Visual working memory (VWM) has been thought to operate in active and passive states, but whether these states differentially engage sensory storage remains debated. The current study aims to delve into this debate further by testing whether increasing the load of active/passive states in VWM affects detection sensitivity to an incoming visual stimulus, a psychophysiological probing method which has been verified to specifically uncover the sensory nature of working memory storage. Across Experiments 1-3, we consistently found that loading either active or passive state impaired visual detection to a similar degree, indicating comparable sensory demands for both states. In Experiment 4, we further validated the manipulation of VWM states by observing dissociative memory-driven attentional bias effect of different states. Experiment 5 showed that information released from VWM no longer impaired visual detection, further confirming the specific role of working memory storage (in either active or passive state) in interfering with concurrent sensory processing. Together, these findings suggest that both active and passive states in VWM engage sensory storage, with comparable functional consequences for ongoing sensory processing.

ObjectiveThis study examined whether cognitive load produces selective effects on different trust updating pathways in AI-assisted decision making.BackgroundAlthough cognitive load affects trust in automation, its influence on the mechanisms of trial-by-trial trust updating remains unclear.MethodsA dual-task paradigm embedded in a mining exploration task manipulated cognitive load while capturing dynamic trust calibration. Guided by a dual-pathway framework, we operationalized process-based (analytical evaluation of AI recommendation correctness) and outcome-based (heuristic reliance on task outcomes) trust updating pathways. Trust dynamics and behavioral reliance were examined using linear mixed-effects models.ResultsCognitive load shifted the relative influence of the two trust updating pathways. Process-based updating was attenuated under high cognitive load, indicating reduced sensitivity to AI recommendation correctness during trust updating. Outcome-based information gained greater influence under high load, amplifying outcome-driven bias regardless of recommendation correctness. Asymmetric trust updating was evident overall, although the influence of cognitive load on this asymmetry depended on task outcomes. Overall, high cognitive load elevated both subjective trust and behavioral reliance on AI.ConclusionCognitive load shapes trust calibration through mechanism-level reconfiguration rather than global impairment. By revealing how cognitive constraints rebalance dual trust pathways-weakening analytic evaluation while amplifying heuristic outcome reliance-this study advances theoretical understanding of dynamic trust in human-AI collaboration.ApplicationThe results provide practical guidance for the design of AI systems in high-stakes settings, highlighting the need to support analytic trust updating and mitigate over-reliance under cognitive strain.

As robots increasingly act as collaborative partners, their behavior must be not only functional but also understandable. Although robot control typically follows a Perception-Decision-Action (PDA) sequence, the causal link between perception and behavior is often not observable to users. To support explainability, users must intuitively grasp how perceptual input leads to action. Drawing on Hume's (1739) principles of causal perception-contiguity and contingency-we examined how temporal delays and perception-behavior alignment shape causal understanding. Results show that delays exceeding 600 ms, or actions preceding perception, disrupt perceived causality. Moreover, perception-behavior alignment must exceed 90% to maintain causal coherence and user trust. Together, these findings define empirically grounded design ranges: perceptual signals should precede actions by 0-600 ms, and alignment accuracy should remain above 90%. Meeting these criteria enhances robot explainability and improves human-robot interaction.

Dynamic social interactions form an important component of experienced “now,” which is not an infinitesimally small moment but a “specious” present lasting roughly three seconds. Our study demonstrates how the temporal integration within experienced “now” imposes a boundary that distinguishes perceived animacy from post-perceptual reasoning in the rich social interpretations elicited by Heider-Simmel-style displays, and how this temporal dissociation further reveals a semantic boundary of perceived animacy. We hypothesized that social interpretations reflected in spontaneous perception would be severely and selectively impaired when the temporal integration within the experienced “now” is disrupted by temporally stretching the animate displays. To test this idea, we pitted spontaneous perceived animacy against post-perceptual reasoning in two ways: through self-censorship instructions that prohibited mentalizing language (Experiment 1) and a dual-task paradigm in which animacy recognition was measured as an unprepared secondary task without focal attention (Experiments 2–3). Results showed that reports of intentional actions producing tangible physical outcomes (e.g., chasing, attacking) were severely disrupted by temporal stretching but largely immune to self-censorship and dual-task distraction, indicating that they arise from spontaneous perception and are part of experienced “now.” In contrast, interpretations reflecting an agent’s epistemic states (e.g., seeing), emotion (e.g., being angry) or personality (e.g., bully) were unaffected by temporal stretching but were easily inhibited by self-censorship and dual-task distraction, suggesting that they arise from post-perceptual reasoning. These findings establish the temporal and semantic boundaries of perceived animacy, revealing both its contributions and limitations in the visually grounded intuitive understanding of social agency.

Atypical social information processing is a core characteristic of autism spectrum disorder (ASD), yet the mechanisms underlying these differences remain debated. Recent neurophysiological findings suggesting heightened responses to social stimuli have challenged traditional hypo-arousal accounts and instead support hyper-arousal frameworks, which makes a temporal prediction: social information should evoke enhanced early processing before its behavioral influence is subsequently attenuated. To provide direct evidence of this dynamic sequence, we employed an irrelevant change detection paradigm across five experiments combining behavioral measures, pupillometry, and electroencephalography (EEG). Behaviorally, although autistic children showed reduced interference from task-irrelevant social features at long encoding durations (1000 ms), they exhibited reliable interference at short durations (250 ms), indicating intact early automatic encoding. Pupillometry revealed enhanced early pupil dilation to social stimuli, indicating elevated physiological arousal, without evidence of spatial avoidance. These findings further suggest the engagement of a time-dependent regulatory process that reduces the impact of social representations at long durations: behavioral interference re-emerged when adopting highly salient emotional expressions, and EEG data showed increased theta-band activity in left temporo-occipital regions around 300 ms post-stimulus. Together, these findings support a dynamic account in which early heightened arousal is followed by rapid, localized regulation of social representations.

As AI shifts from a tool to a teammate, achieving effective human-AI collaboration becomes a central challenge. Human-AI teaming (HAT) promises to combine the strengths of humans and AI through close cooperation. However, recent studies reveal that mixed human-AI teams do not consistently outperform all-human or all-AI teams. The performance gap arises when AI does not align well with human cognitive processes. This highlights the need to understand human cognition better and to design teaming patterns that adapt to the nature of human-AI collaboration. To address this gap, we introduce a Three-Layer Cognitive Framework for HAT (TriC). It is composed of three layers: the task layer depicts human cognitive processes, serving as the core around which the entire framework is structured. The interaction layer classifies ten teaming patterns by combining initiator (human vs. AI) with five function-allocation strategies (Delegation, Parallel, AI-guided, User-guided, Collaboration). The mental layer analyzes how human mental structures, such as static psychological characteristics, shape teaming outcomes. By integrating these three layers, TriC is the first HAT framework to derive a complete, domain-general teaming pattern taxonomy directly from human cognitive processes. We further validate this taxonomy through systematic mapping of relevant empirical studies conducted to date, while deriving preliminary design guidelines for teaming pattern selection, and outlining future research directions.

Attention and working memory are thought to be closely linked, with the assumption that attended information can be remembered and reported within the time window of working memory. However, the phenomenon of attribute amnesia challenges this view by showing that participants fail to report a specific attended information (termed as the key feature), even though they had just used it for a task. Such observation raises questions about whether the report failure of key feature reflects a lack of working memory encoding for attended information. While previous studies have yielded contradictory results on this issue, the present study aimed to reconcile these seemingly conflicting findings by identifying future relevance as a critical factor driving the memory encoding of attended key features. The current results showed that when the key feature became irrelevant after being attended, it was not automatically encoded into working memory; however, when the key feature might become useful in future tasks, it was intentionally encoded into working memory. Overall, we concluded that the attended information could not automatically enter into working memory but would be actively encoded into working memory when expected to be useful in the future, supporting that working memory encoding is adaptive to future relevance. (PsycInfo Database Record (c) 2026 APA, all rights reserved).

Working memory and attention play pivotal roles in navigating the external environment and constructing our understanding of the world. Extensive research has demonstrated that items stored in visual working memory (VWM) can capture attention during visual search tasks, a phenomenon known as VWM-guided attention. However, a debate remains regarding the number of items that can guide attention. While early findings support a single-template limitation, more recent studies suggest that multiple items may guide attention when perceptually grouped. Yet this work has largely focused on static stimuli. In contrast, dynamic, temporally unfolding interactions, which are more representative of real-world perception, remain underexplored. Although previous research has shown that a single motion cue can capture attention, it remains unclear whether two separate motions, when bound by a meaningful dynamic relation, can function as a unified attentional template. In this study, we examine whether causality, a spontaneously perceived and ecologically valid dynamic relation, can restructure memory representations to modulate attention. We manipulated the presence of phenomenal causality using variations in time intervals, motion paths, and motion order. The results revealed that when causality existed between two memory items, singleton distractors matching either of the memory items could guide attention in visual search tasks, whereas this effect disappeared in the absence of phenomenal causality. These findings suggest that causality can indeed modulate VWM-guided attention.

Source amnesia refers to the failure to remember the source format of information despite remembering the content itself. While well-documented in long-term memory, recent studies have revealed that source amnesia can also occur in short-term or working memory. Across four experiments, the present study aimed to investigate why short-term source amnesia arises, focusing on whether it results from source misattribution between items or item-specific interference caused by repeated exposure to the same content in different formats. We found that source misattribution persisted even for a single item presented per trial, suggesting that item-source misbinding between simultaneously presented items is not necessary for source-amnesia effect. Source misattribution was significantly reduced when the test item was novel or had consistently appeared in a single format across trials, but reliably emerged when the same item had been presented in different formats. These findings suggest that short-term source amnesia reflects item-specific source misattribution, driven by the coexistence of conflicting source traces for the same content. We propose that the task-irrelevant source information for target stimuli is stored in an intermediate representational state-activated long-term memory-which maintains weak bindings to its content but lacks robust contextual indexing.

Despite widespread adoption of large language models (LLMs), our understanding of these competent assistants remains limited. Since personality serves as a key predictor of behavioral patterns, this study investigates whether LLMs have implicitly learned to display personality through a rigorous psychometric approach. We employed role-playing prompts to assess the "personality" of 100 fictional characters enacted by ChatGPT-4o and Claude Haiku 3.5 using two validated personality measures: a self-evaluation scale (NEO-FFI-R) and a situational judgment test (SJT). Performance was benchmarked against human participants (N = 113). Results revealed limitations in LLMs' personality coherence: both LLMs demonstrated inferior test-retest reliability to human participants, especially in terms of behavioral tendencies which showed particularly pronounced instability. More critically, construct validity analysis revealed false inference and inconsistencies between self-evaluated personality and behavioral tendencies. These discrepancies demonstrate that current LLMs are responding based on personality-related knowledge to pretend they have specific personality, rather than having a stable and consistent inner personality as humans.

Working memory (WM) capacity predicts a wide range of cognitive abilities, yet little is known about how individuals with different WM capacities regulate memory use during natural tasks in which memory use is voluntarily determined. To address this question, we conducted two experiments (N = 120) in which participants completed a copying task and a change-detection task. Experiment 1 manipulated sampling cost to increase reliance on internal memory storage, whereas Experiment 2 manipulated time pressure to increase demands for maintaining consistent memory use. Increasing sampling cost led participants to use more memory but did not make individual differences in WM capacity more apparent in mean memory usage. In contrast, time pressure revealed a reliable association between WM capacity and the consistency of memory usage across sampling episodes. Higher-capacity individuals exhibited more consistent memory use. These findings suggest that WM capacity shapes voluntary memory use not simply by increasing how much information people remember, but by influencing how consistently they regulate memory use under different task demands.

Joint action involves more than coordinated activity; it is cooperation grounded in shared intentionality, whereby partners represent an activity as something "we" are doing together. This "we-mode" stance should shape attention and memory, making partner-relevant information psychologically significant because it supports a collective goal. Using a joint-search paradigm, we tested whether people automatically attend to and remember partner goals. Pairs of participants searched for targets from different item categories, and trials were successful only when both responded correctly. A surprise recognition test followed the joint-search task assessing memory for the items. Across Experiments 1 (animate stimuli) and 2 (inanimate stimuli), participants showed better recognition of partner-goal items compared to distractors. Participants also showed enhanced attention to partner-goal items in Experiment 2. In Experiment 3, participants completed the same task, and returned three days later for a recognition test first followed by a second joint-search task with switched targets. Participants continued to show superior recognition for partner-goal items, and search efficiency improved after targets switched, indicating that partner-goal was retained over time and supported later cooperation. Together, these findings demonstrate that human cognition supports joint actions over time by organizing attention and memory around what "we" are doing together.

What determines the quality of human memory? Recent studies suggest that memory success depends on an intrinsic property of stimuli-memorability, which has predominantly been examined in item recognition tests. However, memory is an integrated system that involves not only individual elements but also the relational contexts in which they are embedded, an aspect that has been largely overlooked. In the present study, we investigate the impact of memorability on both item memory and relational memory. Through three experiments with 90 adult participants, we demonstrate that the memorability of stimuli not only facilitates item memory but also profoundly boosts their relational memory, encompassing temporal, spatial, and item-item binding memory. These findings provide robust evidence that memorability is a systematic phenomenon that goes beyond the stimulus itself, highlighting the broader relational nature of memorability. (PsycInfo Database Record (c) 2026 APA, all rights reserved).

The underlying mechanism of visual perspective-taking (VPT)-the ability to represent what others see-remains contested. Perceptual simulation theory proposes that VPT involves reconstructing others' visual experiences, whereas heuristic accounts argue that it relies on symbolic inference grounded in naïve optics. Evidence for heuristics largely comes from explicit report tasks, leaving open whether spontaneous (implicit) VPT in an agent-irrelevant task is driven by the same mechanism. A further possibility is that apparent "simulation failures" arise because observers lack prior visual information about what the other sees from their viewpoint. Across two experiments, participants performed an agent-irrelevant line-length judgment task while receiving plausible, absent, or implausible prior visual information from the agent's viewpoint. Experiment 1 showed a robust perspective-consistent bias under plausible priors, no bias without priors, and a weaker bias under implausible priors. A control experiment ruled out priming. Experiment 2 parametrically varied implausibility in a Ponzo-style layout and revealed a boundary condition: priors ranging from plausible to moderately implausible continued to bias judgments, whereas highly implausible priors were discounted. These results support a bounded, resource-rational heuristic account in which others' visual information acts as plausibility-weighted cues integrated with one's own visual input, rather than being reconstructed via perceptual simulation.

Previous studies have shown that sensory information matching the content of visual working memory (VWM) gains prioritized access into awareness. While these studies primarily focused on a single stimulus, it remains unclear whether the prioritization persists when multiple items are memorized. Using a breaking continuous flash suppression paradigm, the current study systematically investigated the time taken to detect a suppressed stimulus when two items were maintained in VWM. The results demonstrated that multiple items stored in VWM did not prioritize the matched stimuli into awareness, regardless of whether the stimuli presented during suppression were partially matched (Experiment 1) or fully matched (Experiment 2) to the memorized items. Furthermore, no prioritization was observed when the memorized items were either integrated into a single object (Experiment 3) or remembered with increased precision (Experiment 4). After confirming the validity of the current experimental paradigm (Experiments 5a and 5b), we found that the item assigned with a higher priority through a retro cue broke into awareness faster than the uncued and the new items (Experiment 6). These findings suggest that when multiple items are retained in VWM, only one single item stored in the active state can facilitate matched stimuli into awareness, indicating a limited capacity for the modulation of VWM on access to visual awareness. (PsycInfo Database Record (c) 2025 APA, all rights reserved).