PROJECT SUMMARY (See instructions): Virtually every aspect of human behavior is governed by beliefs about ourselves and the world around us. These beliefs are continuously updated in response to the deluge of information we encounter in everyday life. The mechanisms that underlie belief updating may hold the key to understanding neuropsychiatric conditions involving abnormal belief updating, such as delusions in psychosis or cognitive distortions in depression. Despite active research, progress in our understanding of the neural mechanisms of normal and abnormal belief updating has been limited, owing in part to i) a focus on individual tasks rather than cross-domain constructs, ii) oversimplified models that cannot scale to real world behavior, and iii) a failure to link computational descriptions to the neural circuitry that performs them. Here, we propose to bridge these gaps by implementing our recently developed computational framework of belief updating across tasks from different neuroscience domains. We recently established a belief updating framework based on contextual inference, which embraces the complexity of real-world belief updating, and can, in principle, be applied broadly to tasks of different neuroscience domains. Contextual inference assumes that belief updating includes two components: 1) learning what to expect in a given context, and 2) figuring out which context you are actually in. We have developed a biological instantiation of this model, based on cortico-striato-thalamic loops that can perform this contextual inference, capture belief updating behavior in canonical tasks, and shed light on pathological belief updating behaviors observed in schizophrenia. In our model, individual and clinical differences in belief updating could emerge through differences in the stability of context representations, which are stored in cortical attractor networks that are sensitive to both stable structural features, such as the ratio of excitation to inhibition within the network, as well as dynamic interactions with thalamic inputs. Applying lessons from our model to schizophrenia, where prefrontal inhibition is impaired, cortical attractors would be expected to be less stable, promoting spurious attractor switches and state transitions. We will extend our contextual inference model to measure the degree to which it can generalize behavioral tendencies of belief updating across neuroscience domains. We will examine how these tendencies relate to real-world behavior and mental health traits. We will then test the model’s core predictions regarding the biological origins of