Emergent social transmission of model-based representations without inference
Speaker: Silja Keßler
Abstract: “How do people acquire rich, flexible knowledge about their environment from others despite limited cognitive capacity? Humans are often thought to rely on computationally costly mentalizing, such as inferring others’ beliefs. In contrast, cultural evolution emphasizes that behavioral transmission can be supported by simple social cues. Using reinforcement learning simulations, we show how minimal social learning can indirectly transmit higher-level representations. We simulate a naïve agent searching for rewards in a reconfigurable environment, learning either alone or by observing an expert---crucially, without inferring mental states. Instead, the learner heuristically selects actions or boosts value representations based on observed actions.
Our results demonstrate that these cues bias the learner's experience, causing its representation to converge toward the expert’s. Model-based learners benefit most from social exposure, showing faster learning and more expert-like representations. These findings show how cultural transmission can arise from simple, non-mentalizing processes exploiting asocial learning mechanisms.”

Modeling Human Pouring Behavior using System Identification and Optimal Control Methods
Speaker: Niteesh Midlagajni M. Sc. (Psychology of Information Processing)
Abstract: Optimal control methods under uncertainty have been proven effective in modeling a variety of human sensorimotor behaviors. However, these algorithms require the specification of underlying dynamics. In many naturalistic, sequential tasks, deriving these dynamics from first principles is non-trivial. Here, we propose using system identification methods to learn the dynamics directly from behavioral data, and then leveraging optimal control models to understand the task. We focus on the task of pouring: a highly practiced, everyday activity involving continuous visuomotor control. We designed a novel experimental setup combining mobile eye tracking, object tracking, and a custom-made digital scale, and collected data of participants performing the pouring task under two conditions: self-paced, where they poured naturally, and fast, where they were instructed to pour as quickly as possible. Participants were given no explicit instructions on the target fill level or time constraints in either condition. Using Sparse Identification of Nonlinear Dynamics (SINDy), we extract a low-dimensional dynamical model from the data and apply iterative Linear Quadratic Gaussian (iLQG) control to infer the cost function governing the behavior. The resulting cost function is highly interpretable: the pouring task reflects a trade-off between reaching a preferred fill level, while minimizing effort, and regulating flow to avoid spillage. Taken together, this approach offers an interpretable framework for analyzing naturalistic behavior using tools from optimal control theory.

Joint Learning in the Cooperative Game Hanabi
Speaker: Dominik Magiera (Models of Higher Cognition)
Abstract: Human cooperation often requires jointly learning a task while simultaneously establishing shared conventions. We study this process using a simplified version of Hanabi, a cooperative game that captures core features of real-world collaboration under partial observability and constrained communication. Thirty participants played over 300 repeated games, either with another human or with a rule-based agent, while providing think-aloud verbalizations. Performance improved steadily but peaked only late in the session, after 40 minutes on average. Participants learned to use limited communication more efficiently, requiring fewer explicit hints to support play under partial knowledge. This indicates the emergence of effective coordination and Theory-of-Mind–based reasoning. Subjective experience was analyzed using think-aloud reports with automated transcription and LLM-based sentiment analysis. Overall, criticism was more directed to oneself than to the partner, and increased with task understanding.