Pain Expectations: the neural representations of predicted pain in self and others
Inferring another persons pain is a challenging computational problem: unlike self-related pain, individuals have no access to others nociceptive input and must instead rely only on indirect cues under substantial uncertainty. Using a predictive-processing framework, we modeled how expectation, uncertainty (variance-based risk), prediction error (PE) and surprise jointly shape pain-related decisions for oneself and for a stranger. During fMRI, participants received cues signaling the intensity and probability of possible painful events, then made decisions to mitigate the anticipated pain via a lottery and a wager. Behaviorally, participants were more risk-averse toward others pain than their own, prioritizing pain reduction over economic gain. Neurally, anticipatory risk signals in anterior insula (AI) and dorsal striatum were selectively amplified when pain concerned another person, as opposed to self-pain. By contrast, PE and surprise were reliably represented in AIns for both targets following pain delivery. However, representational similarity analysis suggest that error-signal in AIns is represented through reliable, but dissociated, patterns for self and others. Together, these findings show that expectation, uncertainty and error processing are differentially affected by the identity of the pain recipient, and underscore how models of prediction under uncertainty are powerful tools for explaining how we assess pain in other people. Significance StatementWhen we make decisions about another persons pain, we cannot directly access what they feel: we must infer it under uncertainty. Through a rigorous computational framework of prediction under uncertainty, we assessed the joint role played by expectancy, uncertainty and error-signal to model peoples decisions about their own and others pain. Deciding on anothers behalf increased brain and behavioral sensitivity to uncertainty; likewise error signals following a painful event were represented in a specific code, distinct from self-related pain. These findings offer a computational account of how humans navigate the inherent uncertainty of inferring others internal states, with direct relevance to clinical settings where surrogate decision-makers must estimate a patients pain despite lacking direct access to it.