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.

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.

Superior colliculus neuronal ensemble activity signals optimal rather than subjective confidence.

Recent studies suggest that neurons in sensorimotor circuits involved in perceptual decision-making also play a role in decision confidence. In these studies, confidence is often considered to be an optimal readout of the probability that a decision is correct. However, the information leading to decision accuracy and the report of confidence often covaried, leaving open the possibility that there are actually two dissociable signal types in the brain: signals that correlate with decision accuracy (optimal confidence) and signals that correlate with subjects' behavioral reports of confidence (subjective confidence). We recorded neuronal activity from a sensorimotor decision area, the superior colliculus (SC) of monkeys, while they performed two different tasks. In our first task, decision accuracy and confidence covaried, as in previous studies. In our second task, we implemented a motion discrimination task with stimuli that were matched for decision accuracy but produced different levels of confidence, as reflected by behavioral reports. We used a multivariate decoder to predict monkeys' choices from neuronal population activity. As in previous studies on perceptual decision-making mechanisms, we found that neuronal decoding performance increased as decision accuracy increased. However, when decision accuracy was matched, performance of the decoder was similar between high and low subjective confidence conditions. These results show that the SC likely signals optimal decision confidence similar to previously reported cortical mechanisms, but is unlikely to play a critical role in subjective confidence. The results also motivate future investigations to determine where in the brain signals related to subjective confidence reside.

Social relevance modulates multivariate neural representations of threat generalization in children and adults.

Few studies have examined threat generalization across development and no developmental studies have compared the generalization of social versus nonsocial threat, making it difficult to identify contextual factors that contribute to threat learning across development. The present study assessed youth and adults' multivoxel neural representations of social versus nonsocial threat stimuli. Twenty adults (Mage  = 25.7 ± 4.9) and 16 youth (Mage  = 14.1 ± 1.7) completed two conditioning and extinction recall paradigms: one social and one nonsocial paradigm. Three weeks after conditioning, participants underwent a functional magnetic resonance imaging extinction recall task that presented the extinguished threat cue (CS+), a safety cue (CS-), and generalization stimuli (GS) consisting of CS-/CS+ blends. Across age groups, neural activity patterns and self-reported fear and memory ratings followed a linear generalization gradient for social threat stimuli and a quadratic generalization gradient for nonsocial threat stimuli, indicating enhanced threat/safety discrimination for social relative to nonsocial threat stimuli. The amygdala and ventromedial prefrontal cortex displayed the greatest neural pattern differentiation between the CS+ and GS/CS-, reinforcing their role in threat learning and extinction recall. Contrary to predictions, age did not influence threat representations. These findings highlight the importance of the social relevance of threat on generalization across development.

Anodal Transcranial Direct Current Stimulation to the Left Rostrolateral Prefrontal Cortex Selectively Improves Source Memory Retrieval.

Functional neuroimaging studies have consistently implicated the left rostrolateral prefrontal cortex (RLPFC) as playing a crucial role in the cognitive operations supporting episodic memory and analogical reasoning. However, the degree to which the left RLPFC causally contributes to these processes remains underspecified. We aimed to assess whether targeted anodal stimulation-thought to boost cortical excitability-of the left RLPFC with transcranial direct current stimulation (tDCS) would lead to augmentation of episodic memory retrieval and analogical reasoning task performance in comparison to cathodal stimulation or sham stimulation. Seventy-two healthy adult participants were evenly divided into three experimental groups. All participants performed a memory encoding task on Day 1, and then on Day 2, they performed continuously alternating tasks of episodic memory retrieval, analogical reasoning, and visuospatial perception across two consecutive 30-min experimental sessions. All groups received sham stimulation for the first experimental session, but the groups differed in the stimulation delivered to the left RLPFC during the second session (either sham, 1.5 mA anodal tDCS, or 1.5 mA cathodal tDCS). The experimental group that received anodal tDCS to the left RLPFC during the second session demonstrated significantly improved episodic memory source retrieval performance, relative to both their first session performance and relative to performance changes observed in the other two experimental groups. Performance on the analogical reasoning and visuospatial perception tasks did not exhibit reliable changes as a result of tDCS. As such, our results demonstrate that anodal tDCS to the left RLPFC leads to a selective and robust improvement in episodic source memory retrieval.

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.

Behavioral and neural measures of confidence using a novel auditory pitch identification task.

Observers can discriminate between correct versus incorrect perceptual decisions with feelings of confidence. The centro-parietal positivity build-up rate (CPP slope) has been suggested as a likely neural signature of accumulated evidence, which may guide both perceptual performance and confidence. However, CPP slope also covaries with reaction time, which also covaries with confidence in previous studies, and performance and confidence typically covary; thus, CPP slope may index signatures of perceptual performance rather than confidence per se. Moreover, perceptual metacognition-including neural correlates-has largely been studied in vision, with few exceptions. Thus, we lack understanding of domain-general neural signatures of perceptual metacognition outside vision. Here we designed a novel auditory pitch identification task and collected behavior with simultaneous 32-channel EEG in healthy adults. Participants saw two tone labels which varied in tonal distance on each trial (e.g., C vs D, C vs F), then heard a single auditory tone; they identified which label was correct and rated confidence. We found that pitch identification confidence varied with tonal distance, but performance, metacognitive sensitivity (trial-by-trial covariation of confidence with accuracy), and reaction time did not. Interestingly, however, while CPP slope covaried with performance and reaction time, it did not significantly covary with confidence. We interpret these results to mean that CPP slope is likely a signature of first-order perceptual processing and not confidence-specific signals or computations in auditory tasks. Our novel pitch identification task offers a valuable method to examine the neural correlates of auditory and domain-general perceptual confidence.

Unconscious Perception of Vernier Offsets.

The comparison between conscious and unconscious perception is a cornerstone of consciousness science. However, most studies reporting above-chance discrimination of unseen stimuli do not control for criterion biases when assessing awareness. We tested whether observers can discriminate subjectively invisible offsets of Vernier stimuli when visibility is probed using a bias-free task. To reduce visibility, stimuli were either backward masked or presented for very brief durations (1-3 milliseconds) using a modern-day Tachistoscope. We found some behavioral indicators of perception without awareness, and yet, no conclusive evidence thereof. To seek more decisive proof, we simulated a series of Bayesian observer models, including some that produce visibility judgements alongside type-1 judgements. Our data are best accounted for by observers with slightly suboptimal conscious access to sensory evidence. Overall, the stimuli and visibility manipulations employed here induced mild instances of blindsight-like behavior, making them attractive candidates for future investigation of this phenomenon.

A predictor-informed multi-subject bayesian approach for dynamic functional connectivity.

Dynamic functional connectivity investigates how the interactions among brain regions vary over the course of an fMRI experiment. Such transitions between different individual connectivity states can be modulated by changes in underlying physiological mechanisms that drive functional network dynamics, e.g., changes in attention or cognitive effort. In this paper, we develop a multi-subject Bayesian framework where the estimation of dynamic functional networks is informed by time-varying exogenous physiological covariates that are simultaneously recorded in each subject during the fMRI experiment. More specifically, we consider a dynamic Gaussian graphical model approach where a non-homogeneous hidden Markov model is employed to classify the fMRI time series into latent neurological states. We assume the state-transition probabilities to vary over time and across subjects as a function of the underlying covariates, allowing for the estimation of recurrent connectivity patterns and the sharing of networks among the subjects. We further assume sparsity in the network structures via shrinkage priors, and achieve edge selection in the estimated graph structures by introducing a multi-comparison procedure for shrinkage-based inferences with Bayesian false discovery rate control. We evaluate the performances of our method vs alternative approaches on synthetic data. We apply our modeling framework on a resting-state experiment where fMRI data have been collected concurrently with pupillometry measurements, as a proxy of cognitive processing, and assess the heterogeneity of the effects of changes in pupil dilation on the subjects' propensity to change connectivity states. The heterogeneity of state occupancy across subjects provides an understanding of the relationship between increased pupil dilation and transitions toward different cognitive states.

Tests for consciousness in humans and beyond.

Which systems/organisms are conscious? New tests for consciousness ('C-tests') are urgently needed. There is persisting uncertainty about when consciousness arises in human development, when it is lost due to neurological disorders and brain injury, and how it is distributed in nonhuman species. This need is amplified by recent and rapid developments in artificial intelligence (AI), neural organoids, and xenobot technology. Although a number of C-tests have been proposed in recent years, most are of limited use, and currently we have no C-tests for many of the populations for which they are most critical. Here, we identify challenges facing any attempt to develop C-tests, propose a multidimensional classification of such tests, and identify strategies that might be used to validate them.

Toward 'Computational-Rationality' Approaches to Arbitrating Models of Cognition: A Case Study Using Perceptual Metacognition.

Perceptual confidence results from a metacognitive process which evaluates how likely our percepts are to be correct. Many competing models of perceptual metacognition enjoy strong empirical support. Arbitrating these models traditionally proceeds via researchers conducting experiments and then fitting several models to the data collected. However, such a process often includes conditions or paradigms that may not best arbitrate competing models: Many models make similar predictions under typical experimental conditions. Consequently, many experiments are needed, collectively (sub-optimally) sampling the space of conditions to compare models. Here, instead, we introduce a variant of optimal experimental design which we call a computational-rationality approach to generative models of cognition, using perceptual metacognition as a case study. Instead of designing experiments and post-hoc specifying models, we began with comprehensive model comparison among four competing generative models for perceptual metacognition, drawn from literature. By simulating a simple experiment under each model, we identified conditions where these models made maximally diverging predictions for confidence. We then presented these conditions to human observers, and compared the models' capacity to predict choices and confidence. Results revealed two surprising findings: (1) two models previously reported to differently predict confidence to different degrees, with one predicting better than the other, appeared to predict confidence in a direction opposite to previous findings; and (2) two other models previously reported to equivalently predict confidence showed stark differences in the conditions tested here. Although preliminary with regards to which model is actually 'correct' for perceptual metacognition, our findings reveal the promise of this computational-rationality approach to maximizing experimental utility in model arbitration while minimizing the number of experiments necessary to reveal the winning model, both for perceptual metacognition and in other domains.

A Novel Hidden Markov Approach to Studying Dynamic Functional Connectivity States in Human Neuroimaging.

Introduction: Hidden Markov models (HMMs) are a popular choice to extract and examine recurring patterns of activity or functional connectivity in neuroimaging data, both in terms of spatial patterns and their temporal progression. Although many diverse HMMs have been applied to neuroimaging data, most have defined states based on activity levels (intensity-based [IB] states) rather than patterns of functional connectivity between brain areas (connectivity-based states), which is problematic if we want to understand connectivity dynamics: IB states are unlikely to provide comprehensive information about dynamic connectivity patterns. Methods: We addressed this problem by introducing a new HMM that defines states based on full functional connectivity (FFC) profiles among brain regions. We empirically explored the behavior of this new model in comparison to existing approaches based on IB or summed functional connectivity states using the Human Connectome Project unrelated 100 functional magnetic resonance imaging "resting-state" dataset. Results: Our FFC model discovered connectivity states with more distinguishable (i.e., unique and separable from each other) patterns than previous approaches, and recovered simulated connectivity-based states more faithfully than the other models tested. Discussion: Thus, if our goal is to extract and interpret connectivity states in neuroimaging data, our new model outperforms previous methods, which miss crucial information about the evolution of functional connectivity in the brain. Impact statement Hidden Markov models (HMMs) can be used to investigate brain states noninvasively. Previous models "recover" connectivity from intensity-based hidden states, or from connectivity "summed" across nodes. In this study, we introduce a novel connectivity-based HMM and show how it can reveal true connectivity hidden states under minimal assumptions.

Locus coeruleus integrity predicts ease of attaining and maintaining neural states of high attentiveness.

The locus coeruleus (LC), a small subcortical structure in the brainstem, is the brain's principal source of norepinephrine. It plays a primary role in regulating stress, the sleep-wake cycle, and attention, and its degradation is associated with aging and neurodegenerative diseases associated with cognitive deficits (e.g., Parkinson's, Alzheimer's). Yet precisely how norepinephrine drives brain networks to support healthy cognitive function remains poorly understood - partly because LC's small size makes it difficult to study noninvasively in humans. Here, we characterized LC's influence on brain dynamics using a hidden Markov model fitted to functional neuroimaging data from healthy young adults across four attention-related brain networks and LC. We modulated LC activity using a behavioral paradigm and measured individual differences in LC magnetization transfer contrast. The model revealed five hidden states, including a stable state dominated by salience-network activity that occurred when subjects actively engaged with the task. LC magnetization transfer contrast correlated with this state's stability across experimental manipulations and with subjects' propensity to enter into and remain in this state. These results provide new insight into LC's role in driving spatiotemporal neural patterns associated with attention, and demonstrate that variation in LC integrity can explain individual differences in these patterns even in healthy young adults.

Consciousness beyond the human case.

There is growing interest in the relationship been AI and consciousness. Joseph LeDoux and Jonathan Birch thought it would be a good moment to put some of the big questions in this area to some leading experts. The challenge of addressing the questions they raised was taken up by Kristin Andrews, Nicky Clayton, Nathaniel Daw, Chris Frith, Hakwan Lau, Megan Peters, Susan Schneider, Anil Seth, Thomas Suddendorf, and Marie Vanderkerckhoeve.

0 + 1 > 1: How adding noninformative sound improves performance on a visual task.

It is well known that the nervous system combines information from different cues within and across sensory modalities to improve performance on perceptual tasks. In this article, we present results showing that in a visual motion-detection task, concurrent auditory motion stimuli improve accuracy even when they do not provide any useful information for the task. When participants judged which of two stimulus intervals contained visual coherent motion, the addition of identical moving sounds to both intervals improved accuracy. However, this enhancement occurred only with sounds that moved in the same direction as the visual motion. Therefore, it appears that the observed benefit of auditory stimulation is due to auditory-visual interactions at a sensory level. Thus, auditory and visual motion-processing pathways interact at a sensory-representation level in addition to the level at which perceptual estimates are combined.

Towards characterizing the canonical computations generating phenomenal experience.

Science and philosophy have long struggled with how to even begin studying the neural or computational basis of qualitative experience. Here I review psychological, neuroscience, and philosophical literature to reveal how perceptual metacognition possesses five unique properties that provide a powerful opportunity for studying the neural and computational correlates of subjective experience: (1) Metacognition leads to subjective experiences (we "feel" confident); (2) Metacognition is "about" internal representations, formalizing introspection; (3) Metacognitive computations are "recursive" (applying to meta-cognition and meta-meta-cognition), so we might discover "canonical computations" preserved across processing levels and implementations; (4) Metacognition is anchored to observable behavior; and (5) Metacognitive computations are unobservable yet hierarchically dependent, requiring development of sensitive, specific models. Given these properties, computational models of metacognition provide an empirically-tractable early step in characterizing the generative process that constructs qualitative experience. I also present practical ways to make progress in this vein, applying decades of developments in nearby fields to perceptual metacognition to reveal new and exciting insights about how the brain constructs subjective conscious experiences.

Variance misperception under skewed empirical noise statistics explains overconfidence in the visual periphery.

Perceptual confidence typically corresponds to accuracy. However, observers can be overconfident relative to accuracy, termed "subjective inflation." Inflation is stronger in the visual periphery relative to central vision, especially under conditions of peripheral inattention. Previous literature suggests inflation stems from errors in estimating noise (i.e., "variance misperception"). However, despite previous Bayesian hypotheses about metacognitive noise estimation, no work has systematically explored how noise estimation may critically depend on empirical noise statistics, which may differ across the visual field, with central noise distributed symmetrically but peripheral noise positively skewed. Here, we examined central and peripheral vision predictions from five Bayesian-inspired noise-estimation algorithms under varying usage of noise priors, including effects of attention. Models that failed to optimally estimate noise exhibited peripheral inflation, but only models that explicitly used peripheral noise priors-but used them incorrectly-showed increasing peripheral inflation under increasing peripheral inattention. Further, only one model successfully captured previous empirical results, which showed a selective increase in confidence in incorrect responses under performance reductions due to inattention accompanied by no change in confidence in correct responses; this was the model that implemented Bayesian estimation of peripheral noise, but using an (incorrect) symmetric rather than the correct positively skewed peripheral noise prior. Our findings explain peripheral inflation, especially under inattention, and suggest future experiments that might reveal the noise expectations used by the visual metacognitive system.

Interaction of prior category knowledge and novel statistical patterns during visual search for real-world objects.

Two aspects of real-world visual search are typically studied in parallel: category knowledge (e.g., searching for food) and visual patterns (e.g., predicting an upcoming street sign from prior street signs). Previous visual search studies have shown that prior category knowledge hinders search when targets and distractors are from the same category. Other studies have shown that task-irrelevant patterns of non-target objects can enhance search when targets appear in locations that previously contained these irrelevant patterns. Combining EEG (N2pc ERP component, a neural marker of target selection) and behavioral measures, the present study investigated how search efficiency is simultaneously affected by prior knowledge of real-world objects (food and toys) and irrelevant visual patterns (sequences of runic symbols) within the same paradigm. We did not observe behavioral differences between locating items in patterned versus random locations. However, the N2pc components emerged sooner when search items appeared in the patterned location, compared to the random location, with a stronger effect when search items were targets, as opposed to non-targets categorically related to the target. A multivariate pattern analysis revealed that neural responses during search trials in the same time window reflected where the visual patterns appeared. Our finding contributes to our understanding of how knowledge acquired prior to the search task (e.g., category knowledge) interacts with new content within the search task.

Comparing random dot motion in MATLAB vs. Inquisit Millisecond.

Random Dot Motion (RDM) displays refer to clouds of independently moving dots that can be parametrically manipulated to provide a perception of the overall cloud moving coherently in a specified direction of motion. As a well-studied probe of motion perception, RDMs have been widely employed to understand underlying neural mechanisms of motion perception, perceptual decision-making, and perceptual learning, among other processes. Despite their wide use, RDM stimuli implementation is highly dependent on the parameters and the generation algorithm of the stimuli; both can greatly influence behavioral performance on RDM tasks. With the advent of the COVID pandemic and an increased need for more accessible platforms, we aimed to validate a novel RDM paradigm on Inquisit Millisecond, a platform for the online administration of cognitive and neuropsychological tests and assessments. We directly compared, in the same participants using the same display, a novel RDM paradigm on both Inquisit Millisecond and MATLAB with Psychtoolbox. We found that psychometric functions of Coherence largely match between Inquisit Millisecond and MATLAB, as do the effects of Duration. These data demonstrate that the Millisecond RDM provides data largely consistent with those previously found in laboratory-based systems, and the present findings can serve as a reference point for expected thresholds for when these procedures are used remotely on different platforms.

An investigation of how relative precision of target encoding influences metacognitive performance.

Detection failures in perceptual tasks can result from different causes: sometimes we may fail to see something because perceptual information is noisy or degraded, and sometimes we may fail to see something due to the limited capacity of attention. Previous work indicates that metacognitive capacities for detection failures may differ depending on the specific stimulus visibility manipulation employed. In this investigation, we measured metacognition while matching performance in two visibility manipulations: phase-scrambling and the attentional blink. As in previous work, metacognitive asymmetries emerged: despite matched type 1 performance, metacognitive ability (measured by area under the ROC curve) for reporting stimulus absence was higher in the attentional blink condition, which was mainly driven by metacognitive ability in correct rejection trials. We performed Signal Detection Theoretic (SDT) modeling of the results, showing that differences in metacognition under equal type I performance can be explained when the variance of the signal and noise distributions are unequal. Specifically, the present study suggests that phase scrambling signal trials have a wider distribution (more variability) than attentional blink signal trials, leading to a larger area under the ROC curve for attentional blink trials where subjects reported stimulus absence. These results provide a theoretical basis for the origin of metacognitive differences on trials where subjects report stimulus absence, and may also explain previous findings where the absence of evidence during detection tasks results in lower metacognitive performance when compared to categorization.