The computational relationship between reinforcement learning, social inference, and paranoia

Abstract

Theoretical accounts suggest heightened uncertainty about the state of the world underpin aberrant belief updates, which in turn increase the risk of developing a persecutory delusion. However, this raises the question as to how an agent’s uncertainty may relate to the precise phenomenology of paranoia, as opposed to other qualitatively different forms of belief. We tested whether the same population (n=750) responded similarly to non-social and social contingency changes in a probabilistic reversal learning task and a modified serial Dictator game, and the influence of paranoia on both. We fitted computational models that included closely related parameters that quantified the rigidity across contingency reversals, and the uncertainty of individuals about the environment/their partner. Consistent with prior work we show that, after reversals, pre-existing paranoia was independently associated with uncertainty around a partner’s behavioural policy and rigidity in harmful intent attributions in the social task, and in the probabilistic task, larger decision temperatures and commitment to suboptimal cards. We show relationships between decision temperature in the non-social task and priors over harmful intent attributions and uncertainty over beliefs about partners in the social task. Our results converge across both classes of model, suggesting paranoia is associated with a general uncertainty over the state of the world (and agents within it) that takes longer to resolve, although we demonstrate that this uncertainty is expressed asymmetrically in social contexts. Our model and data allow clarification of cognitive mechanisms that explain persecutory delusions and provide testable, phenomenologically relevant predictions for causal experiments.

Assumed agreement about paranoid conspiracies in social networks protects against distress

“…But our data continue to suggest that paranoia and conspiracy theorizing -even the social aspects thereofare related to low-level domain-general primary learning mechanisms (30) which may be dopaminergically and noradrenergically mediated (31,32). This domain general contribution to paranoia involves learning expectancies about the environment and the agents within it (33), a process that may be guided by using oneself as a source of prior beliefs (18). Our data represent an extension of those processes into real world social relationships.…”

Rapid classification vs. refined learning about harm intent: Roles of serotonin and racial bias

“…Belief-based models have performed well in attribution tasks, over and above associative models (3,4), but the initial characterization of others appears faster than most models suggest (5,6). We took inspiration from recent work on black-and-white thinking (7) to hypothesize that in interpersonal situations, people attempt to quickly classify others' attributes as 'beneficial or detrimental' to themselves.…”

Aberrant perception of environmental volatility during social learning in emerging psychosis

“…Therefore, this mechanism may not be speciftied to social contexts, but instead may be related to a more general deficit in learning under uncertainty (Reed et al, 2020;Suthaharan et al, 2021). However, we do note that the social control task employed by Suthaharan et al (2021) was not as ecologically valid as other tasks that were used to study paranoia such as the dictator game (Raihani and Bell, 2017;Barnby et al, 2020Barnby et al, , 2022 or our task which was adapted from empirically-observed human-human interactions in a previous study using videos of human advisers intending to either help or deceive players (Diaconescu et al, 2014). Finally, it is also possible that there are both domain-general and domainspecific changes, but that these can only be studied at the neuronal level and converge on the same behavioural model parameters.…”