Psychological Review

Bayesian decision theory proposes that people make statistically rational decisions by combining prior knowledge with sensory information (likelihoods). This framework successfully explains many aspects of human behaviour. However, debate persists over whether people perform precise Bayesian computations (i.e., explicit Bayesian strategy) or rely on less demanding strategies - such as approximations or heuristics - that produce Bayesian-like behaviour (i.e., implicit Bayesian strategy). To address this, we examined people's sensitivity to metamers: different prior-likelihood combinations yielding identical optimal policies. An explicit Bayesian observer would show a temporary performance drop immediately after a switch of prior-likelihood combination, followed by recovery, reflecting prior updating. In two studies, we trained participants to estimate hidden target locations drawn from a Gaussian prior. On each trial, scattered dots provided likelihood information. Over time, participants learned the prior and combined it with likelihood information to infer target locations. We then covertly introduced an untrained prior-likelihood metamer. Unlike explicit Bayesian observers, participants' performance declined after the switch and persisted throughout the untrained pair presentation. This finding challenges strict Bayesian interpretations of task performance and suggests that participants rely instead on likelihood-sensitive strategy that is neither explicit Bayesian nor does it not fully integrate prior information. Our study demonstrates how metamer manipulations can distinguish behaviour that merely appears Bayesian, from behaviour genuinely produced by Bayesian computations, and calls for the use of metamers for ruling out alternative explanations of Bayesian-like behaviours.

Previous work showed that observers integrate audiovisual duration cues optimally when cue-conflict is small. Does causal inference lead to a breakdown of audiovisual integration when duration conflicts are large? We addressed this by testing a wide range of duration cue-conflicts. Participants compared the auditory durations of a test and a standard stimulus. Audiovisual durations were consistent in the test stimulus, but differed by seven conflict durations (up to 250 ms) in the standard. Two levels of auditory noise were tested. Auditory duration percepts shifted systematically toward the visual duration, especially with high auditory noise. The shift was proportional to cue-conflict magnitude, inconsistent with causal inference. We compared several models. A heuristic model in which the observer probabilistically switches between the visual and auditory cues was preferred for most participants, although performance differences across models were small. Within the tested conflict range, the forced fusion, causal inference, and probabilistic cue switching models produced overlapping, near-linear shifts as a function of cue-conflict. Model simulations further revealed that given the measured sensory noise, forced fusion and causal inference can be discriminated only with unreasonably large conflicts. Together, while our results suggest that observers do not rely on causal inference when judging auditory durations under our conditions, high sensory encoding noise in auditory duration limits the discriminability of competing computational models.

Reinforcement learning theory formulates distinct decision-making strategies, including reactive model-free and deliberative model-based strategies. This study investigates how mice adjust their reinforcement learning strategies while learning decision-making in dynamic environments. Unlike previous studies that focused on behaviors after extensive training periods, we analyzed changes in learning strategies in the course of training of a two-step decision-making task with probabilistic state transition and fluctuating reward probabilities. Our statistical behavioral analysis showed that the stay-probability following common and rare transitions diverged with training, a signature of strategies that utilize knowledge of task structure. We fit various reinforcement learning strategies to behavioral data and found that structure-informed strategies became increasingly dominant in their behaviors during training. Whereas previous studies emphasized transition from goal-directed to habitual strategies after extensive training, which were often associated with model-based and model-free strategies, respectively, our results newly demonstrate a shift from model-free to structure-informed strategies in early training in mice. Author summaryReinforcement learning theory allows us to examine how we make decisions and what approaches we use to optimize rewards. Most previous research, however, has examined animal behavior only after extensive training. Here we analyzed how mice adjust their reinforcement learning strategies as they are trained in a two-step decision-making task. Initially, mice relied on reactive model-free strategies, but as training progressed, their behavior began to incorporate knowledge of task structure. While previous studies suggested transition from model-based to model-free strategies with extensive training, our study revealed the opposite in the early stage of training.

Confidence evaluates the likely accuracy of a current decision. However, to be maximally informative about accuracy, confidence judgments should incorporate information about ones broader decision tendencies, such as their propensity to favor specific alternatives. We distinguish bias-aware confidence, which considers such tendencies, from bias-blind confidence, which relies only on evidence available on the current trial. To adjudicate between bias-aware and bias-blind confidence, we identified a signature of bias-aware confidence: the down-weighting of confidence for alternatives that a participant is biased toward. We then used a large dataset (N = 200) spanning 4- and 8-choice digit-classification tasks to show that humans reliably exhibit this signature of bias-aware confidence. This effect was reduced under speed pressure and could not be explained by guessing. In contrast to the human results, artificial neural networks (ANNs) trained for object recognition lacked this signature of bias- aware confidence. Importantly, augmenting ANNs with a metacognitive module that allows confidence to take the networks biases into account led to the emergence of human-like bias- aware confidence. These findings show that human confidence incorporates not only information from the current trial but also longer-term decision tendencies, and that this capacity - absent in standard ANNs - can be conferred through specialized metacognitive mechanisms.

Response inhibition is a key control function that allows humans to perform safe, goal-directed behaviors. The dominant behavioral-computational model holds that response inhibition fails when the inhibition process is too slow to intercept unwanted movements. However, many have hypothesized that some inhibitory failures instead result from a failure to launch the inhibition process altogether. Since no method exists to identify individual trigger failure (TF) trials, they are hitherto a largely hypothetical phenomenon. We combined Bayesian process-mixture modeling with likelihood ratio testing and jackknife resampling to identify TF trials in 253 humans. We find that TF indeed represent a separate category from other inhibitory failures. Unlike other inhibitory failures, TF do not result from premature responses. Furthermore, TF lack a stop-signal P3 event-related potential, a neural index of response inhibition. Surprisingly, TF do not result from perceptual/attentional lapses. Thus, TF are a qualitatively distinct class of executive failure during response inhibition.

Behavioral experiments are often infeasible when stimulus spaces have many dimensions or when testing time is limited. One way to address this challenge is adaptive stimulus selection, where informative stimuli are chosen dynamically based on participants responses. However, in high-dimensional spaces, identifying such stimuli is computationally demanding. Here, we describe High-dimensional Online Particle Estimation (HOPE), which selects informative stimuli in less than a second for up to 50 dimensions, enabling efficient estimation of high-dimensional psychometric functions. We validate HOPE through simulations and a face-categorization experiment in an 18-dimensional parameter space with human participants. Compared to uniform stimulus presentation, HOPE reduces uncertainty over model parameters two-to three-times faster, reaching the same certainty in half the trials or fewer. This efficiency enables psychophysical studies that were previously impractical due to the exponential scaling of trial requirements.

Deciding when to stop gathering information and commit to a choice is a fundamental challenge in decision-making under uncertainty. Normative characterizations such as Partially Observable Markov Decision Processes (POMDPs) prescribe mathematically optimal stopping rules; however, human evidence gathering systematically departs from optimality. Pathological departures -- such as the excessive indecisiveness characteristic of obsessive-compulsive disorder (OCD) -- offer an important opportunity to investigate the cognitive mechanisms involved in stopping. We extend a POMDP framework to incorporate key candidate suboptimalities: a biased prior belief, transient evidence exaggeration, progressive forgetting, boosted costs of error, temporal regulation (patience and urgency), and misperception of a deadline. We evaluate this model in a pre-existing dataset comprising 105 participants spanning healthy controls, generalised anxiety disorder, and the OCD spectrum performing an information gathering task with controlled, stochastic, deadlines. Model comparison reveals that human sequential choices are broadly governed by subjective risk penalties and time-dependent urgency, with a smaller and less certain contribution from an over-weighting of recent evidence, which a random-effects comparison does not support at the population level. Individuals differ in how that over-weighting is implemented: in one deadline condition, subjects divide almost evenly between models carrying a transient exaggeration of the newest sample, models carrying progressive forgetting of older evidence, and models carrying no recency mechanism at all. Crucially, while risk sensitivity and choice stochasticity act as shared mechanisms across conditions, mechanisms such as belief bias and patience are more variable. Finally, using OCD as a clinical case study, we demonstrate that simulating choices from the fitted exaggeration model reproduces model-agnostic regression signatures of clinical indecision, which the forgetting and no-recency accounts do not. These findings offer a generative foundation for dissecting clinical departures in information gathering across the obsessive-compulsive spectrum.

Action-selection is determined by a combination of goal-directed and habitual processes. Habits are defined as the reward-independent, stimulus-response relationships which form when an action is regularly executed in the same context, regardless of outcome. An influential computational model proposes that habit formation is driven by action prediction errors which occur when non-habitual actions are taken. It has been further suggested that action prediction errors are encoded in activity of specific dopamine neurons, and it has been recently observed that dopamine activity in the tail of the striatum follows a pattern consistent with the action prediction errors. However, the original models capture changes in habits across trials, but do not describe the time-course of action prediction errors within trials, hence it is difficult to directly compare them with dopamine activity. We begin by outlining the temporal-difference action learning algorithm, which uses biologically-plausible mechanisms to determine how dynamic changes in action intensity influence the resultant prediction errors across near-continuous time. We then demonstrate that dopaminergic data recently collected from the tail of the striatum is better represented by action prediction errors than reward prediction errors. Overall, our results support the existence of value-free action prediction errors and associated habitual behaviour in dopaminergic signals. Author summaryWhenever we choose one action over another, there are two ways that the selection can be made. We could take the time to consider what we want to achieve, calculate which action is the most likely to give us that outcome and balance it against the possible negative consequences. These goal-directed calculations are very time-consuming and our brains could not possibly do it for every choice. Instead, we often rely on the second method, habits, which learn to copy the actions that were most often chosen in the past. In this paper, we present a new model of learning that is based on biologically plausible brain networks and applies action prediction errors to update our habits across continuous time. Using simulations, we reveal testable predictions that are specific to our temporal-difference action learning model and build an intuition for its behaviour. Finally, this model is tested against real dopaminergic data from the tail of the striatum, and we show that our model provides better explanation for these data, than classic reward-based reinforcement learning models.

Working memory is capacity-limited, but interactions with episodic memory may offset this constraint. We tested moment-by-moment contributions of episodic representations to working memory by combining the N-back and Mnemonic Similarity tasks. Thirty-one participants, undergoing eye-tracking, first encoded items in a one-back task, classifying them as "same" or "similar" to their predecessor. In a subsequent two-back task, mnemonic discrimination showed a graded, item-specific benefit of prior experience: performance was best for previously compared items, whereas recognition of identical repeats was unaffected. Successful discrimination of previously compared items was accompanied by greater pupil dilation, gradually emerging gaze patterns resembling those elicited by their similar pair-mate, and higher gaze-similarity between one-back and two-back target viewing. Diverging gaze patterns between pair-mates during one-back further predicted two-back discrimination. These findings challenge working memory's characterization as an isolated system, demonstrating how it recruits episodic computations - encoding distinct traces, predicting upcoming content, and reinstating it at retrieval.

BACKGROUND: An individual's adaptation to threatening environments under uncertainty is reflected in stress responses. Predictability (the ability to anticipate events) and controllability (the ability to control outcomes) are central to how one adapts, yet their joint influence on aversive learning remains unclear. METHODS: Thirty healthy adults completed a probabilistic aversive learning task in which cue-outcome contingencies varied across levels of predictability and controllability, i.e. whether shock intensity depended on prediction accuracy. Prediction accuracy, reaction time, subjective stress ratings, and skin conductance responses were recorded throughout. Trial-wise learning dynamics were estimated using the Volatile Kalman Filter. RESULTS: Prediction accuracy reduced as environments became less predictable and negatively associated with higher learning rates across predictability levels, with the strongest relationship observed in highly predictable blocks. Skin conductance responses showed that moderately predictable environments elicited responses like those in highly predictable environments when accurate predictions reduced shock intensity, but resembled responses in unpredictable environments when shock intensity was uncontrollable. Model comparison revealed a double dissociation between subjective stress ratings and skin conductance responses. Subjective ratings were best explained by model-derived volatility when prediction accuracy determined shock intensity and by belief uncertainty when it was independent of prediction accuracy, whereas skin conductance responses showed the reverse pattern. Reaction times were best explained by belief uncertainty when predictions influenced shock intensity. Higher anxiety was associated with elevated learning rates in highly and moderately predictable blocks when predictions did not control shock intensity.

IIt has been suggested that courtship signals reflect a potential mate's learning ability or nervous system competence. However, there is no rigorous theory that explains how features of sexual signals represent a nervous system's "quality". Such a theory may provide a mechanism for mate assessment via sexual signals and offer an explanation for why courtship signals are rhythmic and stereotypic. In our paper, first we use a general model of optimal neural decision-making to show that variance in an organism's solution time for a given fitness problem lowers the fitness gain rate; more specifically, in well-supported "competing accumulator" models of decision making, we show that noise in the slope of spike rate increase in evidence accumulators increases both reaction time and the probability of a sub-optimal decision. In conclusion, higher timing regularity leads to quicker and better decisions. This finding accords with extensive human study data showing that variance in reaction times is negatively associated with various measures of motor and cognitive performance. Thus, selection should favor individuals that require potential mates to advertise courtship signal regularity to indicate their nervous system's general timing consistency (the timing-consistency signaling theory). The focus on signal consistency (rather than on signal duration or power) may account for why courtship signals are typically rhythmic, are often multi-modal, and why rhythmic signals are also employed in territorial contests. One of the model's several predictions is that individuals should favor potential mates with lower noise in courtship signal features such as inter-pulse intervals.

Human language processing can be studied through both behavior and brain activity, yet it remains unclear whether these two data types reflect sensitivity to the same information. One influential view holds that both behavioral and neural responses are largely determined by processing effort, often estimated by word surprisal together with the context-independent properties of word frequency and length. At the same time, neural responses have been shown to encode richer aspects of linguistic content, including meaning. Here, we use neural network language models to operationalize these alternatives and systematically compare, within the same analytic computational framework, the predictive power of low-dimensional effort-based predictors and high-dimensional embedding representations that encode contextualized linguistic content, including meaning. Across 8 behavioral datasets and 5 neural datasets (4 fMRI and 1 ERP), we find that processing effort captures substantial variance in both behavioral and neural measures of language processing, in line with much previous work. However, for brain responses---but not for behavioral measures---embedding representations carry substantial predictive power beyond the estimates of processing effort. These results therefore suggest that neural data provide access to rich, high-dimensional dynamics of language comprehension, whereas behavioral data reflect a bottlenecking of these dynamics into a small set of theoretically motivated properties of contextualized linguistic input.

Longitudinal animal experiments generate behavior that is individual, history-dependent, and sometimes affected by ordinary procedural irregularities, yet analyses commonly reduce such records to pooled averages or synchronous trial-level explanations. We introduce history-structured forecasting as an auditable framework for determining whether an animals own preceding behavior carries predictive information beyond current-trial context. We applied the framework to 213,990 events from rats performing an auditory duration-discrimination task, using leakage-safe chronological forward-chaining, explicit trivial baselines, and controls that reset, exchange, or disrupt behavioral history. A transparent gradient-boosted model achieved 51.7% four-class accuracy, exceeding last-action persistence (40.9%) and prefix-derived subject-modal prediction (37.2%); decline-class AUPRC was 0.525 against a prevalence baseline of 0.303. Validation showed that the predictive advantage depended predominantly on each animals short-range sequential action history rather than group-level history, subject identity alone, or reward/correctness features, and strengthened on genuine choice trials. Forecasting remained informative across all 23 labeled animals, including six with recoverable records affected by incorrect training programming. These results revise the interpretation of rewarded give-up behavior while demonstrating how recoverable irregular records can be retained in transparent robustness analyses. History-structured forecasting offers a reusable open-science strategy for extracting reproducible evidence from imperfect longitudinal animal records without creating an artificially clean cohort.

Adults who stutter can speak fluently when speech is not addressed to another person, but stuttering emerges when they aim to convey information to a listener. The value of the information being conveyed to the listener also affects the likelihood of stuttering. Why should the mere absence of a listener neutralise a profound motor deficit, and why does a word's predictability affect whether it is spoken fluently? To resolve this socio-motor paradox, we develop a computational model of stuttering within an active inference architecture. The model represents the communicative context, including whether a listener is present and whether the agent is speaking or listening. It was designed around two candidate mechanisms for stuttering, a prior for silence and rigid phoneme sequencing precision. Using both, the model produced fluent private speech and more stuttering-like events during social speech. In the same parameter regime, the model also showed more stuttering-like events on words with higher information value, and produced a word-length effect, in which disfluency increased with longer words. To our knowledge, this is the first model of stuttering to generate both the private speech and the information-value effect from inferred communicative context. By representing the listener as a hidden state that makes the sensory consequences of resuming speech ambiguous, the model offers a computational link between social cognition and speech-motor instability, and suggests that speech fluency depends on whether the speaker believes anyone is present. Clinically, it may offer testable hypotheses and a route to personalising treatment, since the same overt severity can arise from different combinations of parameters.

Biological neurons can perform nonlinear computations within their dendrites and support branch-localized plasticity. This raises the possibility that single cells can store memories more efficiently and with less interference by confining synaptic modifications to specific dendrites. We study a parallel-dendrite model performing online familiarity detection and compare three dendrite-update rules during learning: (i) independent thresholding, (ii) an interacting rule that adapts the target local dendritic activation per item, and (iii) an interacting n-winners-take-all (WTA) rule that constrains the number of updated branches per item. The interacting rules substantially improve capacity by limiting variance in memory responses and decorrelating weights across branches -- even when inputs are strongly correlated. These results suggest that competition among dendrites, consistent with resource-limited plasticity mechanisms, can enhance single-cell memory beyond non-interacting schemes.

Learning in uncertain environments requires identifying the relevant associations between stimuli, actions, and outcomes and determining how strongly to update these associations. Although often treated separately, these components likely interact in the brain. We hypothesized that this interaction shapes individual learning rates according to cue-choice alignment and reward outcome, thereby improving discrimination between competing cues. We tested this hypothesis using a probabilistic learning task in which human participants predicted outcomes based on multiple cues and reward feedback. We measured gaze and manipulated cue saliency to assess and influence which cues were preferentially processed during choice and feedback. Computational modeling revealed that learning rates were selectively enhanced for cues supporting the chosen option after reward and for cues opposing it after no reward. This learning-rate asymmetry based on cue-choice alignment sharpened discrimination among predictive cues, increased robustness to noise, and improved performance. Moreover, differential gaze toward supporting and opposing cues predicted this asymmetry, which was causally altered by manipulating cue saliency. Together, our results suggest that attention provides a unifying mechanism for coordinating what we learn from with how much we learn, helping preserve distinctions among competing cues and bringing several learning asymmetries within a common framework.

Several theories propose that perception and attention are governed by rhythmic processes that give rise to periodic fluctuations in behavior. However, empirical support for behavioral rhythms has been derived largely from paradigms involving brief, static stimuli. Here, we introduce a temporal averaging task requiring integration of rapidly unfolding visual features. Across three experiments, we tested averaging of orientation, size, and color under different eccentricity conditions. We used a temporally weighted averaging model to assess whether the influence of individual stimulus samples on perceptual estimates exhibits periodic modulation over time. We found no common rhythmic signature across tasks. Instead, orientation and size judgments showed reliable low-frequency modulations (<2.5 Hz), whereas color judgments showed only weak trends. Higher-frequency components (~3.5-8 Hz), often linked to theta and alpha rhythms, were observed only in a subset of participants and were limited to parafoveal orientation processing. These findings challenge the notion of universal behavioral rhythms and instead suggest that temporal dynamics are task-dependent, with slow oscillatory processes emerging as the most consistent feature.

Newly acquired memories are initially fragile and are consolidated into long-term memory over time. Although consolidated memories were once thought to be resistant to interference, accumulating evidence shows that memory reactivation renders them transiently labile, making them susceptible to modification or interference before reconsolidation. Crucially, however, there is substantial variation in whether reactivated memories are disrupted by new information, remain protected from it, or even strengthened by it. The factors that guide this modification are largely unknown. To systematically investigate this, we examined motor memory interference using a classic A-B-A visuomotor rotation paradigm. Participants adapted to a 30-degree clockwise rotation (A) on Day1. On Day2, an interfering 30-degree counter-clockwise rotation (B) was introduced under varied conditions: directly without reactivation, after brief reactivation of A, after expression of A without feedback, or following a gradual transition from A to B. The final experiment used explicit contextual cues (a secondary follow-through target) to distinguish A and B trials. Contrary to the simple prediction that reactivation should increase vulnerability to interference, reactivating the original memory before introducing interference protected it, as evidenced by significant savings during relearning on Day3. In contrast, introducing interference directly, without reactivation, disrupted the original memory. This protection was consistent with a contextual-inference account: the large sensory prediction error experienced during the abrupt transition from A to B served as a latent contextual cue, signaling a new context and thereby shielding the original memory from being overwritten. Eliminating this prediction error through an immediate washout session with a similar error profile or through a gradual A-to-B transition abolished the protective effect and disrupted the original memory. Furthermore, when explicit contextual cues distinguished the two perturbations, memories were protected even in the absence of a salient prediction error. Our findings are consistent with a contextual inference account in which the fate of a consolidated memory, whether it is modified or protected, is shaped by the availability of explicit cues or latent signals such as sensory prediction error at the time when interference is introduced.

How conceptual knowledge is organized in the mind remains difficult to observe directly in natural behavior. Here we show that explaining knowledge to others brings out spatially structured representations of conceptual memory in naturalistic speech. Participants with varying schizotypy traits learned associations between two conceptual dimensions through either image-based or language-based input, and subsequently described their memory strategies or explained how they would teach the information to others. Spatial structure was more likely to emerge during teaching than reflection, particularly following image-based learning. As predicted by prior evidence that visual input requires the active construction of internal relational scaffolding, cognitive disorganization selectively attenuated spatial expression after image-based, but not language-based, learning. These findings establish naturalistic speech as a behavioral readout of conceptual map structure, and suggest that a common mechanism links map construction and cognitive disorganization across task performance and verbal communication.

Working memory (WM) is often assumed to play a stronger role in mental operations than in pure storage. However, much of the evidence comes from tasks using novel stimuli requiring active maintenance. Everyday cognition, in contrast, often involves operating on information retrieved from long-term memory (LTM), which may not always require sustained WM storage. Moreover, prior evidence for enhanced WM involvement relies on univariate measures, which cannot separate procedural demands of operations from representational strength of operation-relevant items. Here, we used EEG to test how WM supports mental operations on LTM. First, participants studied color-position associations. Then, on each trial, a color cue prompted retrieval of its associated position, followed by a novel position. Across blocks, participants either performed a mental operation to compute the positions' spatial midpoint or judged whether the probe matched one of the memory positions. Representations of task type and memory position were assessed using MVPA and inverted encoding models on alpha-band power, respectively. Task type was decoded throughout the trial, reflecting persistent task-set representations. In contrast, LTM position was represented in WM more strongly for integration than recognition early in the retention and operation periods, but these differences were transient. These findings challenge the view that mental operations inherently demand enhanced WM engagement: when information is available in LTM, increased WM involvement is transient, not sustained.