Mesolimbic confidence signals guide perceptual learning in the absence of external feedback

Optimal metacognitive decision strategies in signal detection theory

Signal detection theory (SDT) has long provided the field of psychology with a simple but powerful model of how observers make decisions under uncertainty. SDT can distinguish sensitivity from response bias and characterize optimal decision strategies. Whereas classical SDT pertains to "type 1" judgments about the world, recent work has extended SDT to quantify sensitivity for metacognitive or "type 2" judgments about one's own type 1 processing, e.g. confidence ratings. Here we further advance the application of SDT to the study of metacognition by providing a formal account of normative metacognitive decision strategies - i.e., type 2 (confidence) criterion setting - for ideal observers. Optimality is always defined relative to a given objective. We use SDT to derive formulae for optimal type 2 criteria under four distinct objectives: maximizing type 2 accuracy, maximizing type 2 reward, calibrating confidence to accuracy, and maximizing the difference between type 2 hit rate and false alarm rate. Where applicable, we consider these optimization contexts alongside their type 1 counterparts (e.g. maximizing type 1 accuracy) to deepen understanding. We examine the different strategies implied by these formulae and further consider how optimal type 2 criterion setting differs when metacognitive sensitivity deviates from SDT expectation. The theoretical framework provided here can be used to better understand the metacognitive decision strategies of real observers. Possible applications include characterizing observers' spontaneously chosen metacognitive decision strategies, assessing their ability to fine-tune metacognitive decision strategies to optimize a given outcome when instructed, determining over- or under-confidence relative to an optimal standard, and more. This framework opens new avenues for enriching our understanding of metacognition.

Empirically Integrating the Evidence for Different Predictive Coding Components Using Auditory False Perception

Perception is a probabilistic estimation of the sensory information we receive at any given time and is shaped by an internal model generated by the brain by assimilating information over the life course. This predictive system in the brain has several components-(i) the internal model, (ii) the model-based prediction called priors, (iii) the weighted difference between the prior and sensory input called prediction error (PE) and (iv) the weighted sum of the prior and input called perceptual inference. Until now, different studies have explored the independent components of this predictive coding system, and we, for the first time to our knowledge, integrate them. To do this, we induce a conditioned hallucination (CH) illusion by means of a multisensory integration paradigm and use this as a model to study the behavioral and electrophysiological responses to this experience. Additionally, we also probe their predictive coding system using a well-established local-global auditory oddball paradigm. By comparing the behavioral and electrophysiological components of people more and less likely to perceive an illusion in the two paradigms, we observed that high perceivers place more confidence in their internal model and low perceivers in the sensory information. Furthermore, high perceivers were more sensitive than low perceivers to PEs that were generated by a change in the context of the sensory information, which served as a measure of a change in the internal model itself. As an exploratory analysis, we also observed that the objective likelihood of perceiving an illusion was corrected to the self-reported likelihood of perceiving an illusion in a day-to-day setting, which disappears when controlled for the perceptual threshold. These results taken together start to give us an idea as to how a person's innate bias-either towards a learned model or external information may-affect their perception in a sensory context.

Perceptual learning improves discrimination but does not reduce distortions in appearance

Human perceptual sensitivity often improves with training, a phenomenon known as "perceptual learning." Another important perceptual dimension is appearance, the subjective sense of stimulus magnitude. Are training-induced improvements in sensitivity accompanied by more accurate appearance? Here, we examined this question by measuring both discrimination (sensitivity) and estimation (appearance) responses to near-horizontal motion directions, which are known to be repulsed away from horizontal. Participants performed discrimination and estimation tasks before and after training in either the discrimination or the estimation task or none (control group). Human observers who trained in either discrimination or estimation exhibited improvements in discrimination accuracy, but estimation repulsion did not decrease; instead, it either persisted or increased. Hence, distortions in perception can be exacerbated after perceptual learning. We developed a computational observer model in which perceptual learning arises from increases in the precision of underlying neural representations, which explains this counterintuitive finding. For each observer, the fitted model accounted for discrimination performance, the distribution of estimates, and their changes with training. Our empirical findings and modeling suggest that learning enhances distinctions between categories, a potentially important aspect of real-world perception and perceptual learning.

Information-Theoretic Measures of Metacognitive Efficiency: Empirical Validation with the Face Matching Task

The ability of participants to monitor the correctness of their own decisions by rating their confidence is a form of metacognition. This introspective act is crucial for many aspects of cognition, including perception, memory, learning, emotion regulation, and social interaction. Researchers assess the quality of confidence ratings according to bias, sensitivity, and efficiency. To do so, they deploy quantities such as meta-d'-d' or the M-ratio These measures compute the expected accuracy level of performance in the primary task (Type 1) from the secondary confidence rating task (Type 2). However, these measures have several limitations. For example, they are based on unwarranted parametric assumptions, and they fall short of accommodating the granularity of confidence ratings. Two recent papers by Dayan and by Fitousi have proposed information-theoretic measures of metacognitive efficiency that can address some of these problems. Dayan suggested meta-I and Fitousi proposed meta-U, meta-KL, and meta-J. These authors demonstrated the convergence of their measures on the notion of metacognitive efficiency using simulations, but did not apply their measures to real empirical data. The present study set to test the construct validity of these measures in a concrete behavioral task-the face-matching task. The results supported the viability of these novel indexes of metacognitive efficiency, and provide substantial empirical evidence for their convergence. The results also adduce considerable evidence that participants in the face-matching task acquire valuable metaknowledge about the correctness of their own decisions in the task.

Neural mechanisms of metacognitive improvement under speed pressure

The ability to accurately monitor the quality of one's choices, or metacognition, improves under speed pressure, possibly due to changes in post-decisional evidence processing. Here, we investigate the neural processes that regulate decision-making and metacognition under speed pressure using time-resolved analyses of brain activity recorded using electroencephalography. Participants performed a motion discrimination task under short and long response deadlines and provided a metacognitive rating following each response. Behaviourally, participants were faster, less accurate, and showed superior metacognition with short deadlines. These effects were accompanied by a larger centro-parietal positivity (CPP), a neural correlate of evidence accumulation. Crucially, post-decisional CPP amplitude was more strongly associated with participants' metacognitive ratings following errors under short relative to long response deadlines. Our results suggest that superior metacognition under speed pressure may stem from enhanced metacognitive readout of post-decisional evidence.

Estimating self-performance when making complex decisions

Metacognition, the ability to monitor and reflect on our own mental states, enables us to assess our performance at different levels - from confidence in individual decisions to overall self-performance estimates (SPEs). It plays a particularly important part in computationally complex decisions that require a high level of cognitive resources, as the allocation of such limited resources presumably is based on metacognitive evaluations. However, little is known about metacognition in complex decisions, in particular, how people construct SPEs. Here, we examined how SPEs are modulated by task difficulty and feedback in cognitively complex economic decision-making, with reference to simple perceptual decision-making. We found that, in both types of decision-making, participants' objective performance was only affected by task difficulty but not by the presence of feedback. In complex economic decision-making, participants had lower SPEs in the absence of feedback (compared to the presence of feedback) in easy trials only but not in hard trials, while in simple perceptual decision-making, SPEs were lower in the absence of feedback in both easy and hard trials. Our findings suggest that people estimate their performance in complex economic decision-making through distinct metacognitive mechanisms for easy and hard instances.

Metacognitive confidence and affect - two sides of the same coin?

Decision confidence is a prototypical metacognitive representation that is thought to approximate the probability that a decision is correct. The perception of being correct has also been associated with affective valence such that being correct feels more positive and being mistaken more negative. This suggests that, similarly to confidence, affective valence reflects the probability that a decision is correct. However, both fields of research have seen very little interaction. Here, we test if affect, similarly to confidence reflects probability that a decision is correct in two perceptual decision-making experiments where we compare the relationships of theoretically relevant variables (e.g. evidence, accuracy, and expectancy) with both confidence and affect ratings. The findings indicate that confidence and affect ratings are similarly sensitive to changes in accuracy, evidence, and expectancy, indicating that both track the subjective probability that a decision is correct. We identify various mechanisms that can explain these results. We also envision future research for clarifying the role of cognitive and affective aspects of metacognition relying on deeper integration of the respective research fields.

A comprehensive assessment of current methods for measuring metacognition

One of the most important aspects of research on metacognition is the measurement of metacognitive ability. However, the properties of existing measures of metacognition have been mostly assumed rather than empirically established. Here I perform a comprehensive empirical assessment of 17 measures of metacognition. First, I develop a method of determining the validity and precision of a measure of metacognition and find that all 17 measures are valid and most show similar levels of precision. Second, I examine how measures of metacognition depend on task performance, response bias, and metacognitive bias, finding only weak dependences on response and metacognitive bias but many strong dependencies on task performance. Third, I find that all measures have very high split-half reliabilities, but most have poor test-retest reliabilities. This comprehensive assessment paints a complex picture: no measure of metacognition is perfect and different measures may be preferable in different experimental contexts.

No gender difference in confidence or metacognitive ability in perceptual decision-making

Prior research has found inconsistent results regarding gender differences in confidence and metacognitive ability. Different studies have shown that men are either more or less confident and have either higher or lower metacognitive abilities than women. However, this research has generally not used well-controlled tasks or used performance-independent measures of metacognitive ability. Here, we test for gender differences in performance, confidence, and metacognitive ability using data from 10 studies from the Confidence Database (total N = 1,887, total number of trials = 633,168). We find an absence of strong gender differences in performance and no gender differences in either confidence or metacognitive ability. These results were further confirmed by meta-analyses of the 10 datasets. These findings show that it is unlikely that gender has a strong effect on metacognitive evaluation in low-level perceptual decision-making and suggest that previously observed gender differences in confidence and metacognition are likely domain-specific.

Demystifying unsupervised learning: how it helps and hurts

Humans and machines rarely have access to explicit external feedback or supervision, yet manage to learn. Most modern machine learning systems succeed because they benefit from unsupervised data. Humans are also expected to benefit and yet, mysteriously, empirical results are mixed. Does unsupervised learning help humans or not? Here, we argue that the mixed results are not conflicting answers to this question, but reflect that humans self-reinforce their predictions in the absence of supervision, which can help or hurt depending on whether predictions and task align. We use this framework to synthesize empirical results across various domains to clarify when unsupervised learning will help or hurt. This provides new insights into the fundamentals of learning with implications for instruction and lifelong learning.

Confidence control for efficient behaviour in dynamic environments

Signatures of confidence emerge during decision-making, implying confidence may be of functional importance to decision processes themselves. We formulate an extension of sequential sampling models of decision-making in which confidence is used online to actively moderate the quality and quantity of evidence accumulated for decisions. The benefit of this model is that it can respond to dynamic changes in sensory evidence quality. We highlight this feature by designing a dynamic sensory environment where evidence quality can be smoothly adapted within the timeframe of a single decision. Our model with confidence control offers a superior description of human behaviour in this environment, compared to sequential sampling models without confidence control. Using multivariate decoding of electroencephalography (EEG), we uncover EEG correlates of the model's latent processes, and show stronger EEG-derived confidence control is associated with faster, more accurate decisions. These results support a neurobiologically plausible framework featuring confidence as an active control mechanism for improving behavioural efficiency.

Reliable, rapid, and remote measurement of metacognitive bias

Metacognitive biases have been repeatedly associated with transdiagnostic psychiatric dimensions of 'anxious-depression' and 'compulsivity and intrusive thought', cross-sectionally. To progress our understanding of the underlying neurocognitive mechanisms, new methods are required to measure metacognition remotely, within individuals over time. We developed a gamified smartphone task designed to measure visuo-perceptual metacognitive (confidence) bias and investigated its psychometric properties across two studies (N = 3410 unpaid citizen scientists, N = 52 paid participants). We assessed convergent validity, split-half and test-retest reliability, and identified the minimum number of trials required to capture its clinical correlates. Convergent validity of metacognitive bias was moderate (r(50) = 0.64, p < 0.001) and it demonstrated excellent split-half reliability (r(50) = 0.91, p < 0.001). Anxious-depression was associated with decreased confidence (β =  - 0.23, SE = 0.02, p < 0.001), while compulsivity and intrusive thought was associated with greater confidence (β = 0.07, SE = 0.02, p < 0.001). The associations between metacognitive biases and transdiagnostic psychiatry dimensions are evident in as few as 40 trials. Metacognitive biases in decision-making are stable within and across sessions, exhibiting very high test-retest reliability for the 100-trial (ICC = 0.86, N = 110) and 40-trial (ICC = 0.86, N = 120) versions of Meta Mind. Hybrid 'self-report cognition' tasks may be one way to bridge the recently discussed reliability gap in computational psychiatry.

Distinct basal ganglia contributions to learning from implicit and explicit value signals in perceptual decision-making

Metacognitive evaluations of confidence provide an estimate of decision accuracy that could guide learning in the absence of explicit feedback. We examine how humans might learn from this implicit feedback in direct comparison with that of explicit feedback, using simultaneous EEG-fMRI. Participants performed a motion direction discrimination task where stimulus difficulty was increased to maintain performance, with intermixed explicit- and no-feedback trials. We isolate single-trial estimates of post-decision confidence using EEG decoding, and find these neural signatures re-emerge at the time of feedback together with separable signatures of explicit feedback. We identified these signatures of implicit versus explicit feedback along a dorsal-ventral gradient in the striatum, a finding uniquely enabled by an EEG-fMRI fusion. These two signals appear to integrate into an aggregate representation in the external globus pallidus, which could broadcast updates to improve cortical decision processing via the thalamus and insular cortex, irrespective of the source of feedback.

What the eyes, confidence, and partner's identity can tell about change of mind

Perceptual confidence reflects the ability to evaluate the evidence that supports perceptual decisions. It is thought to play a critical role in guiding decision-making. However, only a few empirical studies have actually investigated the function of perceptual confidence. To address this issue, we designed a perceptual task in which participants provided a confidence judgment on the accuracy of their perceptual decision. Then, they viewed the response of a machine or human partner, and they were instructed to decide whether to keep or change their initial response. We observed that confidence predicted participants' changes of mind more than task difficulty and perceptual accuracy. Additionally, interacting with a machine, compared to a human, decreased confidence and increased participants tendency to change their initial decision, suggesting that both confidence and changes of mind are influenced by contextual factors, such as the identity of a partner. Finally, variations in confidence judgments but not change of mind were correlated with pre-response pupil dynamics, indicating that arousal changes are linked to confidence computations. This study contributes to our understanding of the factors influencing confidence and changes of mind and also evaluates the possibility of using pupil dynamics as a proxy of confidence.

Hands of confidence: When gestures increase confidence in spatial problem-solving

This study examined whether the metacognitive system monitors the potential positive effects of gestures on spatial thinking. Participants (N = 59, 31F, Mage = 21.67) performed a mental rotation task, consisting of 24 problems varying in difficulty, and they evaluated their confidence in their answers to problems in either gesture or control conditions. The results revealed that performance and confidence were higher in the gesture condition, in which the participants were asked to use their gestures during problem-solving, compared with the control condition, extending the literature by evidencing gestures' role in metacognition. Yet, the effect was only evident for females, who already performed worse than males, and when the problems were difficult. Encouraging gestures adversely affected performance and confidence in males. Such results suggest that gestures selectively influence cognition and metacognition and highlight the importance of task-related (i.e., difficulty) and individual-related variables (i.e., sex) in elucidating the links between gestures, confidence, and spatial thinking.

Think twice: Re-assessing confidence improves visual metacognition

Metacognition is a fundamental feature of human behavior that has adaptive functional value. Current understanding of the factors that influence metacognition remains incomplete, and we lack protocols to improve metacognition. Here, we introduce a two-step confidence choice paradigm to test whether metacognitive performance may improve by asking subjects to reassess their initial confidence. Previous work on perceptual and mnemonic decision-making has shown that (type 1) perceptual sensitivity benefits from reassessing the primary choice, however, it is not clear whether such an effect occurs for type 2 confidence choices. To test this hypothesis, we ran two separate online experiments, in which participants completed a type 1 task followed by two consecutive confidence choices. The results of the two experiments indicated that metacognitive sensitivity improved after re-evaluation. Since post-decisional evidence accumulation following the first confidence choice is likely to be minimal, this metacognitive improvement is better accounted for by an attenuation of metacognitive noise during the process of confidence generation. Thus, here we argue that metacognitive noise may be filtered out by additional post-decisional processing, thereby improving metacognitive sensitivity. We discuss the ramifications of these findings for models of metacognition and for developing protocols to train and manipulate metacognitive processes.

A confidence framing effect: Flexible use of evidence in metacognitive monitoring

Human behavior is flexibly regulated by specific goals of cognitive tasks. One notable example is goal-directed modulation of metacognitive behavior, where logically equivalent decision-making problems can yield different patterns of introspective confidence depending on the frame in which they are presented. While this observation highlights the important heuristic nature of metacognitive monitoring, computational mechanisms underlying this phenomenon remain elusive. We confirmed the confidence framing effect in two-alternative dot-number discrimination and in previously published preference-choice data, demonstrating distinctive confidence patterns between "choose more" or "choose less" frames. Formal model comparisons revealed a simple confidence heuristic behind this phenomenon, which assigns greater weight to chosen than unchosen stimulus evidence. This computation appears to be based on internal evidence constituted under specific task demands rather than physical stimulus intensity itself, a view justified in terms of ecological rationality. These results shed light on the adaptive nature of human decision-making and metacognitive monitoring.

Metacognition and Confidence: A Review and Synthesis

Determining the psychological, computational, and neural bases of confidence and uncertainty holds promise for understanding foundational aspects of human metacognition. While a neuroscience of confidence has focused on the mechanisms underpinning subpersonal phenomena such as representations of uncertainty in the visual or motor system, metacognition research has been concerned with personal-level beliefs and knowledge about self-performance. I provide a road map for bridging this divide by focusing on a particular class of confidence computation: propositional confidence in one's own (hypothetical) decisions or actions. Propositional confidence is informed by the observer's models of the world and their cognitive system, which may be more or less accurate-thus explaining why metacognitive judgments are inferential and sometimes diverge from task performance. Disparate findings on the neural basis of uncertainty and performance monitoring are integrated into a common framework, and a new understanding of the locus of action of metacognitive interventions is developed.

Differences in metacognitive functioning between obsessive-compulsive disorder patients and highly compulsive individuals from the general population

BACKGROUND

Our confidence, a form of metacognition, guides our behavior. Confidence abnormalities have been found in obsessive-compulsive disorder (OCD). A first notion based on clinical case-control studies suggests lower confidence in OCD patients compared to healthy controls. Contrarily, studies in highly compulsive individuals from general population samples showed that obsessive-compulsive symptoms related positively or not at all to confidence. A second notion suggests that an impairment in confidence estimation and usage is related to compulsive behavior, which is more often supported by studies in general population samples. These opposite findings call into question whether findings from highly compulsive individuals from the general population are generalizable to OCD patient populations.

METHODS

To test this, we investigated confidence at three hierarchical levels: local confidence in single decisions, global confidence in task performance and higher-order self-beliefs in 40 OCD patients (medication-free, no comorbid diagnoses), 40 controls, and 40 matched highly compulsive individuals from the general population (HComp).

RESULTS

In line with the first notion we found that OCD patients exhibited relative underconfidence at all three hierarchical levels. In contrast, HComp individuals showed local and global overconfidence and worsened metacognitive sensitivity compared with OCD patients, in line with the second notion.

CONCLUSIONS

Metacognitive functioning observed in a general highly compulsive population, often used as an analog for OCD, is distinct from that in a clinical OCD population, suggesting that OC symptoms in these two groups relate differently to (meta)cognitive processes. These findings call for caution in generalizing (meta)cognitive findings from general population to clinical samples.

Common computations for metacognition and meta-metacognition

Recent evidence shows that people have the meta-metacognitive ability to evaluate their metacognitive judgments of confidence. However, it is unclear whether meta-metacognitive judgments are made by a different system and rely on a separate set of computations compared to metacognitive judgments. To address this question, we asked participants (N = 36) to perform a perceptual decision-making task and provide (i) an object-level, Type-1 response about the identity of the stimulus; (ii) a metacognitive, Type-2 response (low/high) regarding their confidence in their Type-1 decision; and (iii) a meta-metacognitive, Type-3 response (low/high) regarding the quality of their Type-2 rating. We found strong evidence for the existence of Type-3, meta-metacognitive ability. In a separate condition, participants performed an identical task with only a Type-1 response followed by a Type-2 response given on a 4-point scale. We found that the two conditions produced equivalent results such that the combination of binary Type-2 and binary Type-3 responses acts similar to a 4-point Type-2 response. Critically, while Type-2 evaluations were subject to metacognitive noise, Type-3 judgments were made at no additional cost. These results suggest that it is unlikely that there is a distinction between Type-2 and Type-3 systems (metacognition and meta-metacognition) in perceptual decision-making and, instead, a single system can be flexibly adapted to produce both Type-2 and Type-3 evaluations recursively.

Studying the neural representations of uncertainty

The study of the brain's representations of uncertainty is a central topic in neuroscience. Unlike most quantities of which the neural representation is studied, uncertainty is a property of an observer's beliefs about the world, which poses specific methodological challenges. We analyze how the literature on the neural representations of uncertainty addresses those challenges and distinguish between 'code-driven' and 'correlational' approaches. Code-driven approaches make assumptions about the neural code for representing world states and the associated uncertainty. By contrast, correlational approaches search for relationships between uncertainty and neural activity without constraints on the neural representation of the world state that this uncertainty accompanies. To compare these two approaches, we apply several criteria for neural representations: sensitivity, specificity, invariance and functionality. Our analysis reveals that the two approaches lead to different but complementary findings, shaping new research questions and guiding future experiments.

Metacognition, public health compliance, and vaccination willingness

Metacognition, our ability to reflect on our own beliefs, manifests itself in the confidence we have in these beliefs, and helps us guide our behavior in complex and uncertain environments. Here, we provide empirical tests of the importance of metacognition during the pandemic. Bayesian and frequentist analyses demonstrate that citizens with higher metacognitive sensitivity-where confidence differentiates correct from incorrect COVID-19 beliefs-reported higher willingness to vaccinate against COVID-19, and higher compliance with recommended public health measures. Notably, this benefit of accurate introspection held controlling for the accuracy of COVID-19 beliefs. By demonstrating how vaccination willingness and compliance may relate to insight into the varying accuracy of beliefs, rather than only the accuracy of the beliefs themselves, this research highlights the critical role of metacognitive ability in times of crisis. However, we do not find sufficient evidence to conclude that citizens with higher metacognitive sensitivity were more likely to comply with recommended public health measures when controlling for the absolute level of the confidence citizens had in their COVID-19 beliefs.

Contributions of the Basal Ganglia to Visual Perceptual Decisions

The basal ganglia (BG) make up a prominent nexus between visual and motor-related brain regions. In contrast to the BG's well-established roles in movement control and value-based decision making, their contributions to the transformation of visual input into an action remain unclear, especially in the context of perceptual decisions based on uncertain visual evidence. This article reviews recent progress in our understanding of the BG's contributions to the formation, evaluation, and adjustment of such decisions. From theoretical and experimental perspectives, the review focuses on four key stations in the BG network, namely, the striatum, pallidum, subthalamic nucleus, and midbrain dopamine neurons, which can have different roles and together support the decision process.

Navigating the COVID-19 infodemic: the influence of metacognitive efficiency on health behaviours and policy attitudes

The COVID-19 pandemic has been accompanied by an infodemic of misinformation and increasing polarization around public health measures, such as social distancing and national lockdowns. In this study, I examined metacognitive efficiency-the extent to which the subjective feeling of knowing predicts the objective accuracy of knowledge-as a tool to understand and measure the assimilation of misleading misinformation in a balanced sample of Great Britain's population (N = 1689), surveyed at the end of the third national lockdown. Using a signal-detection theory approach to quantify metacognitive efficiency, I found that at the population level, metacognitive efficiency for COVID-19 knowledge was impaired compared with general knowledge, indicating a worse alignment between confidence levels and the actual ability to discern true and false statements. Crucially, individual differences in metacognitive efficiency related to COVID-19 knowledge predicted health-protective behaviours, vaccination intentions and attitudes towards public health measures, even after accounting for the level of knowledge itself and demographic covariates, such as education, income and political alignment. These results reveal the significant impact of misinformation on public beliefs and suggest that fostering confidence in accurate knowledge should be a key target for science communication efforts aimed at promoting compliance with public health and social measures.

Preserved Metacognition for Undetected Visuomotor Deviations

Humans can successfully correct deviations of movements without conscious detection of such deviations, suggesting limited awareness of movement details. We ask whether such limited awareness impairs confidence (metacognition). We recorded functional magnetic resonance imaging data while 31 human female and male participants detected cursor deviations during a visuomotor reaching task and rated their confidence retrospectively. We show that participants monitor a summary statistic of the unfolding visual feedback (the peak cursor error) to detect visuomotor deviations and adjust their confidence ratings, even when they report being unaware of a deviation. Crucially, confidence ratings were as metacognitively efficient for aware and unaware deviations. At the neural level, activity in the ventral striatum tracks high confidence, whereas a broad network encodes cursor error but not confidence. These findings challenge the notion of limited conscious action monitoring and uncover how humans monitor their movements as they unfold, even when unaware of ongoing deviations.SIGNIFICANCE STATEMENT We are unaware of the small corrections we apply to our movements as long as our goals are achieved. Here, although we replicate the finding that participants deny perceiving small deviations they correct, we show that their confidence reliably reflects the presence or absence of a deviation. This observation shows they can metacognitively monitor the presence of a deviation, even when they deny perceiving it. We also describe the hemodynamic correlates of confidence ratings. Our study questions the extent to which humans are unaware of the details of their movements; describes a plausible mechanism for metacognition in a visuomotor task, along with its neural correlates; and has important implications for the construction of the sense of self.

Expectations about precision bias metacognition and awareness

Bayesian models of the mind suggest that we estimate the reliability or "precision" of incoming sensory signals to guide perceptual inference and to construct feelings of confidence or uncertainty about what we are perceiving. However, accurately estimating precision is likely to be challenging for bounded systems like the brain. One way observers could overcome this challenge is to form expectations about the precision of their perceptions and use these to guide metacognition and awareness. Here we test this possibility. Participants made perceptual decisions about visual motion stimuli, while providing confidence ratings (Experiments 1 and 2) or ratings of subjective visibility (Experiment 3). In each experiment, participants acquired probabilistic expectations about the likely strength of upcoming signals. We found these expectations about precision altered metacognition and awareness-with participants feeling more confident and stimuli appearing more vivid when stronger sensory signals were expected, without concomitant changes in objective perceptual performance. Computational modeling revealed that this effect could be well explained by a predictive learning model that infers the precision (strength) of current signals as a weighted combination of incoming evidence and top-down expectation. These results support an influential but untested tenet of Bayesian models of cognition, suggesting that agents do not only "read out" the reliability of information arriving at their senses, but also take into account prior knowledge about how reliable or "precise" different sources of information are likely to be. This reveals that expectations about precision influence how the sensory world appears and how much we trust our senses. (PsycInfo Database Record (c) 2023 APA, all rights reserved).

Blocking D2/D3 dopamine receptors in male participants increases volatility of beliefs when learning to trust others

The ability to learn about other people is crucial for human social functioning. Dopamine has been proposed to regulate the precision of beliefs, but direct behavioural evidence of this is lacking. In this study, we investigate how a high dose of the D2/D3 dopamine receptor antagonist sulpiride impacts learning about other people's prosocial attitudes in a repeated Trust game. Using a Bayesian model of belief updating, we show that in a sample of 76 male participants sulpiride increases the volatility of beliefs, which leads to higher precision weights on prediction errors. This effect is driven by participants with genetically conferred higher dopamine availability (Taq1a polymorphism) and remains even after controlling for working memory performance. Higher precision weights are reflected in higher reciprocal behaviour in the repeated Trust game but not in single-round Trust games. Our data provide evidence that the D2 receptors are pivotal in regulating prediction error-driven belief updating in a social context.

Flexible use of confidence to guide advice requests

Subjective confidence plays an important role in guiding behavior, for example, people typically commit to decisions immediately if high in confidence and seek additional information if not. The present study examines whether people are flexible in their use of confidence, such that the mapping between confidence and behavior is not fixed but can instead vary depending on the specific context. To investigate this proposal, we tested the hypothesis that the seemingly natural relationship between low confidence and requesting advice varies according to whether people know, or do not know, the quality of the advice. Participants made an initial perceptual judgement and then chose between re-sampling evidence or receiving advice from a virtual advisor, before committing to a final decision. The results indicated that, when objective information about advisor reliability was not available, participants selected advice more often when their confidence was high rather than when it was low. This pattern reflects the use of confidence as a feedback proxy to learn about advisor quality: Participants were able to learn about the reliability of advice even in the absence of feedback and subsequently requested more advice from better advisors. In contrast, when participants had prior knowledge about the reliability of advisors, they requested advice more often when their confidence was low, reflecting the use of confidence as a self-monitoring tool signaling that help should be solicited. These findings indicate that people use confidence in a way that is context-dependent and directed towards achieving their current goals.

Metacognition tracks sensitivity following involuntary shifts of visual attention

Salient, exogenous cues have been shown to induce a temporary boost of perceptual sensitivity in their immediate vicinity. In two experiments involving uninformative exogenous cues presented at various times before a target stimulus, we investigated whether human observers (N = 100) were able to monitor the involuntary increase in performance induced by such transients. We found that an increase of perceptual sensitivity (in a choice task) and encoding precision (in a free-estimation task) occurred approximately 100 ms after cue onset, and was accompanied by an increase in confidence about the perceptual response. These simultaneous changes in sensitivity and confidence resulted in stable metacognition across conditions. These results suggest that metacognition efficiently tracks the effects of a reflexive attentional mechanism known to evade voluntary control, and illustrate a striking ability of high-level cognition to capture fleeting, low-level sensory modulations.

Recent Advances at the Interface of Neuroscience and Artificial Neural Networks

Biological neural networks adapt and learn in diverse behavioral contexts. Artificial neural networks (ANNs) have exploited biological properties to solve complex problems. However, despite their effectiveness for specific tasks, ANNs are yet to realize the flexibility and adaptability of biological cognition. This review highlights recent advances in computational and experimental research to advance our understanding of biological and artificial intelligence. In particular, we discuss critical mechanisms from the cellular, systems, and cognitive neuroscience fields that have contributed to refining the architecture and training algorithms of ANNs. Additionally, we discuss how recent work used ANNs to understand complex neuronal correlates of cognition and to process high throughput behavioral data.

Consensus Goals in the Field of Visual Metacognition

Despite the tangible progress in psychological and cognitive sciences over the last several years, these disciplines still trail other more mature sciences in identifying the most important questions that need to be solved. Reaching such consensus could lead to greater synergy across different laboratories, faster progress, and increased focus on solving important problems rather than pursuing isolated, niche efforts. Here, 26 researchers from the field of visual metacognition reached consensus on four long-term and two medium-term common goals. We describe the process that we followed, the goals themselves, and our plans for accomplishing these goals. If this effort proves successful within the next few years, such consensus building around common goals could be adopted more widely in psychological science.

Confidence at the limits of human nested cognition

Metacognition is the ability to weigh the quality of our own cognition, such as the confidence that our perceptual decisions are correct. Here we ask whether metacognitive performance can itself be evaluated or else metacognition is the ultimate reflective human faculty. Building upon a classic visual perception task, we show that human observers are able to produce nested, above-chance judgements on the quality of their decisions at least up to the fourth order (i.e. meta-meta-meta-cognition). A computational model can account for this nested cognitive ability if evidence has a high-resolution representation, and if there are two kinds of noise, including recursive evidence degradation. The existence of fourth-order sensitivity suggests that the neural mechanisms responsible for second-order metacognition can be flexibly generalized to evaluate any cognitive process, including metacognitive evaluations themselves. We define the theoretical and practical limits of nested cognition and discuss how this approach paves the way for a better understanding of human self-regulation.

The value of confidence: Confidence prediction errors drive value-based learning in the absence of external feedback

Reinforcement learning algorithms have a long-standing success story in explaining the dynamics of instrumental conditioning in humans and other species. While normative reinforcement learning models are critically dependent on external feedback, recent findings in the field of perceptual learning point to a crucial role of internally generated reinforcement signals based on subjective confidence, when external feedback is not available. Here, we investigated the existence of such confidence-based learning signals in a key domain of reinforcement-based learning: instrumental conditioning. We conducted a value-based decision making experiment which included phases with and without external feedback and in which participants reported their confidence in addition to choices. Behaviorally, we found signatures of self-reinforcement in phases without feedback, reflected in an increase of subjective confidence and choice consistency. To clarify the mechanistic role of confidence in value-based learning, we compared a family of confidence-based learning models with more standard models predicting either no change in value estimates or a devaluation over time when no external reward is provided. We found that confidence-based models indeed outperformed these reference models, whereby the learning signal of the winning model was based on the prediction error between current confidence and a stimulus-unspecific average of previous confidence levels. Interestingly, individuals with more volatile reward-based value updates in the presence of feedback also showed more volatile confidence-based value updates when feedback was not available. Together, our results provide evidence that confidence-based learning signals affect instrumentally learned subjective values in the absence of external feedback.

Reinforcement learning models successfully simulate value-based learning processes (e.g., “How worthwhile is it to choose the same option again?”) when external reward feedback is provided (e.g., drops of sweet liquids or money). But does learning stagnate if such feedback is no longer provided? Recently, a number of studies have shown that subjective confidence can likewise act as an internal reward signal, when external feedback is not available. These results are in line with the intuitive experience that being confident about choices and actions comes with a satisfying feeling of accomplishment. To better understand the role of confidence in value-based learning, we designed a study in which participants had to learn the value of choice options in phases with and without external feedback. Behaviorally, we found signatures of self-reinforcement, such as increased confidence and choice consistency, in phases without feedback. To examine the underlying mechanisms, we compared computational models, in which learning was guided by confidence signals, with more standard reinforcement learning models. A statistical comparison of these models showed that a confidence-based model in which generic confidence prediction errors (e.g., “Am I as confident as expected?”) guide learning indeed outperformed the standard models.

Reverse engineering of metacognition

The human ability to introspect on thoughts, perceptions or actions − metacognitive ability − has become a focal topic of both cognitive basic and clinical research. At the same time it has become increasingly clear that currently available quantitative tools are limited in their ability to make unconfounded inferences about metacognition. As a step forward, the present work introduces a comprehensive modeling framework of metacognition that allows for inferences about metacognitive noise and metacognitive biases during the readout of decision values or at the confidence reporting stage. The model assumes that confidence results from a continuous but noisy and potentially biased transformation of decision values, described by a confidence link function. A canonical set of metacognitive noise distributions is introduced which differ, amongst others, in their predictions about metacognitive sign flips of decision values. Successful recovery of model parameters is demonstrated, and the model is validated on an empirical data set. In particular, it is shown that metacognitive noise and bias parameters correlate with conventional behavioral measures. Crucially, in contrast to these conventional measures, metacognitive noise parameters inferred from the model are shown to be independent of performance. This work is accompanied by a toolbox (ReMeta) that allows researchers to estimate key parameters of metacognition in confidence datasets.

Metacognition is a person’s ability to think about their own thoughts. For example, imagine you are walking in a dark forest when you see an elongated object. You think it is a stick rather than a snake, but how sure are you? Reflecting on one’s certainty about own thoughts or perceptions – confidence – is a prime example of metacognition. While our ability to think about our own thoughts in this way provides many, perhaps uniquely human, advantages, confidence judgements are prone to biases. Often, humans tend to be overconfident: we think we are right more often than we actually are. Internal noise of neural processes can also affect confidence.

Understanding these imperfections in metacognition could shed light on how humans think, but studying this phenomenon is challenging. Current methods are lacking either mechanistic insight about the sources of metacognitive biases and noise or rely on unrealistic assumptions. A better model for how metacognition works could provide a clearer picture.

Guggenmos developed a mathematical model and a computer toolbox to help researchers investigate how humans or animals estimate confidence in their own thoughts and resulting decisions . The model splits metacognition apart, allowing scientists to explore biases and sources of noise at different phases in the process. It takes two kinds of data: the decisions study participants make, and how sure they are about their decision being correct. It then recreates metacognition in three phases: the primary decision, the metacognitive readout of the evidence, and the confidence report. This allows investigators to see where and when noise and bias come into play. Guggenmos tested the model using independent data from a visual discrimination task and found that it was able to predict how confident participants reported to be in their decisions.

Metacognitive ability can change in people with mental illness. People with schizophrenia have often been found to be overconfident in their decisions, while people with depression can be underconfident. Using this model to separate the various facets of metacognition could help to explain why. It could also shed light on human thinking in general.

On the assumptions behind metacognitive measurements: Implications for theory and practice

Theories of visual confidence have largely been grounded in the gaussian signal detection framework. This framework is so dominant that idiosyncratic consequences from this distributional assumption have remained unappreciated. This article reports systematic comparisons of the gaussian signal detection framework to its logistic counterpart in the measurement of metacognitive accuracy. Because of the difference in their distribution kurtosis, these frameworks are found to provide different perspectives regarding the efficiency of confidence rating relative to objective decision (the logistic model intrinsically gives greater meta-d'/d' ratio than the gaussian model). These frameworks can also provide opposing conclusions regarding the metacognitive inefficiency along the internal evidence continuum (whether meta-d' is larger or smaller for higher levels of confidence). Previous theories developed on these lines of analysis may need to be revisited as the gaussian and logistic metacognitive models received somewhat equivalent support in our quantitative model comparisons. Despite these discrepancies, however, we found that across-condition or across-participant comparisons of metacognitive measures are relatively robust against the distributional assumptions, which provides much assurance to conventional research practice. We hope this article promotes the awareness for the significance of hidden modeling assumptions, contributing to the cumulative development of the relevant field.

Partially Overlapping Neural Correlates of Metacognitive Monitoring and Metacognitive Control

Metacognition describes the process of monitoring one's own mental states, often for the purpose of cognitive control. Previous research has investigated how metacognitive signals are generated (metacognitive monitoring), for example, when people (both female/male) judge their confidence in their decisions and memories. Research has also investigated how metacognitive signals are used to influence behavior (metacognitive control), for example, setting a reminder (i.e., cognitive offloading) for something you are not confident you will remember. However, the mapping between metacognitive monitoring and metacognitive control needs further study on a neural level. We used fMRI to investigate a delayed-intentions task with a reminder element, allowing human participants to use their metacognitive insight to engage metacognitive control. Using multivariate pattern analysis, we found that we could separately decode both monitoring and control, and, to a lesser extent, cross-classify between them. Therefore, brain patterns associated with monitoring and control are partially, but not fully, overlapping.SIGNIFICANCE STATEMENT Models of metacognition commonly distinguish between monitoring (how metacognition is formed) and control (how metacognition is used for behavioral regulation). Research into these facets of metacognition has often happened in isolation. Here, we provide a study which directly investigates the mapping between metacognitive monitoring and metacognitive control at a neural level. We applied multivariate pattern analysis to fMRI data from a novel task in which participants separately rated their confidence (metacognitive monitoring) and how much they would like to use a reminder (metacognitive control). We find support for the notion that the two aspects of metacognition overlap partially but not fully. We argue that future research should focus on how different metacognitive signals are selected for control.

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

While often seeming to investigate rather different problems, research into value-based decision making and cognitive control have historically offered parallel insights into how people select thoughts and actions. While the former studies how people weigh costs and benefits to make a decision, the latter studies how they adjust information processing to achieve their goals. Recent work has highlighted ways in which decision-making research can inform our understanding of cognitive control. Here, we provide the complementary perspective: how cognitive control research has informed understanding of decision-making. We highlight three particular areas of research where this critical interchange has occurred: (1) how different types of goals shape the evaluation of choice options, (2) how people use control to adjust the ways they make their decisions, and (3) how people monitor decisions to inform adjustments to control at multiple levels and timescales. We show how adopting this alternate viewpoint offers new insight into the determinants of both decisions and control; provides alternative interpretations for common neuroeconomic findings; and generates fruitful directions for future research.

Confidence-Controlled Hebbian Learning Efficiently Extracts Category Membership From Stimuli Encoded in View of a Categorization Task

In experiments on perceptual decision making, individuals learn a categorization task through trial-and-error protocols. We explore the capacity of a decision-making attractor network to learn a categorization task through reward-based, Hebbian-type modifications of the weights incoming from the stimulus encoding layer. For the latter, we assume a standard layer of a large number of stimulus-specific neurons. Within the general framework of Hebbian learning, we have hypothesized that the learning rate is modulated by the reward at each trial. Surprisingly, we find that when the coding layer has been optimized in view of the categorization task, such reward-modulated Hebbian learning (RMHL) fails to extract efficiently the category membership. In previous work, we showed that the attractor neural networks' nonlinear dynamics accounts for behavioral confidence in sequences of decision trials. Taking advantage of these findings, we propose that learning is controlled by confidence, as computed from the neural activity of the decision-making attractor network. Here we show that this confidence-controlled, reward-based Hebbian learning efficiently extracts categorical information from the optimized coding layer. The proposed learning rule is local and, in contrast to RMHL, does not require storing the average rewards obtained on previous trials. In addition, we find that the confidence-controlled learning rule achieves near-optimal performance. In accordance with this result, we show that the learning rule approximates a gradient descent method on a maximizing reward cost function.

Examining the robustness of the relationship between metacognitive efficiency and metacognitive bias

We recently found a positive relationship between estimates of metacognitive efficiency and metacognitive bias. However, this relationship was only examined on a within-subject level and required binarizing the confidence scale, a technique that introduces methodological difficulties. Here we examined the robustness of the positive relationship between estimates of metacognitive efficiency and metacognitive bias by conducting two different types of analyses. First, we developed a new within-subject analysis technique where the original n-point confidence scale is transformed into two different (n-1)-point scales in a way that mimics a naturalistic change in confidence. Second, we examined the across-subject correlation between metacognitive efficiency and metacognitive bias. Importantly, for both types of analyses, we not only established the direction of the effect but also computed effect sizes. We applied both techniques to the data from three tasks from the Confidence Database (N > 400 in each). We found that both approaches revealed a small to medium positive relationship between metacognitive efficiency and metacognitive bias. These results demonstrate that the positive relationship between metacognitive efficiency and metacognitive bias is robust across several analysis techniques and datasets, and have important implications for future research.

Confidence guides priority between forthcoming tasks

Humans can estimate confidence in their decisions, and there is increasing interest on how this feeling of confidence regulates future behavior. Here, we investigate whether confidence in a perceptual task affects prioritizing future trials of that task, independently of task performance. To do so, we experimentally dissociated confidence from performance. Participants judged whether an array of differently colored circles was closer to blue or red, and we manipulated the mean and variability of the circles' colors across the array. We first familiarized participants with a low mean low variability condition and a high mean high variability condition, which were matched in performance despite participants being more confident in the former. Then we made participants decide in which order to complete forthcoming trials for both conditions. Crucially, prioritizing one condition was associated with being more confident in that condition compared to the other. This relationship was observed both across participants, by correlating inter-individual heterogeneity in prioritization and in confidence, and within participants, by assessing how changes in confidence with accuracy, condition and response times could predict prioritization choices. Our results suggest that confidence, above and beyond performance, guides prioritization between forthcoming tasks, strengthening the evidence for its role in regulating behavior.

Metacognitive resources for adaptive learning⋆

Biological organisms display remarkably flexible behaviours. This is an area of active investigation, in particular in the fields of artificial intelligence, computational and cognitive neuroscience. While inductive biases and broader cognitive functions are undoubtedly important, the ability to monitor and evaluate one's performance or oneself -- metacognition -- strikes as a powerful resource for efficient learning. Often measured as decision confidence in neuroscience and psychology experiments, metacognition appears to reflect a broad range of abstraction levels and downstream behavioural effects. Within this context, the formal investigation of how metacognition interacts with learning processes is a recent endeavour. Of special interest are the neural and computational underpinnings of confidence and reinforcement learning modules. This review discusses a general hierarchy of confidence functions and their neuro-computational relevance for adaptive behaviours. It then introduces novel ways to study the formation and use of meta-representations and nonconscious mental representations related to learning and confidence, and concludes with a discussion on outstanding questions and wider perspectives.

Cholinergic and dopaminergic effects on prediction error and uncertainty responses during sensory associative learning

Navigating the physical world requires learning probabilistic associations between sensory events and their change in time (volatility). Bayesian accounts of this learning process rest on hierarchical prediction errors (PEs) that are weighted by estimates of uncertainty (or its inverse, precision). In a previous fMRI study we found that low-level precision-weighted PEs about visual outcomes (that update beliefs about associations) activated the putative dopaminergic midbrain; by contrast, precision-weighted PEs about cue-outcome associations (that update beliefs about volatility) activated the cholinergic basal forebrain. These findings suggested selective dopaminergic and cholinergic influences on precision-weighted PEs at different hierarchical levels. Here, we tested this hypothesis, repeating our fMRI study under pharmacological manipulations in healthy participants. Specifically, we performed two pharmacological fMRI studies with a between-subject double-blind placebo-controlled design: study 1 used antagonists of dopaminergic (amisulpride) and muscarinic (biperiden) receptors, study 2 used enhancing drugs of dopaminergic (levodopa) and cholinergic (galantamine) modulation. Pooled across all pharmacological conditions of study 1 and study 2, respectively, we found that low-level precision-weighted PEs activated the midbrain and high-level precision-weighted PEs the basal forebrain as in our previous study. However, we found pharmacological effects on brain activity associated with these computational quantities only when splitting the precision-weighted PEs into their PE and precision components: in a brainstem region putatively containing cholinergic (pedunculopontine and laterodorsal tegmental) nuclei, biperiden (compared to placebo) enhanced low-level PE responses and attenuated high-level PE activity, while amisulpride reduced high-level PE responses. Additionally, in the putative dopaminergic midbrain, galantamine compared to placebo enhanced low-level PE responses (in a body-weight dependent manner) and amisulpride enhanced high-level precision activity. Task behaviour was not affected by any of the drugs. These results do not support our hypothesis of a clear-cut dichotomy between different hierarchical inference levels and neurotransmitter systems, but suggest a more complex interaction between these neuromodulatory systems and hierarchical Bayesian quantities. However, our present results may have been affected by confounds inherent to pharmacological fMRI. We discuss these confounds and outline improved experimental tests for the future.

Visual attention modulates the integration of goal-relevant evidence and not value

When choosing between options, such as food items presented in plain view, people tend to choose the option they spend longer looking at. The prevailing interpretation is that visual attention increases value. However, in previous studies, 'value' was coupled to a behavioural goal, since subjects had to choose the item they preferred. This makes it impossible to discern if visual attention has an effect on value, or, instead, if attention modulates the information most relevant for the goal of the decision-maker. Here, we present the results of two independent studies-a perceptual and a value-based task-that allow us to decouple value from goal-relevant information using specific task-framing. Combining psychophysics with computational modelling, we show that, contrary to the current interpretation, attention does not boost value, but instead it modulates goal-relevant information. This work provides a novel and more general mechanism by which attention interacts with choice.

Learning from Others, but with What Confidence?

A recent study by Zhang and Gläscher (2020) in humans examines learning from one's own versus others' actions under reward uncertainty. Comparing findings from this and non-human studies on learning under perceptual uncertainty suggests a unified role for confidence in learning under different types of uncertainty across mammalian brains.

Formation of global self-beliefs in the human brain

Momentary feelings of confidence accompany many of our actions and decisions. In addition to such “local” feelings of confidence, we also construct “global” confidence estimates about our skills and abilities (global self-performance estimates or SPEs). Distorted SPEs may have a pervasive impact on motivation and self-evaluation, for instance affecting estimates of our competitiveness at work or in a sports team. Here, we found that components of a brain network previously implicated in the tracking of local confidence was additionally modulated by SPE level, whereas ventral striatum tracked SPEs irrespective of confidence. Our findings of a neurocognitive basis for global SPEs lay the groundwork for understanding how distorted SPEs arise in educational and clinical settings.

Humans create metacognitive beliefs about their performance across many levels of abstraction—from local confidence in individual decisions to global estimates of our skills and abilities. Despite a rich literature on the neural basis of local confidence judgements, how global self-performance estimates (SPEs) are constructed remains unknown. Using functional magnetic resonance imaging, we scanned human subjects while they performed several short blocks of tasks and reported on which task they think they performed best, providing a behavioral proxy for global SPEs. In a frontoparietal network sensitive to fluctuations in local confidence, we found that activity within ventromedial prefrontal cortex and precuneus was additionally modulated by global SPEs. In contrast, activity in ventral striatum was associated with subjects’ global SPEs irrespective of fluctuations in local confidence, and predicted the extent to which global SPEs tracked objective task difficulty across individuals. Our findings reveal neural representations of global SPEs that go beyond the tracking of local confidence, and lay the groundwork for understanding how a formation of global self-beliefs may go awry in conditions characterized by distorted self-evaluation.

Unconscious reinforcement learning of hidden brain states supported by confidence

Can humans be trained to make strategic use of latent representations in their own brains? We investigate how human subjects can derive reward-maximizing choices from intrinsic high-dimensional information represented stochastically in neural activity. Reward contingencies are defined in real-time by fMRI multivoxel patterns; optimal action policies thereby depend on multidimensional brain activity taking place below the threshold of consciousness, by design. We find that subjects can solve the task within two hundred trials and errors, as their reinforcement learning processes interact with metacognitive functions (quantified as the meaningfulness of their decision confidence). Computational modelling and multivariate analyses identify a frontostriatal neural mechanism by which the brain may untangle the 'curse of dimensionality': synchronization of confidence representations in prefrontal cortex with reward prediction errors in basal ganglia support exploration of latent task representations. These results may provide an alternative starting point for future investigations into unconscious learning and functions of metacognition.