Confirmation bias is adaptive when coupled with efficient metacognition

Abstract

Selective consideration of information is a prominent feature of human behaviour, and recent studies have identified proneness to confirmation bias as a cognitive feature underlying dogmatic beliefs. While such altered information processing typically leads to detrimental performance in laboratory tasks, the ubiquitous nature of confirmation bias makes it unlikely that selective information processing is universally detrimental. Here we suggest that confirmation bias is adaptive to the extent that agents have good metacognition, and are able to seek information when they realise they are wrong. Using simulation-based modelling, we explore how the adaptiveness of holding a confirmation bias depends on such metacognitive insight. We find that the behavioural consequences of selective information processing are highly dependent on agents’ introspective abilities. Strikingly, selective information processing can even improve decision-making when compared to unbiased evidence accumulation, as long as it is accompanied by good metacognition. These results further suggest that interventions which boost people’s metacognition might be efficient in alleviating the negative effects of selective information processing.

Insight into the accuracy of COVID-19 beliefs predicts behavior during the pandemic

“…Rather, the infodemic is worrisome because it produces informational noise that reduces the reliability of evidence. When citizens form confidence judgments by accumulating lowreliability evidence, this necessarily decreases metacognitive efficiency, that is, the sensitivity with which confidence distinguishes between true and false belief (Rollwage & Fleming, 2021). In line with a mechanism whereby the reliability of evidence determines the accuracy of confidence judgments, it was found that citizens' metacognitive accuracy is lower for the politicized science of climate change, compared to non-politicized science (Fischer et al, 2019).…”

“…There is considerable research on how human metacognition, our ability to reflect upon, and evaluate own beliefs, enables us to avoid making decisions based on unreliable evidence in the absence of external feedback (Hainguerlot et al, 2018;Schulz et al, 2020;Yeung & Summerfield, 2012). Metacgonitive reflection expresses itself in confidence, an awareness of the validity and fallibility of our beliefs, which can be used as an internal control signal to guide behavior (Balsdon et al, 2020;Desender et al, 2018;Rollwage & Fleming, 2021). Ideally, then, citizens' confidence should have high sensitivity in that it distinguishes correct from incorrect beliefs.…”

Metacognitive Computations for Information Search: Confidence in Control

“…Such a confirmation bias can be straightforwardly modelled within our framework and might might assist in explaining behaviour . Recent simulation work (Rollwage & Fleming, 2021) has shown that this an apparent confidenceinduced confirmation bias can in fact be adaptive when an agent posseses second-order metacognitive hypersensitivity. Notably, Rollwage and Fleming (2021) used a different information flow for the final decision.…”

“…Recent simulation work (Rollwage & Fleming, 2021) has shown that this an apparent confidenceinduced confirmation bias can in fact be adaptive when an agent posseses second-order metacognitive hypersensitivity. Notably, Rollwage and Fleming (2021) used a different information flow for the final decision. However, this still raises interesting questions about what constitutes optimality in both the passive and active sampling of information.…”

SalemGarcia_2021

“…In other terms, learning rates analysis in the context of simple reinforcement learning suggest that subjects behave as if they were neglecting outcomes that disconfirm their current choice, which inevitably leads to an overestimation of the value of the currently chosen option and therefore the probability of being correct. Although recent simulation studies suggests that this confirmation bias may be adaptative in some situations (Lefebvre et al, 2020;Rollwage and Fleming, 2021), it could be detrimental, contributing to inaccurate estimates of the average reward of options, and hindering performancefor instance in high reward environments (Cazé and van der Meer, 2013).…”