Large language models (LLMs) are being used as agents in behavioral science and agent-based modeling. In this workshop, we would like to discuss the opportunities and challenges of LLMs for modeling human behavior. This workshop will bring together scientists from psychology, sociology, computational social science, computer science, and applied mathematics, to discuss the following questions:

1. Which kinds of individual behaviors and collective phenomena can be modeled using LLMs, and how can LLMs be combined with models of different aspects of cognition (memory, emotions, beliefs, decisions, perception) to study different individual and collective phenomena in online environments and beyond?
2. What methodological challenges need to be overcome, for example, biases in representativeness of agents stemming from incomplete training samples that do not include some parts of the population and some aspects of human behavior?
3. How can LLMs be used beyond studying collective behavior on social media, to better understand human cognition and sociality more generally?

The workshop would take place over 3 days. During the first two days, everyone will give a 10-15 min talk on 1) their current experience in using LLMs, and 2) a new challenge/perplexing problem related to the use of LLMs in behavioral science. After each pair of related talks, there will be a discussion. Following the talks, there will be group discussions about several of the proposed problems. On the last day, everyone will be invited to submit a couple of paragraphs reflecting on the problems discussed. The organizers will turn those paragraphs into a white paper, which will serve as a background for a peer-reviewed publication and/or further collaborations.