Tuned inhibition in perceptual decision-making circuits can explain seemingly suboptimal confidence behavior.

Current dominant views hold that perceptual confidence reflects the probability that a decision is correct. Although these views have enjoyed some empirical support, recent behavioral results indicate that confidence and the probability of being correct can be dissociated. An alternative hypothesis suggests that confidence instead reflects the magnitude of evidence in favor of a decision while being relatively insensitive to the evidence opposing the decision. We considered how this alternative hypothesis might be biologically instantiated by developing a simple neural network model incorporating a known property of sensory neurons: tuned inhibition. The key idea of the model is that the level of inhibition that each accumulator unit receives from units with the opposite tuning preference, i.e. its inhibition 'tuning', dictates its contribution to perceptual decisions versus confidence judgments, such that units with higher tuned inhibition (computing relative evidence for different perceptual interpretations) determine perceptual discrimination decisions, and units with lower tuned inhibition (computing absolute evidence) determine confidence. We demonstrate that this biologically plausible model can account for several counterintuitive findings reported in the literature where confidence and decision accuracy dissociate. By comparing model fits, we further demonstrate that a full complement of behavioral data across several previously published experimental results-including accuracy, reaction time, mean confidence, and metacognitive sensitivity-is best accounted for when confidence is computed from units without, rather than units with, tuned inhibition. Finally, we discuss predictions of our results and model for future neurobiological studies. These findings suggest that the brain has developed and implements this alternative, heuristic theory of perceptual confidence computation by relying on the diversity of neural resources available.

Neural Integration of Stimulus History Underlies Prediction for Naturalistically Evolving Sequences.

Forming valid predictions about the environment is crucial to survival. However, whether humans are able to form valid predictions about natural stimuli based on their temporal statistical regularities remains unknown. Here, we presented subjects with tone sequences with pitch fluctuations that, over time, capture long-range temporal dependence structures prevalent in natural stimuli. We found that subjects were able to exploit such naturalistic statistical regularities to make valid predictions about upcoming items in a sequence. Magnetoencephalography (MEG) recordings revealed that slow, arrhythmic cortical dynamics tracked the evolving pitch sequence over time such that neural activity at a given moment was influenced by the pitch of up to seven previous tones. Importantly, such history integration contained in neural activity predicted the expected pitch of the upcoming tone, providing a concrete computational mechanism for prediction. These results establish humans' ability to make valid predictions based on temporal regularities inherent in naturalistic stimuli and further reveal the neural mechanisms underlying such predictive computation.SIGNIFICANCE STATEMENT A fundamental question in neuroscience is how the brain predicts upcoming events in the environment. To date, this question has primarily been addressed in experiments using relatively simple stimulus sequences. Here, we studied predictive processing in the human brain using auditory tone sequences that exhibit temporal statistical regularities similar to those found in natural stimuli. We observed that humans are able to form valid predictions based on such complex temporal statistical regularities. We further show that neural response to a given tone in the sequence reflects integration over the preceding tone sequence and that this history dependence forms the foundation for prediction. These findings deepen our understanding of how humans form predictions in an ecologically valid environment.

Limited Cognitive Resources Explain a Trade-Off between Perceptual and Metacognitive Vigilance.

Why do experimenters give subjects short breaks in long behavioral experiments? Whereas previous studies suggest it is difficult to maintain attention and vigilance over long periods of time, it is unclear precisely what mechanisms benefit from rest after short experimental blocks. Here, we evaluate decline in both perceptual performance and metacognitive sensitivity (i.e., how well confidence ratings track perceptual decision accuracy) over time and investigate whether characteristics of prefrontal cortical areas correlate with these measures. Whereas a single-process signal detection model predicts that these two forms of fatigue should be strongly positively correlated, a dual-process model predicts that rates of decline may dissociate. Here, we show that these measures consistently exhibited negative or near-zero correlations, as if engaged in a trade-off relationship, suggesting that different mechanisms contribute to perceptual and metacognitive decisions. Despite this dissociation, the two mechanisms likely depend on common resources, which could explain their trade-off relationship. Based on structural MRI brain images of individual human subjects, we assessed gray matter volume in the frontal polar area, a region that has been linked to visual metacognition. Variability of frontal polar volume correlated with individual differences in behavior, indicating the region may play a role in supplying common resources for both perceptual and metacognitive vigilance. Additional experiments revealed that reduced metacognitive demand led to superior perceptual vigilance, providing further support for this hypothesis. Overall, results indicate that during breaks between short blocks, it is the higher-level perceptual decision mechanisms, rather than lower-level sensory machinery, that benefit most from rest. SIGNIFICANCE STATEMENT: Perceptual task performance declines over time (the so-called vigilance decrement), but the relationship between vigilance in perception and metacognition has not yet been explored in depth. Here, we show that patterns in perceptual and metacognitive vigilance do not follow the pattern predicted by a previously suggested single-process model of perceptual and metacognitive decision making. We account for these findings by showing that regions of anterior prefrontal cortex (aPFC) previously associated with visual metacognition are also associated with perceptual vigilance. We also show that relieving metacognitive task demand improves perceptual vigilance, suggesting that aPFC may house a limited cognitive resource that contributes to both metacognition and perceptual vigilance. These findings advance our understanding of the mechanisms and dynamics of perceptual metacognition.

Initial-state-dependent, robust, transient neural dynamics encode conscious visual perception.

Recent research has identified late-latency, long-lasting neural activity as a robust correlate of conscious perception. Yet, the dynamical nature of this activity is poorly understood, and the mechanisms governing its presence or absence and the associated conscious perception remain elusive. We applied dynamic-pattern analysis to whole-brain slow (< 5 Hz) cortical dynamics recorded by magnetoencephalography (MEG) in human subjects performing a threshold-level visual perception task. Up to 1 second before stimulus onset, brain activity pattern across widespread cortices significantly predicted whether a threshold-level visual stimulus was later consciously perceived. This initial state of brain activity interacts nonlinearly with stimulus input to shape the evolving cortical activity trajectory, with seen and unseen trials following well separated trajectories. We observed that cortical activity trajectories during conscious perception are fast evolving and robust to small variations in the initial state. In addition, spontaneous brain activity pattern prior to stimulus onset also influences unconscious perceptual making in unseen trials. Together, these results suggest that brain dynamics underlying conscious visual perception belongs to the class of initial-state-dependent, robust, transient neural dynamics.

On a 'failed' attempt to manipulate visual metacognition with transcranial magnetic stimulation to prefrontal cortex.

Rounis, Maniscalco, Rothwell, Passingham, and Lau (2010) reported that stimulation of prefrontal cortex impairs visual metacognition. Bor, Schwartzman, Barrett, and Seth (2017) attempted to replicate this result, but adopted an experimental design that reduced their chanceof obtaining positive findings. Despite that, their results appeared initially consistent with those of Rounis et al., but they subsequently claimed it was necessary to discard ∼30% of their subjects, after which they reported a null result. Using computer simulations, we found that, contrary to their supposed purpose, excluding subjects by Bor et al.'s criteria does not reduce false positive rates. Including both their positive and negative result in a Bayesian framework, we show the correct interpretation is that PFC stimulation likely impaired visual metacognition, exactly contradicting Bor et al.'s claims. That lesion and inactivation studies demonstrate similar positive effects further suggests that Bor et al.'s reported negative finding isn't evidence against the role of prefrontal cortex in metacognition.

Modulating conscious movement intention by noninvasive brain stimulation and the underlying neural mechanisms.

Conscious intention is a fundamental aspect of the human experience. Despite long-standing interest in the basis and implications of intention, its underlying neurobiological mechanisms remain poorly understood. Using high-definition transcranial DC stimulation (tDCS), we observed that enhancing spontaneous neuronal excitability in both the angular gyrus and the primary motor cortex caused the reported time of conscious movement intention to be ∼60-70 ms earlier. Slow brain waves recorded ∼2-3 s before movement onset, as well as hundreds of milliseconds after movement onset, independently correlated with the modulation of conscious intention by brain stimulation. These brain activities together accounted for 81% of interindividual variability in the modulation of movement intention by brain stimulation. A computational model using coupled leaky integrator units with biophysically plausible assumptions about the effect of tDCS captured the effects of stimulation on both neural activity and behavior. These results reveal a temporally extended brain process underlying conscious movement intention that spans seconds around movement commencement.

Action-specific disruption of perceptual confidence.

Theoretical models of perception assume that confidence is related to the quality or strength of sensory processing. Counter to this intuitive view, we showed in the present research that the motor system also contributes to judgments of perceptual confidence. In two experiments, we used transcranial magnetic stimulation (TMS) to manipulate response-specific representations in the premotor cortex, selectively disrupting postresponse confidence in visual discrimination judgments. Specifically, stimulation of the motor representation associated with the unchosen response reduced confidence in correct responses, thereby reducing metacognitive capacity without changing visual discrimination performance. Effects of TMS on confidence were observed when stimulation was applied both before and after the response occurred, which suggests that confidence depends on late-stage metacognitive processes. These results place constraints on models of perceptual confidence and metacognition by revealing that action-specific information in the premotor cortex contributes to perceptual confidence.

Anatomical coupling between distinct metacognitive systems for memory and visual perception.

A recent study found that, across individuals, gray matter volume in the frontal polar region was correlated with visual metacognition capacity (i.e., how well one's confidence ratings distinguish between correct and incorrect judgments). A question arises as to whether the putative metacognitive mechanisms in this region are also used in other metacognitive tasks involving, for example, memory. A novel psychophysical measure allowed us to assess metacognitive efficiency separately in a visual and a memory task, while taking variations in basic task performance capacity into account. We found that, across individuals, metacognitive efficiencies positively correlated between the two tasks. However, voxel-based morphometry analysis revealed distinct brain structures for the two kinds of metacognition. Replicating a previous finding, variation in visual metacognitive efficiency was correlated with volume of frontal polar regions. However, variation in memory metacognitive efficiency was correlated with volume of the precuneus. There was also a weak correlation between visual metacognitive efficiency and precuneus volume, which may account for the behavioral correlation between visual and memory metacognition (i.e., the precuneus may contain common mechanisms for both types of metacognition). However, we also found that gray matter volumes of the frontal polar and precuneus regions themselves correlated across individuals, and a formal model comparison analysis suggested that this structural covariation was sufficient to account for the behavioral correlation of metacognition in the two tasks. These results highlight the importance of the precuneus in higher-order memory processing and suggest that there may be functionally distinct metacognitive systems in the human brain.

Late ERP correlates of confidence for auditory categorization of complex sounds.

Recent research suggests that confidence judgments relate to the quality of early sensory representations and later modality independent processing stages. It is not known whether the nature of this finding might vary based on task and/or stimulus characteristics (e.g., detection vs. categorization). The present study investigated the neural correlates of confidence using electroencephalography (EEG) in an auditory categorization task. This allowed us to examine whether the early event-related potentials (ERPs) related to confidence in detection also apply to a more complex auditory task. Participants listened to frequency-modulated (FM) tonal stimuli going up or down in pitch. The rate of FM tones ranged from slow to fast, making the stimuli harder or easier to categorize. Tone-locked late posterior positivity (LPP) but not N1 or P2 amplitudes were larger for (correct-only) trials rated with high than low confidence. These results replicated for trials presenting stimuli at individually identified threshold levels (rate of change producing ∼71.7% correct performance). This finding suggests that, in this task, neural correlates of confidence do not vary based on difficulty level. We suggest that the LPP is a task general indication of the confidence for an upcoming judgment in a variety of paradigms.

Direct injection of noise to the visual cortex decreases accuracy but increases decision confidence.

The relationship between accuracy and confidence in psychophysical tasks traditionally has been assumed to be mainly positive, i.e., the two typically increase or decrease together. However, recent studies have reported examples of exceptions, where confidence and accuracy dissociate from each other. Explanations for such dissociations often involve dual-channel models, in which a cortical channel contributes to both accuracy and confidence, whereas a subcortical channel only contributes to accuracy. Here, we show that a single-channel model derived from signal detection theory (SDT) can also account for such dissociations. We applied transcranial magnetic stimulation (TMS) to the occipital cortex to disrupt the internal representation of a visual stimulus. The results showed that consistent with previous research, occipital TMS decreased accuracy. However, counterintuitively, it also led to an increase in confidence ratings. The data were predicted well by a single-channel SDT model, which posits that occipital TMS increased the variance of the internal stimulus distributions. A formal model comparison analysis that used information theoretic methods confirmed that this model was preferred over single-channel models, in which occipital TMS changed the signal strength or dual-channel models, which assume two different processing routes. Thus our results show that dissociations between accuracy and confidence can, at least in some cases, be accounted for by a single-channel model.

A signal detection theoretic approach for estimating metacognitive sensitivity from confidence ratings.

How should we measure metacognitive ("type 2") sensitivity, i.e. the efficacy with which observers' confidence ratings discriminate between their own correct and incorrect stimulus classifications? We argue that currently available methods are inadequate because they are influenced by factors such as response bias and type 1 sensitivity (i.e. ability to distinguish stimuli). Extending the signal detection theory (SDT) approach of Galvin, Podd, Drga, and Whitmore (2003), we propose a method of measuring type 2 sensitivity that is free from these confounds. We call our measure meta-d', which reflects how much information, in signal-to-noise units, is available for metacognition. Applying this novel method in a 2-interval forced choice visual task, we found that subjects' metacognitive sensitivity was close to, but significantly below, optimality. We discuss the theoretical implications of these findings, as well as related computational issues of the method. We also provide free Matlab code for implementing the analysis.

Awareness-related activity in prefrontal and parietal cortices in blindsight reflects more than superior visual performance.

Many imaging studies report activity in the prefrontal and parietal cortices when subjects are aware as opposed to unaware of visual stimuli. One possibility is that this activity simply reflects higher signal strength or the superior task performance that is associated with awareness. To find out, we studied the hemianope GY who has unilateral destruction of almost all primary visual cortices. He exhibits 'blindsight', that is, he claims to have no conscious visual phenomenology (i.e., no visual qualia), for stationary stimuli presented to his right visual field (the blind field), although he can press keys to distinguish between different stimuli presented there. We presented to him a visual discrimination task, and equated performance for stimuli presented to the left or right visual field by presenting low contrast stimuli to his normal (left) field and high contrast stimuli to his blind (right) field. Superior accuracy can be a serious confound, and our paradigm allows us to control for it and avoid this confound. Even when performance was matched, and the signal strength was lower, visual stimulation to the normal (conscious) field led to higher activity in the prefrontal and parietal cortices. These results indicate that the activity in the prefrontal and parietal areas that has been reported in previous studies of awareness is not just due to a (signal strength or performance) confounds. One possibility is that it reflects the superior 'metacognitive' capacity that is associated with awareness, because GY was better able to distinguish between his own correct and incorrect responses for stimuli presented to his normal field than to his blind field.

Neural integration underlying naturalistic prediction flexibly adapts to varying sensory input rate.

Prediction of future sensory input based on past sensory information is essential for organisms to effectively adapt their behavior in dynamic environments. Humans successfully predict future stimuli in various natural settings. Yet, it remains elusive how the brain achieves effective prediction despite enormous variations in sensory input rate, which directly affect how fast sensory information can accumulate. We presented participants with acoustic sequences capturing temporal statistical regularities prevalent in nature and investigated neural mechanisms underlying predictive computation using MEG. By parametrically manipulating sequence presentation speed, we tested two hypotheses: neural prediction relies on integrating past sensory information over fixed time periods or fixed amounts of information. We demonstrate that across halved and doubled presentation speeds, predictive information in neural activity stems from integration over fixed amounts of information. Our findings reveal the neural mechanisms enabling humans to robustly predict dynamic stimuli in natural environments despite large sensory input rate variations.

Atypical spatial frequency dependence of visual metacognition among schizophrenia patients.

Although altered early stages of visual processing have been reported among schizophrenia patients, how such atypical visual processing may affect higher-level cognition remains largely unknown. Here we tested the hypothesis that metacognitive performance may be atypically modulated by spatial frequency (SF) of visual stimuli among individuals with schizophrenia, given their altered magnocellular function. To study the effect of SF on metacognitive performance, we asked patients and controls to perform a visual detection task on gratings with different SFs and report confidence, and analyzed the data using the signal detection theoretic measure meta-d'. Control subjects showed better metacognitive performance after yes- (stimulus presence) than after no- (stimulus absence) responses ('yes-response advantage') for high SF (HSF) stimuli but not for low SF (LSF) stimuli. The patients, to the contrary, showed a 'yes-response advantage' not only for HSF but also for LSF stimuli, indicating atypical SF dependency of metacognition. An fMRI experiment using the same task revealed that the dorsolateral prefrontal cortex (DLPFC), known to be crucial for metacognition, shows activity mirroring the behavioral results: decoding accuracy of perceptual confidence in DLPFC was significantly higher for HSF than for LSF stimuli in controls, whereas this decoding accuracy was independent of SF in patients. Additionally, the functional connectivity of DLPFC with parietal and visual areas was modulated by SF and response type (yes/no) in a different manner between controls and patients. While individuals without schizophrenia may flexibly adapt metacognitive computations across SF ranges, patients may employ a different mechanism that is independent of SF. Because visual stimuli of low SF have been linked to predictive top-down processing, this may reflect atypical functioning in these processes in schizophrenia.

The Confidence Database.

Understanding how people rate their confidence is critical for the characterization of a wide range of perceptual, memory, motor and cognitive processes. To enable the continued exploration of these processes, we created a large database of confidence studies spanning a broad set of paradigms, participant populations and fields of study. The data from each study are structured in a common, easy-to-use format that can be easily imported and analysed using multiple software packages. Each dataset is accompanied by an explanation regarding the nature of the collected data. At the time of publication, the Confidence Database (which is available at https://osf.io/s46pr/) contained 145 datasets with data from more than 8,700 participants and almost 4 million trials. The database will remain open for new submissions indefinitely and is expected to continue to grow. Here we show the usefulness of this large collection of datasets in four different analyses that provide precise estimations of several foundational confidence-related effects.

The signal processing architecture underlying subjective reports of sensory awareness.

What is the relationship between perceptual information processing and subjective perceptual experience? Empirical dissociations between stimulus identification performance and subjective reports of stimulus visibility are crucial for shedding light on this question. We replicated a finding that metacontrast masking can produce such a dissociation (Lau and Passingham, 2006), and report a novel finding that this paradigm can also dissociate stimulus identification performance from the efficacy with which visibility ratings predict task performance. We explored various hypotheses about the relationship between perceptual task performance and visibility rating by implementing them in computational models and using formal model comparison techniques to assess which ones best captured the unusual patterns in the data. The models fell into three broad categories: Single Channel models, which hold that task performance and visibility ratings are based on the same underlying source of information; Dual Channel models, which hold that there are two independent processing streams that differentially contribute to task performance and visibility rating; and Hierarchical models, which hold that a late processing stage generates visibility ratings by evaluating the quality of early perceptual processing. Taking into account the quality of data fitting and model complexity, we found that Hierarchical models perform best at capturing the observed behavioral dissociations. Because current theories of visual awareness map well onto these different model structures, a formal comparison between them is a powerful approach for arbitrating between the different theories.

Heuristic use of perceptual evidence leads to dissociation between performance and metacognitive sensitivity.

Zylberberg et al. [Zylberberg, Barttfeld, & Sigman (Frontiers in Integrative Neuroscience, 6; 79, 2012), Frontiers in Integrative Neuroscience 6:79] found that confidence decisions, but not perceptual decisions, are insensitive to evidence against a selected perceptual choice. We present a signal detection theoretic model to formalize this insight, which gave rise to a counter-intuitive empirical prediction: that depending on the observer's perceptual choice, increasing task performance can be associated with decreasing metacognitive sensitivity (i.e., the trial-by-trial correspondence between confidence and accuracy). The model also provides an explanation as to why metacognitive sensitivity tends to be less than optimal in actual subjects. These predictions were confirmed robustly in a psychophysics experiment. In a second experiment we found that, in at least some subjects, the effects were replicated even under performance feedback designed to encourage optimal behavior. However, some subjects did show improvement under feedback, suggesting the tendency to ignore evidence against a selected perceptual choice may be a heuristic adopted by the perceptual decision-making system, rather than reflecting inherent biological limitations. We present a Bayesian modeling framework that explains why this heuristic strategy may be advantageous in real-world contexts.

Scale-Free Neural and Physiological Dynamics in Naturalistic Stimuli Processing.

Neural activity recorded at multiple spatiotemporal scales is dominated by arrhythmic fluctuations without a characteristic temporal periodicity. Such activity often exhibits a 1/f-type power spectrum, in which power falls off with increasing frequency following a power-law function: [Formula: see text], which is indicative of scale-free dynamics. Two extensively studied forms of scale-free neural dynamics in the human brain are slow cortical potentials (SCPs)-the low-frequency (<5 Hz) component of brain field potentials-and the amplitude fluctuations of α oscillations, both of which have been shown to carry important functional roles. In addition, scale-free dynamics characterize normal human physiology such as heartbeat dynamics. However, the exact relationships among these scale-free neural and physiological dynamics remain unclear. We recorded simultaneous magnetoencephalography and electrocardiography in healthy subjects in the resting state and while performing a discrimination task on scale-free dynamical auditory stimuli that followed different scale-free statistics. We observed that long-range temporal correlation (captured by the power-law exponent ß) in SCPs positively correlated with that of heartbeat dynamics across time within an individual and negatively correlated with that of α-amplitude fluctuations across individuals. In addition, across individuals, long-range temporal correlation of both SCP and α-oscillation amplitude predicted subjects' discrimination performance in the auditory task, albeit through antagonistic relationships. These findings reveal interrelations among different scale-free neural and physiological dynamics and initial evidence for the involvement of scale-free neural dynamics in the processing of natural stimuli, which often exhibit scale-free dynamics.

Manipulation of working memory contents selectively impairs metacognitive sensitivity in a concurrent visual discrimination task.

What are the cognitive mechanisms underlying perceptual metacognition? Prior research indicates that prefrontal cortex (PFC) contributes to metacognitive performance, suggesting that metacognitive judgments are supported by high-level cognitive operations. We explored this hypothesis by investigating metacognitive performance for a visual discrimination task in the context of a concurrent working memory (WM) task. We found that, overall, high WM load caused a nonspecific decrease in visual discrimination performance as well as metacognitive performance. However, active manipulation of WM contents caused a selective decrease in metacognitive performance without impairing visual discrimination performance. These behavioral findings are consistent with previous neuroscience findings that high-level PFC is engaged by and necessary for (i) visual metacognition, and (ii) active manipulation of WM contents, but not mere maintenance. The selective interference of WM manipulation on metacognition suggests that these seemingly disparate cognitive functions in fact recruit common cognitive mechanisms. The common cognitive underpinning of these tasks may consist in (i) higher-order re-representation of lower-level sensory information, and/or (ii) application of decision rules in order to transform representations in PFC into definite cognitive/motor responses.

Does perceptual confidence facilitate cognitive control?

Our visual perception is typically accompanied by a sense of subjective confidence. Since perceptual confidence is related to prefrontal activity, higher perceptual confidence may enhance cognitive control functions. To examine this interaction, we developed a novel method to selectively manipulate perceptual confidence while keeping stimulus discrimination accuracy constant. In a behavioral experiment, grating stimuli with different orientations were presented as go/no-go signals. Surprisingly, the results showed that confidence in visual discrimination of the signals on its own did not facilitate response inhibition, since when participants were presented with stimuli that yielded higher confidence, they were no better at performing a go/no-go task. These results were replicated with different (dot motion) stimuli, ruling out alternative explanations based on stimulus idiosyncrasy. In a different experiment, when the grating stimuli were presented as cues for task set preparation, we found that higher perceptual confidence also did not enhance task set preparation efficiency. This result was again replicated with dot motion stimuli. Since confidence may relate to perceptual awareness (Peirce & Jastrow, 1885), our findings may put current dominant theories in question, since these theories often suppose the critical involvement of consciousness in cognitive control. As a proof of concept, our method may also provide a new and powerful way to examine other functions of consciousness in future studies.