Dynamically learning the meaning of deceptive cues about pain

Beliefs influence the intensity of felt pain. We previously applied regression models to two independent pain-cueing datasets (https://osf.io/5r6z7/) and found that pain intensity reports were a function of the error between participants’ belief, as determined by cues that informed participants about the level of forthcoming noxious stimulation, and the actual stimulation intensity, such that greater error decreased the influence of prior beliefs. Although this result appeared to present a challenge to established models of perception, our analysis did not formally compare the result against established computational models. This made it difficult to assess the validity of our results against the wider literature and we could not explain why the effect of information cues changed over the course of the task. In the current study (https://osf.io/fj27k/) we compared a model that corresponded to our original interpretation where pain ratings were considered a function of prediction error size, against Bayesian reinforcement learning models. Pain ratings were best explained by a model in which the expected value of each information cue was updated via a Bayesian reinforcement learning algorithm. These new results indicate that deception pain studies should not assume that participants’ belief about the information they are given is fixed. Rather, our results provide evidence that the meaning of cues change dynamically over the course of the session through trial-by-trial updates, even when participants are instructed otherwise.