Confidence resets reveal hierarchical adaptive learning in humans

Learning and metacognition under volatility in GD: Lower learning rates and distorted coupling between action and confidence

Background and aims

Decisions and learning processes are under metacognitive control, where confidence in one's actions guides future behaviour. Indeed, studies have shown that being more confident results in less action updating and learning, and vice versa. This coupling between action and confidence can be disrupted, as has been found in individuals with high compulsivity symptoms. Patients with Gambling Disorder (GD) have been shown to exhibit both higher confidence and deficits in learning.

Methods

In this study, we tested the hypotheses that patients with GD display increased confidence, reduced action updating and lower learning rates. Additionally, we investigated whether the action-confidence coupling was distorted in patients with GD. To address this, 27 patients with GD and 30 control participants performed a predictive inference task designed to assess action and confidence dynamics during learning under volatility. Action-updating, confidence and their coupling were assessed and computational modeling estimated parameters for learning rates, error sensitivity, and sensitivity to environmental changes.

Results

Contrary to our expectations, results revealed no significant group differences in action updating or confidence levels. Nevertheless, GD patients exhibited a weakened coupling between confidence and action, as well as lower learning rates.

Discussion and conclusions

This suggests that patients with GD may underutilize confidence when steering future behavioral choices. Ultimately, these findings point to a disruption of metacognitive control in GD, without a general overconfidence bias in neutral, non-incentivized volatile learning contexts.

Neural and computational underpinnings of biased confidence in human reinforcement learning

While navigating a fundamentally uncertain world, humans and animals constantly evaluate the probability of their decisions, actions or statements being correct. When explicitly elicited, these confidence estimates typically correlates positively with neural activity in a ventromedial-prefrontal (VMPFC) network and negatively in a dorsolateral and dorsomedial prefrontal network. Here, combining fMRI with a reinforcement-learning paradigm, we leverage the fact that humans are more confident in their choices when seeking gains than avoiding losses to reveal a functional dissociation: whereas the dorsal prefrontal network correlates negatively with a condition-specific confidence signal, the VMPFC network positively encodes task-wide confidence signal incorporating the valence-induced bias. Challenging dominant neuro-computational models, we found that decision-related VMPFC activity better correlates with confidence than with option-values inferred from reinforcement-learning models. Altogether, these results identify the VMPFC as a key node in the neuro-computational architecture that builds global feeling-of-confidence signals from latent decision variables and contextual biases during reinforcement-learning.

Probabilistic integration of preceding responses explains response bias in perceptual decision making

Expectations of sensory information change not only how well but also what we perceive. Even in an unpredictable environment, the brain is by default constantly engaged in computing probabilities between sensory events. These estimates are used to generate predictions about future sensory events. Here, we investigated the predictability of behavioral responses using three different learning models in three different one-interval two-alternative forced choice experiments with either auditory, vestibular, or visual stimuli. Results indicate that recent decisions, instead of the sequence of generative stimuli, cause serial dependence. By bridging the gap between sequence learning and perceptual decision making, we provide a novel perspective on sequential choice effects. We propose that serial biases reflect the tracking of statistical regularities of the decision variable, offering a broader understanding of this phenomenon.

From fear of falling to choking under pressure: A predictive processing perspective of disrupted motor control under anxiety

Under the Predictive Processing Framework, perception is guided by internal models that map the probabilistic relationship between sensory states and their causes. Predictive processing has contributed to a new understanding of both emotional states and motor control but is yet to be fully applied to their interaction during the breakdown of motor movements under heightened anxiety or threat. We bring together literature on anxiety and motor control to propose that predictive processing provides a unifying principle for understanding motor breakdowns as a disruption to the neuromodulatory control mechanisms that regulate the interactions of top-down predictions and bottom-up sensory signals. We illustrate this account using examples from disrupted balance and gait in populations who are anxious/fearful of falling, as well as 'choking' in elite sport. This approach can explain both rigid and inflexible movement strategies, as well as highly variable and imprecise action and conscious movement processing, and may also unite the apparently opposing self-focus and distraction approaches to choking. We generate predictions to guide future work and propose practical recommendations.

A characterization of the neural representation of confidence during probabilistic learning

Learning in a stochastic and changing environment is a difficult task. Models of learning typically postulate that observations that deviate from the learned predictions are surprising and used to update those predictions. Bayesian accounts further posit the existence of a confidence-weighting mechanism: learning should be modulated by the confidence level that accompanies those predictions. However, the neural bases of this confidence are much less known than the ones of surprise. Here, we used a dynamic probability learning task and high-field MRI to identify putative cortical regions involved in the representation of confidence about predictions during human learning. We devised a stringent test based on the conjunction of four criteria. We localized several regions in parietal and frontal cortices whose activity is sensitive to the confidence of an ideal observer, specifically so with respect to potential confounds (surprise and predictability), and in a way that is invariant to which item is predicted. We also tested for functionality in two ways. First, we localized regions whose activity patterns at the subject level showed an effect of both confidence and surprise in qualitative agreement with the confidence-weighting principle. Second, we found neural representations of ideal confidence that also accounted for subjective confidence. Taken together, those results identify a set of cortical regions potentially implicated in the confidence-weighting of learning.

Task-evoked pupillary responses track precision-weighted prediction errors and learning rate during interceptive visuomotor actions

In this study, we examined the relationship between physiological encoding of surprise and the learning of anticipatory eye movements. Active inference portrays perception and action as interconnected inference processes, driven by the imperative to minimise the surprise of sensory observations. To examine this characterisation of oculomotor learning during a hand-eye coordination task, we tested whether anticipatory eye movements were updated in accordance with Bayesian principles and whether trial-by-trial learning rates tracked pupil dilation as a marker of 'surprise'. Forty-four participants completed an interception task in immersive virtual reality that required them to hit bouncing balls that had either expected or unexpected bounce profiles. We recorded anticipatory eye movements known to index participants' beliefs about likely ball bounce trajectories. By fitting a hierarchical Bayesian inference model to the trial-wise trajectories of these predictive eye movements, we were able to estimate each individual's expectations about bounce trajectories, rates of belief updating, and precision-weighted prediction errors. We found that the task-evoked pupil response tracked prediction errors and learning rates but not beliefs about ball bounciness or environmental volatility. These findings are partially consistent with active inference accounts and shed light on how encoding of surprise may shape the control of action.

Human inference reflects a normative balance of complexity and accuracy

We must often infer latent properties of the world from noisy and changing observations. Complex, probabilistic approaches to this challenge such as Bayesian inference are accurate but cognitively demanding, relying on extensive working memory and adaptive processing. Simple heuristics are easy to implement but may be less accurate. What is the appropriate balance between complexity and accuracy? Here we model a hierarchy of strategies of variable complexity and find a power law of diminishing returns: increasing complexity gives progressively smaller gains in accuracy. The rate of diminishing returns depends systematically on the statistical uncertainty in the world, such that complex strategies do not provide substantial benefits over simple ones when uncertainty is either too high or too low. In between, there is a complexity dividend. In two psychophysical experiments, we confirm specific model predictions about how working memory and adaptivity should be modulated by uncertainty.

Motivational signals disrupt metacognitive signals in the human ventromedial prefrontal cortex

A growing body of evidence suggests that, during decision-making, BOLD signal in the ventromedial prefrontal cortex (VMPFC) correlates both with motivational variables – such as incentives and expected values – and metacognitive variables – such as confidence judgments – which reflect the subjective probability of being correct. At the behavioral level, we recently demonstrated that the value of monetary stakes bias confidence judgments, with gain (respectively loss) prospects increasing (respectively decreasing) confidence judgments, even for similar levels of difficulty and performance. If and how this value-confidence interaction is reflected in the VMPFC remains unknown. Here, we used an incentivized perceptual decision-making fMRI task that dissociates key decision-making variables, thereby allowing to test several hypotheses about the role of the VMPFC in the value-confidence interaction. While our initial analyses seemingly indicate that the VMPFC combines incentives and confidence to form an expected value signal, we falsified this conclusion with a meticulous dissection of qualitative activation patterns. Rather, our results show that strong VMPFC confidence signals observed in trials with gain prospects are disrupted in trials with no – or negative (loss) – monetary prospects. Deciphering how decision variables are represented and interact at finer scales seems necessary to better understand biased (meta)cognition.

The human ventromedial prefrontal cortex helps to determine value and confidence in certain decisions, but only in situations when there is a potential for a (monetary) reward.

Models of Dynamic Belief Updating in Psychosis-A Review Across Different Computational Approaches

To understand the dysfunctional mechanisms underlying maladaptive reasoning of psychosis, computational models of decision making have widely been applied over the past decade. Thereby, a particular focus has been on the degree to which beliefs are updated based on new evidence, expressed by the learning rate in computational models. Higher order beliefs about the stability of the environment can determine the attribution of meaningfulness to events that deviate from existing beliefs by interpreting these either as noise or as true systematic changes (volatility). Both, the inappropriate downplaying of important changes as noise (belief update too low) as well as the overly flexible adaptation to random events (belief update too high) were theoretically and empirically linked to symptoms of psychosis. Whereas models with fixed learning rates fail to adjust learning in reaction to dynamic changes, increasingly complex learning models have been adopted in samples with clinical and subclinical psychosis lately. These ranged from advanced reinforcement learning models, over fully Bayesian belief updating models to approximations of fully Bayesian models with hierarchical learning or change point detection algorithms. It remains difficult to draw comparisons across findings of learning alterations in psychosis modeled by different approaches e.g., the Hierarchical Gaussian Filter and change point detection. Therefore, this review aims to summarize and compare computational definitions and findings of dynamic belief updating without perceptual ambiguity in (sub)clinical psychosis across these different mathematical approaches. There was strong heterogeneity in tasks and samples. Overall, individuals with schizophrenia and delusion-proneness showed lower behavioral performance linked to failed differentiation between uninformative noise and environmental change. This was indicated by increased belief updating and an overestimation of volatility, which was associated with cognitive deficits. Correlational evidence for computational mechanisms and positive symptoms is still sparse and might diverge from the group finding of instable beliefs. Based on the reviewed studies, we highlight some aspects to be considered to advance the field with regard to task design, modeling approach, and inclusion of participants across the psychosis spectrum. Taken together, our review shows that computational psychiatry offers powerful tools to advance our mechanistic insights into the cognitive anatomy of psychotic experiences.

Gated recurrence enables simple and accurate sequence prediction in stochastic, changing, and structured environments

From decision making to perception to language, predicting what is coming next is crucial. It is also challenging in stochastic, changing, and structured environments; yet the brain makes accurate predictions in many situations. What computational architecture could enable this feat? Bayesian inference makes optimal predictions but is prohibitively difficult to compute. Here, we show that a specific recurrent neural network architecture enables simple and accurate solutions in several environments. This architecture relies on three mechanisms: gating, lateral connections, and recurrent weight training. Like the optimal solution and the human brain, such networks develop internal representations of their changing environment (including estimates of the environment’s latent variables and the precision of these estimates), leverage multiple levels of latent structure, and adapt their effective learning rate to changes without changing their connection weights. Being ubiquitous in the brain, gated recurrence could therefore serve as a generic building block to predict in real-life environments.

Predictive eye movements are adjusted in a Bayes-optimal fashion in response to unexpectedly changing environmental probabilities

This study examined the application of active inference to dynamic visuomotor control. Active inference proposes that actions are dynamically planned according to uncertainty about sensory information, prior expectations, and the environment, with motor adjustments serving to minimise future prediction errors. We investigated whether predictive gaze behaviours are indeed adjusted in this Bayes-optimal fashion during a virtual racquetball task. In this task, participants intercepted bouncing balls with varying levels of elasticity, under conditions of higher or lower environmental volatility. Participants' gaze patterns differed between stable and volatile conditions in a manner consistent with generative models of Bayes-optimal behaviour. Partially observable Markov models also revealed an increased rate of associative learning in response to unpredictable shifts in environmental probabilities, although there was no overall effect of volatility on this parameter. Findings extend active inference frameworks into complex and unconstrained visuomotor tasks and present important implications for a neurocomputational understanding of the visual guidance of action.

Novelty is not surprise: Human exploratory and adaptive behavior in sequential decision-making

Classic reinforcement learning (RL) theories cannot explain human behavior in the absence of external reward or when the environment changes. Here, we employ a deep sequential decision-making paradigm with sparse reward and abrupt environmental changes. To explain the behavior of human participants in these environments, we show that RL theories need to include surprise and novelty, each with a distinct role. While novelty drives exploration before the first encounter of a reward, surprise increases the rate of learning of a world-model as well as of model-free action-values. Even though the world-model is available for model-based RL, we find that human decisions are dominated by model-free action choices. The world-model is only marginally used for planning, but it is important to detect surprising events. Our theory predicts human action choices with high probability and allows us to dissociate surprise, novelty, and reward in EEG signals.

Neural surprise in somatosensory Bayesian learning

Tracking statistical regularities of the environment is important for shaping human behavior and perception. Evidence suggests that the brain learns environmental dependencies using Bayesian principles. However, much remains unknown about the employed algorithms, for somesthesis in particular. Here, we describe the cortical dynamics of the somatosensory learning system to investigate both the form of the generative model as well as its neural surprise signatures. Specifically, we recorded EEG data from 40 participants subjected to a somatosensory roving-stimulus paradigm and performed single-trial modeling across peri-stimulus time in both sensor and source space. Our Bayesian model selection procedure indicates that evoked potentials are best described by a non-hierarchical learning model that tracks transitions between observations using leaky integration. From around 70ms post-stimulus onset, secondary somatosensory cortices are found to represent confidence-corrected surprise as a measure of model inadequacy. Indications of Bayesian surprise encoding, reflecting model updating, are found in primary somatosensory cortex from around 140ms. This dissociation is compatible with the idea that early surprise signals may control subsequent model update rates. In sum, our findings support the hypothesis that early somatosensory processing reflects Bayesian perceptual learning and contribute to an understanding of its underlying mechanisms.

Our environment features statistical regularities, such as a drop of rain predicting imminent rainfall. Despite the importance for behavior and survival, much remains unknown about how these dependencies are learned, particularly for somatosensation. As surprise signalling about novel observations indicates a mismatch between one’s beliefs and the world, it has been hypothesized that surprise computation plays an important role in perceptual learning. By analyzing EEG data from human participants receiving sequences of tactile stimulation, we compare different formulations of surprise and investigate the employed underlying learning model. Our results indicate that the brain estimates transitions between observations. Furthermore, we identified different signatures of surprise computation and thereby provide a dissociation of the neural correlates of belief inadequacy and belief updating. Specifically, early surprise responses from around 70ms were found to signal the need for changes to the model, with encoding of its subsequent updating occurring from around 140ms. These results provide insights into how somatosensory surprise signals may contribute to the learning of environmental statistics.

Learning in Volatile Environments With the Bayes Factor Surprise

Surprise-based learning allows agents to rapidly adapt to nonstationary stochastic environments characterized by sudden changes. We show that exact Bayesian inference in a hierarchical model gives rise to a surprise-modulated trade-off between forgetting old observations and integrating them with the new ones. The modulation depends on a probability ratio, which we call the Bayes Factor Surprise, that tests the prior belief against the current belief. We demonstrate that in several existing approximate algorithms, the Bayes Factor Surprise modulates the rate of adaptation to new observations. We derive three novel surprise-based algorithms, one in the family of particle filters, one in the family of variational learning, and one in the family of message passing, that have constant scaling in observation sequence length and particularly simple update dynamics for any distribution in the exponential family. Empirical results show that these surprise-based algorithms estimate parameters better than alternative approximate approaches and reach levels of performance comparable to computationally more expensive algorithms. The Bayes Factor Surprise is related to but different from the Shannon Surprise. In two hypothetical experiments, we make testable predictions for physiological indicators that dissociate the Bayes Factor Surprise from the Shannon Surprise. The theoretical insight of casting various approaches as surprise-based learning, as well as the proposed online algorithms, may be applied to the analysis of animal and human behavior and to reinforcement learning in nonstationary environments.

Using the past to estimate sensory uncertainty

To form a more reliable percept of the environment, the brain needs to estimate its own sensory uncertainty. Current theories of perceptual inference assume that the brain computes sensory uncertainty instantaneously and independently for each stimulus. We evaluated this assumption in four psychophysical experiments, in which human observers localized auditory signals that were presented synchronously with spatially disparate visual signals. Critically, the visual noise changed dynamically over time continuously or with intermittent jumps. Our results show that observers integrate audiovisual inputs weighted by sensory uncertainty estimates that combine information from past and current signals consistent with an optimal Bayesian learner that can be approximated by exponential discounting. Our results challenge leading models of perceptual inference where sensory uncertainty estimates depend only on the current stimulus. They demonstrate that the brain capitalizes on the temporal dynamics of the external world and estimates sensory uncertainty by combining past experiences with new incoming sensory signals.

Robust valence-induced biases on motor response and confidence in human reinforcement learning

In simple instrumental-learning tasks, humans learn to seek gains and to avoid losses equally well. Yet, two effects of valence are observed. First, decisions in loss-contexts are slower. Second, loss contexts decrease individuals’ confidence in their choices. Whether these two effects are two manifestations of a single mechanism or whether they can be partially dissociated is unknown. Across six experiments, we attempted to disrupt the valence-induced motor bias effects by manipulating the mapping between decisions and actions and imposing constraints on response times (RTs). Our goal was to assess the presence of the valence-induced confidence bias in the absence of the RT bias. We observed both motor and confidence biases despite our disruption attempts, establishing that the effects of valence on motor and metacognitive responses are very robust and replicable. Nonetheless, within- and between-individual inferences reveal that the confidence bias resists the disruption of the RT bias. Therefore, although concomitant in most cases, valence-induced motor and confidence biases seem to be partly dissociable. These results highlight new important mechanistic constraints that should be incorporated in learning models to jointly explain choice, reaction times and confidence.

Brain dynamics for confidence-weighted learning

Learning in a changing, uncertain environment is a difficult problem. A popular solution is to predict future observations and then use surprising outcomes to update those predictions. However, humans also have a sense of confidence that characterizes the precision of their predictions. Bayesian models use a confidence-weighting principle to regulate learning: for a given surprise, the update is smaller when the confidence about the prediction was higher. Prior behavioral evidence indicates that human learning adheres to this confidence-weighting principle. Here, we explored the human brain dynamics sub-tending the confidence-weighting of learning using magneto-encephalography (MEG). During our volatile probability learning task, subjects’ confidence reports conformed with Bayesian inference. MEG revealed several stimulus-evoked brain responses whose amplitude reflected surprise, and some of them were further shaped by confidence: surprise amplified the stimulus-evoked response whereas confidence dampened it. Confidence about predictions also modulated several aspects of the brain state: pupil-linked arousal and beta-range (15–30 Hz) oscillations. The brain state in turn modulated specific stimulus-evoked surprise responses following the confidence-weighting principle. Our results thus indicate that there exist, in the human brain, signals reflecting surprise that are dampened by confidence in a way that is appropriate for learning according to Bayesian inference. They also suggest a mechanism for confidence-weighted learning: confidence about predictions would modulate intrinsic properties of the brain state to amplify or dampen surprise responses evoked by discrepant observations.

Learning in a changing and uncertain world is difficult. In this context, facing a discrepancy between my current belief and new observations may reflect random fluctuations (e.g. my commute train is unexpectedly late, but it happens sometimes), if so, I should ignore this discrepancy and not change erratically my belief. However, this discrepancy could also denote a profound change (e.g. the train company changed and is less reliable), in this case, I should promptly revise my current belief. Human learning is adaptive: we change how much we learn from new observations, in particular, we promote flexibility when facing profound changes. A mathematical analysis of the problem shows that we should increase flexibility when the confidence about our current belief is low, which occurs when a change is suspected. Here, I show that human learners entertain rational confidence levels during the learning of changing probabilities. This confidence modulates intrinsic properties of the brain state (oscillatory activity and neuromodulation) which in turn amplifies or reduces, depending on whether confidence is low or high, the neural responses to discrepant observations. This confidence-weighting mechanism could underpin adaptive learning.

Eye pupil signals information gain

In conditions of constant illumination, the eye pupil diameter indexes the modulation of arousal state and responds to a large breadth of cognitive processes, including mental effort, attention, surprise, decision processes, decision biases, value beliefs, uncertainty, volatility, exploitation/exploration trade-off, or learning rate. Here, I propose an information theoretic framework that has the potential to explain the ensemble of these findings as reflecting pupillary response to information processing. In short, updates of the brain’s internal model, quantified formally as the Kullback–Leibler (KL) divergence between prior and posterior beliefs, would be the common denominator to all these instances of pupillary dilation to cognition. I show that stimulus presentation leads to pupillary response that is proportional to the amount of information the stimulus carries about itself and to the quantity of information it provides about other task variables. In the context of decision making, pupil dilation in relation to uncertainty is explained by the wandering of the evidence accumulation process, leading to large summed KL divergences. Finally, pupillary response to mental effort and variations in tonic pupil size are also formalized in terms of information theory. On the basis of this framework, I compare pupillary data from past studies to simple information-theoretic simulations of task designs and show good correspondance with data across studies. The present framework has the potential to unify the large set of results reported on pupillary dilation to cognition and to provide a theory to guide future research.

Human online adaptation to changes in prior probability

Optimal sensory decision-making requires the combination of uncertain sensory signals with prior expectations. The effect of prior probability is often described as a shift in the decision criterion. Can observers track sudden changes in probability? To answer this question, we used a change-point detection paradigm that is frequently used to examine behavior in changing environments. In a pair of orientation-categorization tasks, we investigated the effects of changing probabilities on decision-making. In both tasks, category probability was updated using a sample-and-hold procedure: probability was held constant for a period of time before jumping to another probability state that was randomly selected from a predetermined set of probability states. We developed an ideal Bayesian change-point detection model in which the observer marginalizes over both the current run length (i.e., time since last change) and the current category probability. We compared this model to various alternative models that correspond to different strategies—from approximately Bayesian to simple heuristics—that the observers may have adopted to update their beliefs about probabilities. While a number of models provided decent fits to the data, model comparison favored a model in which probability is estimated following an exponential averaging model with a bias towards equal priors, consistent with a conservative bias, and a flexible variant of the Bayesian change-point detection model with incorrect beliefs. We interpret the former as a simpler, more biologically plausible explanation suggesting that the mechanism underlying change of decision criterion is a combination of on-line estimation of prior probability and a stable, long-term equal-probability prior, thus operating at two very different timescales.

We demonstrate how people learn and adapt to changes to the probability of occurrence of one of two categories on decision-making under uncertainty. The study combined psychophysical behavioral tasks with computational modeling. We used two behavioral tasks: a typical forced-choice categorization task as well as one in which the observer specified the decision criterion to use on each trial before the stimulus was displayed. We formulated an ideal Bayesian change-point detection model and compared it to several alternative models. We found that the data are explained best by a model that estimates category probability based on recently observed exemplars with a bias towards equal probability. Our results suggest that the brain takes multiple relevant time scales into account when setting category expectations.