populations and demographic heterogeneity. Effects of individual-level variation on aggregate outcomes.
Platform and Network Contexts: Simulating online interaction under realistic exposure and network structures. Feedback loops, incentives, moderation, and governance mechanisms. Platform design and algorithmic mediation.
Interventions and Counterfactual Social Experiments: “What-if” simulations for policy, governance, and system design. Stress-testing interventions and identifying unintended consequences.
Grounding, Realism, and Privacy: Empirical grounding and calibration with real-world data. Reproducibility and evaluation frameworks. Privacy-preserving and responsible simulation practices.
Ethical and Societal Implications: Validity limits of LLM-based social simulations. Misuse risks and dual-use concerns. Applications in sensitive domains (elections, public health, vulnerable communities).

Submission Types

Long Paper (Archival or Non-Archival): May consist of up to 8 pages of content, plus unlimited pages for references and appendix.
Short Paper / Poster Papers (Archival or Non-Archival): May consist of up to 4 pages of content, plus unlimited references and appendix.
Abstract / Demo Papers (Non-Archival): May consist of up to 2 pages of content.

Submission Guidelines

Please check the style guidelines of ICWSM 2026. Accepted papers will be published at ICWSM workshop proceedings.

Submission Policies

The reviewing process will be double-blind.
At least one author of each accepted paper should register and present their work in SocialLLM.

Important Dates

Submissions open on OpenReview.

Note: Submitting authors must have an OpenReview profile. Co-authors are allowed to be added through name and email. New profiles created with an institutional email will be activated automatically. New profiles created without an institutional email will go through a moderation process that can take up to two weeks.

Submission Deadline	April 1, 2026, 11:59 PM AoE
Paper Notification	April 15, 2026, AoE
Camera-Ready Deadline	TBD
Workshop Date	May 26, 2026 (Half-Day Workshop)

Tentative Schedule

Time	Activity
00:00 – 00:10	Opening Remarks + Icebreaker
00:10 – 00:50	Keynote Talk 1
00:50 – 01:30	Keynote Talk 2
01:30 – 02:00	Coffee Break & Networking Session
02:00 – 03:00	Oral Session
03:00 – 03:55	Lightning Talks + Poster / Interactive Presentations
03:55 – 04:00	Closing Remarks

Keynote Speakers

Zhijing Jin

University of Toronto & MPI

Zhijing Jin (she/her) is an Assistant Professor in Computer Science at the University of Toronto, and also a Research Scientist at Max Planck Institute in Germany. She is a faculty member at the Vector Institute, a CIFAR AI Chair, an ELLIS advisor, and faculty affiliate at the Schwartz Reisman Institute in Toronto, CHAI at UC Berkeley, and the Future of Life Institute. She co-chairs the ACL Ethics Committee and the ACL Year-Round Mentorship. Her research focuses on Causal Reasoning with LLMs, and AI Safety in Multi-Agent LLMs. She has received the ELLIS PhD Award, three Rising Star awards, two Best Paper awards at NeurIPS 2024 Workshops, two PhD Fellowships, and a postdoc fellowship. She has authored over 100 papers and her work has been featured in CHIP Magazine, WIRED, and MIT News.

Testing and Improving LLM Cooperation via Multi-Agent Simulation

As AI systems take on more autonomous roles across the economy, governance, and daily life, they'll increasingly interact with each other. However, will the AI agents coordinate for social good, or exploit rival agents and people in ways that put humans at serious risk?

In this talk I will explain how we assess these dangers with large-scale social simulations and game-theoretic analysis. Across thousands of high-stakes scenarios, from arms race escalation to common pool resource depletion, frontier models choose socially beneficial actions in only 62% of cases, with systematic biases in framing and ordering worsening outcomes. Surprisingly, stronger reasoning capabilities often make models more prone to selfish strategies like free-riding, and recent models consistently defect in unmodified social dilemmas regardless of scale or reasoning ability. However, game-theoretic interventions offer a promising path forward: cooperation mechanisms such as mediation, enforceable contracts, and reputation systems improve collective welfare significantly and become more effective under stronger optimization pressures. Beyond formal mechanisms, self-organizing social structures like elected leadership oriented toward group welfare further sustain cooperation in sequential dilemmas. These results suggest that safer multi-agent AI requires principled institutional design rather than reliance on models' inherent prosociality.