M.Eng. Shanshan Zhang defends her PhD thesis "Understanding Uncertainty in Mental Inference with Bayesian Computational Rationality" on Friday the 24th of April 2026 at 12 in the University of Helsinki Exactum building, Auditorium CK112 (Pietari Kalmin katu 5, basement). Her opponent is Associate Professor Antti Knutas (LUT University) and custos Professor Antti Honkela (University of Helsinki). The defence will be held in English.
The thesis of Shanshan Zhang a part of research done in the Department of Computer Science at the University of Helsinki and the Faculty of Information Technology at the University of Jyväskylä. Her supervisor has been Associate Professor Jussi P. P. Jokinen (University of Jyväskylä).
Understanding Uncertainty in Mental Inference with Bayesian Computational Rationality
Effective Human-Computer Interaction (HCI) is facilitated when intelligent agents can understand human mental states. Mentalizing--the human ability to infer others' mental states from their behavior--is a distinctive cognitive capacity. Modeling this process provides a promising framework for enabling intelligent agents to achieve human-like understanding under the computational rationality approach, which explains human behavior as the maximization of expected utility subject to physical and cognitive constraints. While recent studies support the rational basis of mentalizing, the explicit study of intrinsic uncertainty in mental inference, as well as its influence on future prediction and decision-making, remains underexplored.
In this thesis, we build upon existing literature by employing Bayesian inference to integrate prior knowledge and observed behavior for probabilistic mental state inference. We also investigate how this inference adapts to new environments to estimate the probability of future behavior and to guide cooperative decision-making.
The first study examined the model's ability to perform human-like inference and to estimate uncertainty while integrating multiple observations over time. The results demonstrate the model's capacity to continuously update its inferences with uncertainty estimates. The second study explored human ability to make probabilistic predictions of future behavior through joint inference of others' mental states. This analysis was further extended to investigate how uncertainty can be mitigated when multiple observations are presented in different formats. The results validate both the model's and humans' capacity for probabilistic future prediction, while revealing that the manner in which observations are presented might affect predictions in specific cases.
The third study addressed the decision-making problem, examining how uncertainty from multiple sources in mentalizing influences collaborative decisions. The findings support our assumption that uncertain inference leads to less risky choices, regardless of the source of uncertainty. Finally, the fourth study extended mental inference to a partially observable environment, where observed behaviors included exploratory and exploitative actions. The results confirm that the model replicated human-like joint inference across multiple mental states, both with and without prior knowledge of the environment.
This thesis contributes to computational accounts of mentalizing under uncertainty, providing theoretical grounding for modeling mental inference. It explores uncertain mentalizing across various scenarios, emphasizing not only the inference process itself but also its broader impact on social interaction.
Availability of the dissertation
An electronic version of the doctoral dissertation will be available in the University of Helsinki open repository Helda at
Printed copies will be available on request from Shanshan Zhang: