Psychological Review

Resolving the binding problem, i.e., how we integrate perceptual features in space and time to perceive objects, people and events as coherent wholes instead of masses of free-floating features, has been a central goal of cognitive scientists for decades. Conjunction errors occur whenever features are mis-bound leading to illusory percepts that were not actually presented. A specific kind of Rapid Serial Visual Presentation experiment induces temporal illusory conjunctions, called distractor intrusion errors. Despite these errors being well-established, current theories and models provide limited explanations. We present the 2-feature Simultaneous Type/Serial Token (2f-ST2) Model, the latest addition to the family of ST2-models, which simulates the most important behavioural and electrophysiological distractor intrusion findings. These include response distributions, reaction time patterns, and evoked electrophysiology responses. Importantly, we simulate these results across diverse empirical paradigms, implementing a range of stimulus sets, temporal dynamics, and involving multiple responses after each trial, without needing to modify our model's architecture or core parameters. Our findings are discussed in relation to theories of selective attention and working-memory binding. We argue that the 2f-ST2 model is more parsimonious than the leading theory of distractor-intrusions (the focussed-attention/sophisticated-guessing model), since it uses only a single binding mechanism, implying that correct bindings are simply fortunate conjunctions. The 2f-ST2 model further makes predictions for stimulus-competition, subjective-visibility and human electrophysiology. In summary, the ST2-model is a general framework for understanding how perception, temporal attention and working-memory interact, and the 2f-ST2 model extends the framework to distractor intrusions, thus explaining temporal feature binding.

Motivational incentives are known to improve working memory (WM) when items differ in value, but whether they enhance performance when all items are equally valued, and if so, by what mechanism, remains unclear. Improvements could reflect enhanced memory-based representations or speed-accuracy trade-offs, whereby incentives promote slower, more cautious responding without improving memory itself. Here, we tested these alternatives across two experiments using an auditory WM task in which incentives varied across trials, but all items on a trial were equally incentivised. In Experiment 1,we allowed participants to resubmit their responses if desired. Reward-related improvements in WM performance were selectively observed when memory load was high, and only at serial positions associated with greater interference. These improvements were also dissociated from systematic response slowing and resubmission behaviour. This indicated that the benefits of incentives were unlikely to be attributable to response caution. Critically, in Experiment 2, motivational benefits persisted under strict response deadlines, where speed cannot be traded for accuracy. Computational modelling showed that reward selectively improved the precision of memory-based responses without reducing guessing. Together, these findings indicate that motivational incentives improve WM by enhancing the quality of memory-based representations, rather than by strategic slowing or altering response strategies.

In visuomotor sequential tasks, chaining actions smoothly requires tight coordination between sensory and motor systems to deal with competing demands, such as reducing uncertainty about future opportunities for action while controlling ongoing ones. In a football-based decision-making task, experts and novices used different strategies to overcome these competing demands, demonstrating a proactive and a sequential visuomotor strategy, respectively. We propose a computational explanation of expertise-related difference in policy selection, grounded in the active inference framework. We simulated the selection of sequence of actions using a partially observable Markov decision process (POMDP) model, and we manipulated three parameters: the reward structure, the prior beliefs about state transition, and the sensory precision. Simulations showed that selection of the proactive strategy relied on strong prior beliefs, high sensory precision and high valuation of fast passing. Conversely, the sequential policy was selected when rapid performance was not highly valued, or under noisy prior beliefs, or lower level of sensory precision. Decreasing sensory precision and adding noise in prior beliefs also decreased decision accuracy. These results suggest that policy selection leading to novice or expert visuomotor strategy reflects differences in the underlying generative model, particularly in how agents value different aspects of the task.

Learning depends not only on the content of what we learn, but also on how we learn and on how experiences are structured over time. To investigate how task similarity and training regime interact during learning, we trained participants on spatial and conceptual learning tasks that shared either similar or distinct underlying structures, using either interleaved or blocked regimes. Interleaving the two tasks hindered performance when their structures were similar, compared to when they were different. In contrast, blocked training produced the opposite effect: it improved performance and facilitated transfer across similar tasks. This effect, however, emerged only when participants first learned the conceptual task, followed by the spatial task, suggesting an asymmetric interaction between task order and structural similarity. We also replicated our results using a neural network model, providing converging evidence for the computational principles governing the interplay between training regime and structural similarity in multi-task learning.

Procrastination is widespread, costly, and persistent despite strong intentions to change. We show how procrastination can arise in stationary environments, where there are no deadlines, changing incentives, or new information to justify delay. With hyperbolic discounting, stochastic delay, i.e., acting with a fixed probability each period, emerges as the only self-consistent policy: the only policy that resolves belief-policy inconsistency. We further show that standard reinforcement learning algorithms converge to this Pareto-inefficient solution. We extend the self-consistent model to divisible and continuous-action tasks. In these settings, allowing a voluntary break after completing a subtask never increases, and often reduces, expected delay; enforcing such breaks can reduce delay further, often substantially. Optimal task division equalizes completion probabilities across subtasks, implying that early stages of a larger task should be easier when the goal is to reduce overall time to completion. The model therefore gives a formal rationale and exact quantification for familiar remedies such as task division, starting easy, and enforced breaks, while explaining why procrastination can persist even when intentions, incentives, and information remain unchanged.

Reflecting how we organize visual experience in everyday memory, visual working memory is increasingly understood as a system in which individual item representations are organized within structures rather than maintained in isolation. Among these, relational structure may be especially consequential because, by specifying how one item value lies relative to another within a feature space, it could allow information about one remembered value to constrain which values are plausible for the other. Yet demonstrating such constraint is challenging because item-specific mnemonic evidence and relational evidence ordinarily support essentially the same estimate. We broke this equivalence with biased post-encoding feedback for one item, making item-specific and relation-based predictions for the other diverge. Across three experiments, participants remembered two sequentially presented orientations, with feedback for one shifted slightly clockwise or counterclockwise from its actual value. Participants incorporated this bias into memory for the feedback-provided orientation; critically, it also appeared in reports of the other orientation, which received no feedback, in the direction predicted by the signed angular offset linking the two remembered values. This feedback transfer weakened with increasing angular separation but occurred in both directions between the first and second orientations. These findings show that relational structure directly constrains individual value estimates in visual working memory, even for items encountered separately. By dissociating normally coincident item-specific and relation-based predictions, our approach reveals an otherwise hidden relational contribution. A probabilistic account explains these findings through joint inference from uncertain item-specific and relational evidence, with their relative uncertainties governing transfer strength.

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.

The bounded accumulation model is a dominant theoretical framework to characterize evidence accumulation during decision-making. Using a dynamic sequential tone accumulation (DSTA) task in marmosets, we found that the contribution of late tones to decisions was greater for weaker than for stronger evidence strength-an effect we term evidence-strength-dependent temporal weighting (ETW). Rigorous exclusionary analyses and drift-diffusion modeling further revealed that the bounded accumulation mechanism alone fails to account for this effect, and the online modulation of attentional engagement was identified as the necessary mechanism in addition to the bounded accumulation. Our work thus establishes that attentional engagement is adaptively modulated by evidence strength during complex decision-making.

The development of large language models (LLMs) offers promising opportunities for predicting human behavior across diverse contexts. However, most prior work has emphasized behavioral imitation, with limited attention to transparent or interpretable models of the cognitive mechanisms underlying human decisions. In this study, we introduce MindEvolve, an autonomous workflow designed to predict behavior in social interactions by generating interpretable symbolic models of cognition. We systematically evaluate the modeling capabilities of multiple LLMs across a battery of socioeconomic games covering four core domains of social cognition: economic preferences, social preferences, social reasoning (theory of mind), and recursive planning. The symbolic models proposed by the LLMs are assessed by expert human evaluators for interpretability and theoretical coherence. Our results show that while most LLMs robustly capture economic and social preferences in relatively simple strategic settings, and in some cases generate novel models that integrate broader knowledge than those proposed by human experts, their capacity to model more complex psychological processes remains limited. However, a subset of state-of-the-art models demonstrates promising performance in capturing higher-order reasoning processes such as theory of mind and recursive planning. Together, these findings highlight the emerging potential of LLMs not only to imitate behavior, but to generate interpretable, mechanistic accounts of human social decision-making. Our results provide a roadmap for advancing LLM-based cognitive modeling toward human-expert-level theory construction.

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.

Spontaneous eye blinks have long been linked to cognitive processing, yet how task demands shape blink timing and its relationship to behavioral performance remains unclear. We examined spontaneous blink behavior in 576 adults performing two variants of the Continuous Performance Task (CPT). Blink occurrence and timing were most strongly modulated by the experimental condition in the more demanding CPT-AX task, whereas their association with response time was stronger in the CPT-X task, where more consistent blink timing predicted faster responses. This dissociation suggests that task structure changes not only blink behavior but also the behavioral relevance of blink timing. These findings are consistent with an adaptive scheduling account of spontaneous blinking and provide a conceptual framework for understanding when and why blink timing contains chronometric information about ongoing cognition.

Collective decision-making experiments have largely focused on simple scenarios in which groups choose between two options that differ along a single attribute. However, it remains unclear whether principles derived from this single-attribute best-of-2 paradigm generalizes to the multi-attribute, multi-option decisions that groups often face in nature. Here, we use mean-field and agent-based models of cockroach aggregation to study collective decision-making across a range of problem complexities, varying both the number of options (N) and the number of attributes describing each option. Our models make 3 novel predictions: (1) cockroach groups use a compensatory algorithm when integrating attributes, trading off strength in one attribute against weakness in another; (2) decision-making collapses abruptly once N exceeds a critical threshold; and (3) decision time scales non-monotonically with N. Together, these results indicate that dynamics characterized in single-attribute best-of-2 experiments do not extrapolate to multi-attribute best-of-N decision-making. Collective choice under realistic complexity may follow principles not yet captured by existing models.

Recent research has suggested that the availability of non-instrumental information about the outcome of a risky choice increases risk appetite. In this study, we aimed to perform a conceptual replication of these findings and to examine the cognitive mechanisms underlying this effect. Across two experiments (N = 150, 102), we presented participants with mathematically fair gambles and allowed them to choose the size of their bet. Between trials, we varied the presence of non-instrumental information that would reveal the outcome ahead of time. In both experiments, we did not find consistent evidence for an effect of the availability of non-instrumental information on bet size. These findings suggest that the previously reported effects of non-instrumental information on risk appetite may have been an idiosyncratic feature of experimental design, rather than a more general phenomenon that characterises human decision making under risk.

Effective decision-making requires evaluating options based on relevant features while reducing attention to those that are not. However, how attention prioritizes one source of information over another during value-based decisions remains unknown. Across 7 experiments, we show that irrelevant features both constructively and destructively interfered with how values are assigned to choice options. When relevant and irrelevant features agreed, choice accuracy and speed were facilitated; when they conflicted, accuracy decreased and choices slowed in proportion to the value of the irrelevant feature. We modeled this interference as a weighted sum of relevant and irrelevant feature values. This model explained individual differences in choice behavior and, critically, tracked psychiatric symptom severity related to substance abuse and obsessive-compulsive disorder. This suggests that these symptoms may, in part, reflect an impaired ability to weigh relevant against irrelevant information when assigning value to choice options - a process we term feature-based valuation.

Everyday problems often require choosing among actions with different future consequences, even though fully evaluating those consequences is often computationally infeasible. The human brain is remarkable in its ability to solve such complex problems efficiently. In this paper, we hypothesize that this capacity arises from a principle inspired by Bayesian accounts of perception and sensorimotor control: evidence derived from effortful forward planning is integrated with task-specific preferences akin to priors, weighted by the uncertainty of these preferences. To test this hypothesis, we used a sequential decision-making task with a relatively large state space. Model comparison showed that participants' deviations from optimal forward planning are best captured by task-specific preferences, rather than by unsystematic errors, recent action-history effects, or offer-context biases. Moreover, the influence of forward planning on participants' responses depended on preference uncertainty, as evidenced by both choice behavior and reaction times, and preferences were not arbitrary but instead supported above-stochastic performance even in the absence of forward planning. Together, these results are consistent with a Bayesian-inspired interpretation of decision making, in which costly forward planning is integrated with state-dependent priors or preferences. These preferences may provide an inexpensive, quickly available first-pass solution, which is complemented by forward planning when preference-based guidance is uncertain.

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.

How cognitive control influences the correction of motor errors is an important question. In this study, we examined how the slow and fast processes that constitute motor adaptation are modulated by cognitive load. To address this question, we used a novel approach that differed from traditional dual-task experiments that have assessed the role of fast processes by decreasing attention to the motor adaptation task. Here, cognitive load was manipulated using a decision-making task with easy and hard difficulty levels, while simultaneously adapting to a perturbation and maintaining attention on the task. Our results showed that the hard decision-making task increased cognitive load relative to the easy task, resulting in attenuated steady-state learning. The relative contributions of the fast and slow components were assessed using a dual state-space model, which showed that the former was larger under the high-load condition, whereas the latter was larger under the easy-load condition, consistent with a competition model in which a shared error signal is partitioned between the two processes. We also showed that the effect of cognitive load on fast processes was not attributable to increased reaction times or task performance. Taken together, these results support the notion that faster explicit motor adaptation processes derive from cognitive processes involved in action selection.

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.