Agent-based models (ABMs) offer a conceptual framework for simulating the actions and interactions of autonomous agents to understand the behavior of a system and investigate emergent phenomena. ABMs have a long history of application for social simulation, ranging from the modelling of the spread of an infectious disease to opinion polarization. LLMs offer a powerful alternative to traditional ABMs due to their strong text generation capabilities, allowing for more realistic, rich, and detailed simulations, leading to the emergence of the novel field of Generative Agent-Based Modelling (GABM).

This Dagstuhl Seminar convenes the leading research groups that advance GABM methodology or use GABMs in cutting-edge applications to develop standards, exchange best practices, and foster collaborations. Core topics include:

1. Validation & uncertainty quantification: Development of standardized benchmarks and uncertainty quantification approaches that are currently lacking for GABM research.
2. From chatbots to in-silico humans: Organize a coalition to develop purpose-built foundation models specifically engineered for high-quality social simulation research, overcoming limitations of LLMs that are designed for chat-based tasks.
3. Reproducibility and scalability: Establish standardized reporting requirements for reproducible GABM research and explore the potential for small, purpose-built models for GABM research to address the scalability bottleneck.
4. Promising current and future application areas: Convene researchers from currently disjoint GABM research areas to highlight success stories and identify promising future applications.
5. Best practices and tools for GABM research: Facilitate knowledge transfer about best practices and tools established in different GABM application areas.

By bringing together researchers that are involved in GABM development and validation, different GABM application areas, and the development of foundation models, this seminar aims to (1) consolidate the emerging approaches and standards on validation, benchmarking, and uncertainty quantification in the field, (2) compile Open Source tools and best practices for GABM research, (3) survey promising application areas of GABMs, and (4) foster a coalition of researchers willing to drive the development of efficient foundation models for social simulation.

Creative Commons BY 4.0

Jana Lasser

This seminar qualifies for Dagstuhl's LZI Junior Researchers program. Schloss Dagstuhl wishes to enable the participation of junior scientists with a specialisation fitting for this Dagstuhl Seminar, even if they are not on the radar of the organizers. Applications by outstanding junior scientists are possible until October 2, 2026.