Mental Models in the Brain: Representing Salience and Surprise
James Danckert
University of Waterloo

We are incapable of accurately representing all sensory elements of our noisy and chaotic environment. To deal with these challenges, humans make use of mental models to represent the key components of the environment as accurately as possible. To be effective, mental models need at least two key components: First, they must accurately capture regularities and contingencies in the environment – that is, they have to determine, what is salient? Second, they need to be vigilant to change – determining when new information fails to conform to predictions generated by the model – when is current evidence is surprising relative to the model’s predictions? I will present evidence to suggest that the ability to accurately represent environmental regularities and to update those representations in the face of change is supported by a predominantly right hemisphere network of brain regions including the anterior insular, medial frontal cortex, and the inferior parietal lobule. Using a range of statistical learning paradigms, I will demonstrate that right brain damage (RBD) patients fail to accurately represent regularities and, moreover, fail to adapt to changes in regularities. I will claim that such mental model updating deficits in RBD patients are generic. In two fMRI studies, we showed that updating a model of an opponent’s bias in “rock, paper, scissors” or changing a perceptual report of an ambiguous figure activated a highly similar network of brain regions including the anterior insular, medial frontal cortex, and inferior parietal lobule. We hypothesize specific roles for these regions such that the insular maintains the current model, the medial frontal cortex determines when to explore options/hypotheses to developing a new model, and the inferior parietal cortex represents salience and surprise. One challenge in evaluating mental models lies in determining what model a participant has before coming into the lab. I will report data from a novel task (Plinko) we developed that allows us to flexibly assess a participant’s prior expectations, as well as determining their efficiency in building and updating mental models. Pilot data from RBD patients on this task confirms that they are impaired in both processes. This body of work represents a novel conceptualization of right hemisphere brain function for which there is growing supporting evidence.