18th International Conference on Social Computing, Behavioral-Cultural Modeling, & Prediction and Behavior Representation in Modeling and Simulation
October 15-17, 2025, Hybrid.
Carnegie Mellon University, Pittsburgh, PA 15213


Keynote Speaker(s)


Keynote 1: October 15, 2025, 16:40 - 18:00, In-Person

Lynn Smith-Lovin, Ph.D., Robert L. Wilson Distinguished Professor of Sociology, Duke University

Title: ACT meets AI: How Human Participants and Theoretical Predictions Compare to LLM and NLP

Abstract: Affect control theory (or ACT) is a formal, mathematical theory of how cultural meanings for categories of people and types of behavior can predict interpersonal actions, emotional reactions to them, and cognitive relabeling. The simulation program representing the theory typically uses measures from human participants for two core inputs: (1) the culturally shared meanings of labels for identities, actions, emotions, traits, settings and other socially meaningful aspects of social interaction, and (2) the functional forms that predict how social events will modify those meanings within a situation to create impressions of particular people and actions. The last decade has seen efforts to use NLP and (more recently) LLMs to estimate the cultural meanings of identities, actions and modifiers. Recent work is beginning to assess how well such AI tools can estimate impression formation processes. In addition, we are now looking at how the predictions about future actions and labelling of actors compare between ACT and the LLM models. It appears that computational methods trained on large bodies of text can reproduce human subjects’ ratings of evaluation (good-bad), potency (powerful-powerless) and activity (lively-quiet) quite well. However, they have a Western (U.S.) bias, higher variance and higher bimodality than human ratings. Impression change comparisons are at a much earlier stage, but adding context of an Actor-Behavior-Object event seems to produce lower correlations with human ratings. Finally, comparisons of ACT and AI approaches indicate that LLMs don’t do well at predicting future behaviors of actors or objects of actions. LLMs are more likely to predict continuity of meaning, emotion and action than ACT. ACT is more context-sensitive. While ACT has been validated by experiments on human participants, new studies are needed to assess the relative usefulness of the sensitivity to social context that ACT offers. Applications in combination with NLP to represent narrative character development also offer promise for a wide range of applications, but need validation of some form.

Bio: Lynn Smith-Lovin is Robert L. Wilson Professor of Arts and Sciences in the Department of Sociology at Duke University, with additional appointments in Psychology and Neuroscience. Her research examines the relationships among identity, action and emotion. She works within the affect control theory tradition, which examines how cultural information is imported into local social interactions, and leads to behavioral, cognitive and emotional responses. The theory is represented in the simulation programs INTERACT and BayesACT. Her current projects involve (1) experimental studies of how unexpected interpersonal events are relabeled, and (2) validation assessments of the simulation model, and (3) an overview book on affect control theory which should be out in 2027. She has published two books, and numerous articles in American Sociological Review, American Journal of Sociology, Annual Review of Sociology and Social Psychology Quarterly, among others.

Smith-Lovin has received the James S. Coleman Award from the American Sociological Association section on Mathematical Sociology, the Cooley-Mead Award from the ASA section on Social Psychology, and Lifetime Achievement Awards from the ASA sections on the Sociology of Emotions and on Altruism, Morality and Social Solidarity,. She has co-edited Social Psychology Quarterly, and has served as President of the Southern Sociological Society, Vice-President of the ASA, and Chair of the ASA Sections on the Sociology of Emotion, on Social Psychology and on Mathematical Sociology.


Keynote 2: October 16, 2025, 16:00 - 17:20, In-Person

Robert H. Thomson, Ph.D., Research Scientist, Department of Psychology, Carnegie Mellon University

Title: Re-Analyzing Behavioral Theories with a Cognitive Lens

Abstract: In this talk I will review several behavioral and computational modeling studies to describe how cognitive models can be used to provide insights into the cognitive operations underlying prominent social and behavioral theories. I will explain how common cognitive biases and features of persuasive communication emerge as natural properties of memory dynamics, and how these influences may be mitigated at times using explicit strategy selection. Implications for training and education, future research, and designing successful interventions will be discussed.

Bio: Robert Thomson is an award-winning cognitive scientist, cybersecurity researcher, and served as an Associate Professor at the United States Military Academy from 2016-2025 and also as a Principal Research Scientist at the Army Cyber Institute. He is currently a Research Scientist in the Department of Psychology at Carnegie Mellon University and sits on the Board of the Cognitive Security Institute. With over 14 years of postdoctoral experience, Robert specializes in cognitive modeling, behavior prediction, AI-enabled decision support, and human-machine teaming. His research has directly supported national security initiatives funded by DARPA, IARPA, Army Futures Command, the Office of Naval Research, and the NIH. In addition, he has sat on three DARPA Information Science & Technology Study Groups and has served as Subject Matter Expert and reviewer for five DARPA and IARPA efforts. Recognized for excellence in both research and teaching, Robert has earned numerous honors, including several best paper awards, most recently for his work on belief-bias and resilient decision-making; the Army Commander's Award for Civilian Service (2019) for excellence in teaching, and the Army Achievement Medal for Civilian Service (2020) for excellence in mentoring cadets. He also received a DARPA Service Commendation as a performer and for facilitating the DARPA XAI Program Exchange, and was recognized with the SBP-BRiMS Annual Service Award (2018) for leadership in conference organization. His expertise bridges cognitive science and applied cyber- and cognitive security, advancing mission-critical solutions that enhance decision-making and resilience in complex, high-risk environments.