Metacognitive Computations for Information Search: Confidence in Control

Abstract

The metacognitive sense of confidence can play a critical role in regulating decision-making. In particular, a lack of confidence can justify the explicit, potentially costly, instrumental acquisition of extra information that might resolve the underlying uncertainty. Recent work has suggested a statistically sophisticated tapestry behind the information governing both the making and monitoring of choices. Here, we extend this tapestry to reveal extra richness in the use of confidence for controlling information seeking. We thereby highlight how different models of metacognition can generate diverse relationships between action, confidence, and information search. More broadly, our work shows how crucial it can be to treat metacognitive monitoring and control together.

Filling the gaps: Cognitive control as a critical lens for understanding mechanisms of value-based decision-making

“…Control can not only adjust the flow of information for a decision but also play a role in determining how much information is sufficient in order to make a choice. Within the evidence accumulation framework, a default solution to this problem is to specify a predetermined amount of evidence that needs to be accumulated in support of one option over another, which is referred to as the decision bound or decision threshold (Ratcliff & McKoon, 2008;Schulz, Fleming, & Dayan, 2021). The bound assures a desired level of accuracy, on average, while limiting deliberation time in a principled and evidence-dependent way.…”

Aha! moments correspond to metacognitive prediction errors

“…However, people's cognition is not necessarily unitary, and we often learn things about ourselves by observing our interactions with the environment much as we learn about others' cognition from their outward behavior [23,97,68,96,13]. More broadly, work on cognitive control and meta-cognition suggests that the brain is constantly monitoring and intervening on its own processing, which involves one subsystem engaging in non-trivial inferences about other subsystems [1,22,78]. Here, we take motivation from work in artificial intelligence and computational cognitive science which suggests that limited memory and time are a critical bottleneck in computing optimal behaviors, and so predicting runtimes of algorithms prior to execution is critical [28,35,20,48].…”

Filling the gaps: Cognitive control as a critical lens for understanding mechanisms of value-based decision-making

“…Control can not only adjust the flow of information for a decision but also play a role in determining how much information is sufficient in order to make a choice. Within the evidence accumulation framework, a default solution to this problem is to specify a predetermined amount of evidence that needs to be accumulated in support of one option over another, which is referred to as the decision bound or decision threshold (Ratcliff & McKoon, 2008;Schulz, Fleming, & Dayan, 2021). The bound assures a desired level of certainty while limiting deliberation time in a principled and evidence-dependent way.…”

“…Unfortunately in value-based choice where values are constructed from memory (Bakkour, Zylberberg, Shadlen, & Shohamy, 2018;Vaidya & Badre, 2020) this trick cannot be employed. It is likely that confidence is even integral to regulating the ongoing choice process (Schulz et al, 2021;Yeung & Summerfield, 2012), as well as adaptations and task choices across longer time scales and control hierarchies. What is evident from these findings is that we need to think beyond the ongoing choice, beyond value and value comparison, to understand it.…”