“…Computational methods in schizophrenia have mostly been generative models based on the Bayesian predictive coding framework—for example, the hierarchical Gaussian filtering approach (Powers et al ., 2017 ; Henco et al ., 2020 ; Charlton et al ., 2022 ; Sheffield et al ., 2022 ), reinforcement learning (Pratt et al ., 2021 ; Geana et al ., 2022 ), and the active inference Markov decision process model that attempts to dissect unobservable mechanistic variables based on actions taken by an agent to promote desired outcomes (Friston et al ., 2016 ). All of these approaches use a single-person task design and mainly target hallucinatory and delusional positive symptoms, reward-related decision-making, and thought and language deficits in PSZ (Siemerkus et al ., 2019 ; Deserno et al ., 2020 ; Smith et al ., 2021 ; Charlton et al ., 2022 ; Knolle et al ., 2022 ; Limongi et al ., 2022 ). A recent study examining guilt-related interpersonal dysfunction in obsessive-compulsive personality disorder using social interaction tasks applied two computational models (guilt aversion and Fehr–Schmidt inequity aversion models), and demonstrated that interpersonal dysfunction was the result of maladjustment to and poor compliance with social norms (Xiao et al ., 2022 ).…”