Cognition

Automation has been shown to weaken the sense of agency (SoA), the experience of controlling ones actions and their outcomes, by disrupting the predictive link between intention and effect. Explainable AI (XAI) has been proposed as a solution, yet the neurocognitive mechanisms through which explanations restore agency remain unclear. Across three EEG experiments using an autonomous-driving paradigm, we examined how automation and different forms of AI explanations modulate explicit agency judgments and early neural markers of agency-related predictive processing. In Experiment 1, automation reduced explicit feelings of control and was associated with reduced sensory attenuation, as reflected by increased P1-N1 amplitudes, decreased N1-P2 amplitudes, and delayed N1 latencies. In Experiment 2, distal (goal-level) explanations partially restored agency and selectively modulated early auditory responses, decreasing P1-N1 and increasing N1-P2 amplitudes. In Experiment 3, combining distal and proximal (trajectory-level) explanations produced the strongest behavioural and neural restoration of agency, yielding a graded attenuation of P1-N1 and enhanced N1-P2 responses along with accelerated N1 latencies. Across all experiments, mismatch negativity (MMN) remained unaffected, indicating that pre-attentive deviance detection is preserved regardless of agency or explainability. Together, these results identify component-specific EEG markers that track fluctuations in the sense of agency and demonstrate that multi-level intention sharing by AI systems enhances both predictive engagement and explicit control experience. This work provides a neurocognitive foundation for designing explainable autonomous systems capable of maintaining user agency.

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 Sense of Agency (SoA), the experience of being in control of ones own actions, is thought to emerge from the comparison of internal sensorimotor predictions with afferent feedback. Classical comparator models treat SoA as arising from a prediction-feedback comparison. Volitional action likely engages multiple forward models that predict distinct features of an outcome, such as its timing and spatial trajectory. Whether prediction errors arising from these distinct forward models are integrated into a domain-general representation of agency, and if so by what computational logic, remains unresolved. We addressed this question using a Virtual Reality reaching task. Participants observed a virtual hand replicating their movements while we independently manipulated two sensorimotor domains: temporal delay and spatial angle deviation, in isolation and in factorial combinations. After each trial, participants made an SoA judgment. We examined whether SoA responses show computational hallmarks of integration between different features of sensorimotor prediction. Specifically, we pre-registered three computational models (Multiplicative, Minimum, and Mean) and compared their fit to per-trial responses. Across an exploratory sample (N = 16) and a pre-registered replication (N = 38), SoA declined monotonically with conflict magnitude in both domains. Critically, a Multiplicative integration rule consistently outperformed the Minimum and Mean rules. These results provide direct evidence for domain-general integration between prediction errors in SoA, governed by a multiplicative computational logic.

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.

The pleasurable urge to move to music - known as groove - has been linked to rhythmic and harmonic complexity via an inverted-U relationship, consistent with predictive coding theory of music. This link has been demonstrated mainly in Western listeners, leaving its cultural generality unclear. Given reports of heightened pitch and harmony sensitivity in Chinese listeners, we predicted stronger responses to harmonic complexity and weaker responses to rhythmic complexity compared with Western listeners. We conducted two online experiments measuring "pleasure" and "wanting to move" ratings for musical excerpts varying in rhythmic and harmonic complexity, comparing Chinese and Danish participants as representatives of Eastern and Western traditions. In Experiment 1, which manipulated both dimensions, both groups showed inverted-U responses to rhythmic complexity, but Danish participants exhibited a sharper curve, indicating greater affective differentiation. No inverted-U pattern emerged for harmonic complexity, and Chinese participants did not show enhanced harmonic sensitivity. Experiment 2 focused on rhythmic complexity using ecologically valid stimuli with finer gradations. Again, Danes showed greater sensitivity to rhythmic variation, reflected in a sharper curvature of their inverted U responses, though peak responses did not differ between groups. Our findings support the universality of predictive responses to rhythmic complexity, while highlighting the modulatory role of cultural exposure in shaping musical affective experiences.

Exploration is fundamental to how children learn about the world. To explore a new environment comprehensively, a child must navigate without getting lost while simultaneously attending to salient information along the way. Both abilities improve across childhood, yet most studies examine them in isolation, often in constrained spaces with limited sensory complexity. Here, we asked 7- to 14-year-old children (N=42) and young adults (18-22; N=42) to freely explore a large furniture store in immersive virtual reality, which contained many objects and attention-capturing multisensory events. Surprisingly, despite traveling comparable distances, children explored less - they covered less spatial area and examined fewer unique objects than adults. Fine-grained analyses of their behaviors showed that the reduction in exploration was due to children spending more time backtracking and children taking longer to adapt to salient multisensory objects, which were uncorrelated across individuals after controlling for age and independently predicted exploration. However, only backtracking duration predicted success in a subsequent search task, demonstrating a link between navigational success during exploration and the quality of learned spatial representations for later use. Together, the results demonstrate that children explore less comprehensively than adults in large-scale spaces because they are susceptible to getting lost and perseverating on salient events nested within a large environment, but that these behaviors change with age. This suggests that while children are widely considered to be curious explorers within local environments, their ability to do so within larger environments is shaped by the maturation of cognitive systems that integrate experiences over complex contexts.

Despite the ubiquitous nature of visual search in our daily lives, our knowledge of the underlying mechanisms is almost entirely derived from studies using simple, synthetic stimuli briefly flashed on a computer monitor. In real life, however, visual search occurs in three-dimensional environments over much longer timescales and involves the dynamic coordination of attention, memory, and action planning, generating novel predictions that cannot be tested with traditional approaches. Here, we examined how self-guided visual search unfolds under naturalistic conditions that extend over space and time. We asked participants (N=80) to find specific objects (e.g., a green lamp, an orange shelf) in an immersive, visually rich virtual reality environment using range-limited teleportation (2 m steps). Consistent with our hypotheses, we found that self-guided search engaged a hierarchical information-seeking policy characterized at each stage by recurring plan-and-execute cycles paired with subgoal-specific attentional priorities. Subgoal priorities progressed from wayfinding to identification of clusters of target-similar objects, and then to specific target features. These search priorities progressively narrowed the search space through coordinated eye, head, and body movements that recurred throughout the task. We conclude that naturalistic search is hierarchical, built from sequential subgoals with shifting attentional priorities that single-level tasks cannot capture. Moreover, immersive VR with eye-head-body tracking dissociates motor planning from execution moments and reveals how looking content shifts with each task subgoal while the person moves through space, something desktop tasks cannot capture.

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.

Recognising people precedes and informs most social interactions, yet it is not clear why individuals differ so markedly in face recognition ability. Exposure-based accounts propose that it is shaped by visual experience with faces, known as 'face diet', but most prior work treats exposure as passively acquired, thereby masking the active and dynamic nature of visual attention in natural settings. Here we examined whether spontaneous attention to faces in everyday environments predicts face recognition ability. Eighty participants wore mobile eye-tracking glasses while walking through a busy university campus and during a face-to-face interaction, before completing standard measures of face recognition ability. Fixations directed to faces showed stable individual differences in both settings. People who looked more at faces scored higher on our composite measure of face recognition ability, and gaze in the two settings predicted ability independently, together accounting for 22% of its variance. These findings suggest that differences in how people actively sample visual information from their environments may be an overlooked source of variance in perceptual expertise.

Mice are widely used to study the neural basis of perceptual decision-making, often with the ultimate goal of extrapolating insights to humans. Yet, whether humans and mice rely on similar computational strategies remains rarely tested, partly because the two species acquire knowledge differently. Animals learn from experience, while humans usually receive verbal task instructions. Here, we explicitly contrast these modes of learning in humans who perform a perceptual decision-making task adapted from a standardized rodent task. Specifically, we present an open behavioral dataset from 80 human participants who performed 600 trials of a visual contrast-discrimination task, matched to the standardized International Brain Laboratory task for mice (International Brain Laboratory et al., 2021, 2025). Participants either received verbal instructions ('Instructed') or were only told to discover the rules of the task by playing ('Uninstructed'). The dataset contains choices, response times, continuous cursor trajectories, eye-tracking and pupil recordings. Descriptive analyses show that Uninstructed participants were less accurate and had higher lapse rates than Instructed participants. They improved over the first 200 trials, approaching the performance of Instructed participants. These data provide a benchmark for testing whether computational strategies identified in mice also describe human behavior.

Why does an angry face feel uncomfortable? The answer is that it signals a threat. However, a face is only part of an encounter, and distance, facial stimulus type, and gaze may shape discomfort regardless of perceived anger. To separate these cues, we conducted three within-subjects virtual reality experiments. In each experiment, 24 adults viewed avatars at intimate, personal, and social distances (30, 100, and 300 cm, respectively) and rated the faces perceived anger and their own discomfort; head movement was recorded in Experiments 2 and 3. In Experiment 1, the expression (angry, neutral) and facial color (natural, red) were crossed with distance; in Experiment 2, a featureless mannequin served as a nonface comparison; and in Experiment 3, the gaze direction (direct, averted) was manipulated. Expression primarily determined perceived anger, whereas distance predominantly determined discomfort: A nearby neutral face was uncomfortable despite low perceived anger (Experiment 1). A neutral human face was more uncomfortable than a mannequin, although both received similarly low perceived-anger ratings (Experiment 2). Direct gaze increased the discomfort without changing perceived anger (Experiment 3). Backward head movement exhibited a similar pattern, with participants leaning back more from human faces than from the mannequin. These results indicate that the discomfort associated with an angry face is not merely explained by perceived anger. Instead, social discomfort was differentially associated with interpersonal distance and gaze direction and differed between the human-face and mannequin conditions.

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.

When people engage in spontaneous thought under low external constraint, thought content changes continuously and can shift across topics. However, the cognitive computational principles governing these transitions remain unclear, as does whether similar dynamics emerge in artificial language generation. Here, we formalized spontaneous narrative dynamics using a foraging-inspired framework, constructing an agent based on the marginal value theorem (MVT) that balances exploration and exploitation while selecting among informational resources represented in a participant-specific, high-order patch network. Patches were defined as partially overlapping composite structures containing second-order paths and third-order interactions, with their available resources quantified using multilevel structure values. In a multi-theme free spoken narrative task, the model predicted the temporal organization of human patch activation substantially better than random-shuffle baselines, although the strength of model conformity varied across participant-theme samples. Model-predicted patch transitions corresponded to systematic semantic segmentation and reorganization across multiple representational levels. Transition rates were positively associated with the perceived importance and likelihood of the narrated events. These findings were replicated when the modeled memory environment was enriched with additional hub-related narrative resources in a supplemental experiment. Control-model analyses further showed that patch-network topology, resource depletion history, or either order of structure value alone could explain only part of the observed dynamics; the most stable human prediction required foraging-related decisions guided jointly by multilevel structure values. Narratives generated by large language models (LLMs) also showed detectable conformity with the same framework, but the full model captured LLM content less consistently, while simpler control models accounted for a larger proportion of their patch-level dynamics. Together, these findings support a computational account in which patch-level narrative dynamics are organized by value-guided foraging over structured memory resources, while revealing graded differences in the strength and stability of these dynamics across human and artificial language generation.

Understanding other people relies on long-term memory: we use learned information about others' personalities, mental states, and personal histories to make inferences about the causes of their behavior. However, little is known about individual differences in memory for high-level social information. Here, we develop and psychometrically evaluate a behavioral measure of social memory from narrative video content. Participants (N = 103) watched the pilot episode of the TV show Friday Night Lights, and answered multiple choice questions about characters' actions, statements, inferred mental states, relationships, and personalities immediately after viewing and following a 21-day delay. The social memory task showed moderate split-half reliability, demonstrating feasibility as an individual difference measure. Memory for information about specific events in the narrative showed a decline across the delay, while memory for abstract information remained stable. Social memory performance was correlated with separate measures of face-name association and verbal episodic memory, but not with face image recognition memory. These results provide a proof of concept for a novel behavioral measure of social memory and suggest that the measure relates to episodic and relational memory but not image recognition.

Efficient motion perception requires integrating immediate sensory evidence with internal representations shaped by experience. This influence of recent history can produce serial dependence effects (SDEs), in which current judgments are biased toward or away from the source of previous information. We examined SDEs in auditory, visual, and cued audiovisual motion-direction discrimination using aggregate psychometric measures and trial-level generalized linear mixed-effects models (GLMMs), which estimated the independent contributions of the previous stimulus direction and the participant's previous response. Participants completed unisensory (n=57) and cued audiovisual (n=48) motion-discrimination tasks. Aggregate analyses indicated attractive SDEs in auditory conditions, repulsive SDEs in visual conditions, and no net SDE in the audiovisual-cued condition. Trial-level modeling revealed a more nuanced structure: previous stimulus direction produced repulsive effects in both unisensory conditions, with previous visual direction influencing judgments across all conditions containing visual motion, whereas previous response produced attractive effects only when visual information was task-relevant. In the audiovisual-cued condition, the absence of a net psychometric shift was consistent with the opposing influences of an attractive response-history effect and a repulsive visual stimulus-history effect rather than an absence of serial dependence. These findings suggest that serial dependence in motion perception reflects dissociable and concurrently operating processes, including a predominantly visual, adaptation-like stimulus-history effect and a task-dependent, decision-related response-history effect.

Task demand has a long-standing history in experimental psychology, with numerous investigations of the capacity of different systems and processes under varying cognitive load. Whilst demand is a broadly used term in the literature, it is limited by a lack of cross-domain operationalisation, and thus can be a confound in many inter-domain investigations. To address this, we generated and validated two large-scale, open-source, widely applicable stimuli sets that: (i) can probe a wide, continuous range of objective behaviourally-measured processing demand, and (ii) can be directly matched across two highly contrastive cognitive domains (semantic and visuospatial processing). We normed these tasks in a large, online cohort of healthy adults, enabling future studies to utilise and benefit from reliable baseline measures of demand. Using a matching algorithm that takes into account both speed and accuracy, we then closely paired each semantic judgement trial with a trial of comparable difficulty from the corresponding visuospatial task to produce a cross domain matched set of over 400 items. Finally, we analysed the psycholinguistic and perceptual properties of these stimuli to explore some of the underlying factors driving demand.

This study characterizes how people combine visual and tactile directional cues while acting in a fully immersive 360{degrees} virtual environment. Participants used a vibrotactile belt and VR headset to localize targets while we manipulated visual reliability and the spatial discrepancy between visual and tactile signals. Behaviorally, degraded visual input made visual responses slower, less precise, and more susceptible to tactile pull, whereas tactile-guided responses remained comparatively stable. We then asked whether these behavioral changes reflected a change in multisensory binding or a change in sensory uncertainty. A Bayesian Causal Inference (BCI) framework captured the structure of behavior under high visual reliability and continued to track individual differences under low visual reliability, even though its absolute goodness-of-fit decreased. Under extreme visual noise, Bayesian Information Criterion sometimes favored a simpler Maximum Likelihood Estimation (MLE) model, but MLE showed poor absolute fit and did not capture meaningful behavioral variability. This dissociation shows that statistical parsimony and explanatory validity can diverge when behavior becomes highly variable. BCI-derived parameters further indicated that degraded vision increased visual uncertainty, while the prior tendency to bind visual and tactile cues remained stable. Kinematic analyses added a complementary insight: early movement trajectories were strongly shaped by tactile signals, even when final localization was visually guided. Together, these findings suggest that visual-tactile integration in 360{degrees} environments depends on sensory reliability and task demands, with tactile cues providing fast body-centered guidance when visual information is limited.

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.

Certain conscious contents--such as the covert name of a viewed object--arise involuntarily and resist suppression, a phenomenon captured by the Reflexive Imagery Task (RIT). Whether a subliminal prime can bias which of two simultaneously present objects captures such an involuntary naming response has not been established. We adapted a two-object RIT into a self-contained, browser-based instrument. Thirty-eight adults (M age = 21.6 years, SD = 2.3; 26 female) viewed 24 object pairs and were instructed to fixate a central cross, to refrain from thinking of the objects names, and to click an object whenever its name intruded into awareness. On each trial a masked prime (17 ms, flanked by pattern masks) was either the exact name of one object (Exact Word Prime), a semantic associate of one object (Semantic Prime), or a neutral string (No Prime), defining a primed side per trial. A post-experiment debriefing confirmed that participants noticed the masks but none consciously perceived or could identify the prime words, indicating that the primes were subliminal. Analyses excluded 91 of 912 trials (10.0%) with more than five clicks. The priming effect was robust across measures: an omnibus comparison of clicks across conditions was significant (Friedman {chi}{superscript 2}(2) = 6.59, p = .037); within both prime conditions participants clicked the primed side more than the not-primed side (Exact Word, p = .039; Semantic, p = .005); and the primed-to-not-primed ratio exceeded parity by roughly 57-59% (Laplace-smoothed ratio {approx} 1.58; one-sample Wilcoxon p < .001 for each). A Bayesian Poisson generalized linear mixed model with random participant intercepts confirmed a credible primed-side advantage (incidence rate ratio = 1.72, 95% credible interval [1.57, 1.90]) that did not differ between prime types. No general left/right response bias emerged. An exploratory ight-side x Exact Word Prime interaction was inconsistent across model classes and is reported as hypothesis-generating. The findings indicate that subliminal lexical and semantic primes can steer the spatial locus of involuntary object naming.

Attentional selection not only increases behavioural performance and neural information for attended stimuli but, in some instances, induces changes in stimulus appearance. While reported appearance changes have mostly concerned the effects of attending to a particular spatial location, here we report a shift in appearance that occurs when attention is directed to one component of a stimulus rapidly alternating at a single spatial location. Our stimuli included two frames of different colors, which alternated at 3.75Hz or 7.5Hz. Participants reported the perceived dominance of one frame's color in the reference stimulus, by comparison with a target stimulus. We found that when one stimulus frame was more salient than the other, with a greater likelihood of capturing exogenous attention, there was an illusory perception that the frame was physically present for a greater proportion of time. When the frames were matched in salience, a task encouraging participants to endogenously direct their attention to one frame produced a smaller shift in appearance. Our results add to the body of literature demonstrating that attention can change appearance, in this case altering the appearance of attended versus unattended stimuli that are presented at the same spatial location. Furthermore, our results demonstrate that at the 'temporal transparent' alternation rate of 7.5Hz, where rapidly alternating stimuli are perceived as two translucent surfaces, attentional selection of one surface can account for the apparent paradox that feature conjunctions are perceived at stimulus alternation rates higher than predicted by the temporal limits of feature binding for other stimuli.