Call for Papers 

Social Sim'27 invites papers on LLM-based social simulation. This third edition, held at AAAI-27, focuses on the attribution problem. When an LLM population exhibits a collective phenomenon, what caused it, and what evidence supports a claim about it?  

Submission Portal:  https://openreview.net/group?id=AAAI.org/2027/Workshop/Social_Sim
Interested to serve as a reviewer:  Sign-up form

Key Dates

• Submission Deadline: November 20, 2026, AoE

• Accept/Reject Notification: December 2, 2026, AoE

• Camera Ready Deadline: December 9, 2026, AoE

• Workshop Date: February 2027 (exact date TBD)

Topics  of interest include, but are not limited to: 

Axis 1: Population construction: what does the population represent?

• Data-grounded populations: calibration against behavioral and survey data; choice of human comparators

• Persona modeling: fidelity, consistency, and stability of agent personas

• Heterogeneity and bias: demographic distributions, distributional tails, and representational bias

• Human-AI hybrid simulations: combining human participants and generative agents

Axis 2: Collective dynamics: what emerges from interaction, and what causes it?

• Emergence versus recall: separating social dynamics from prompting artifacts, memorization, and data leakage

• Robustness: whether results hold across models, seeds, and environments

• Scaling and structure: population size, network topology, and interaction protocols

• Norms, diffusion, and polarization: how beliefs and behaviors form and spread

• Strategic behavior: repeated interaction, coordination, and collusion

• Methods from network science and game theory

Axis 3: Inference and consequence: what can be concluded?

• Validation: protocols, and which form of validity a result establishes

• Shared testbeds, replication, and negative results

• Simulation as pre-deployment evaluation; population-level monitoring of deployed agent societies

• Ethical, societal, and policy implications: manipulation, dual-use risks, and misrepresentation of identities

We also invite work that challenges the premise. This includes work on when an LLM society should not be trusted, and on when an artificial society is interesting even where it is not human-like. Papers must simulate a social environment. Single-agent work, agentic coding benchmarks, and multi-agent RL without a social interpretation are out of scope. 

Submission Types:

We invite three types of non-archival submissions:

• Regular Papers: Completed work, such as theoretical advances, empirical studies, or novel simulation methods. Up to 7 pages.

• Early-stage Papers: Preliminary findings, prototype systems, or position pieces. Up to 4 pages.

• Replication and Resources Papers: Simulation environments, population datasets, tools, or replication results, including negative ones. Systems built on Social Sandbox or similar frameworks are welcome here. Up to 4 pages plus artifacts.

References do not count toward page limits. All submissions must use the AAAI-27 Author Kit and be anonymized for double-blind review. Each submission receives at least two expert reviews. Accepted submissions will be presented as spotlights or posters and posted on the website.

Questions:

For questions regarding the call or the relevance of specific topics, please contact us at social-simulation@googlegroups.com or on the Slack.