The study, led by Sanaa N. Al-Haj Ali of the Department of Orthodontics and Pediatric Dentistry and Ra’fat I. Farah of the Department of Prosthetic Dental Sciences, moves beyond the familiar accuracy benchmarks that dominate AI evaluation in medical education. Accuracy alone, the authors argue, cannot characterize how AI models engage with assessment tasks. Workload — the cost of achieving a particular level of performance, emerging from the interaction between task requirements and the operator’s skills and perceptions — offers a complementary lens. In humans, working memory is limited in capacity and duration, and when task demands exceed this capacity, performance is compromised. An operator may maintain accurate performance at an unacceptably high mental cost, which is why performance and subjective workload must be measured independently. This dissociation makes workload a valuable evaluative dimension for AI systems, where models and students may achieve similar accuracy while experiencing vastly different processing demands.

To capture that dimension, the researchers turned to the National Aeronautics and Space Administration Task Load Index (NASA-TLX), the most widely used multidimensional subjective workload scale in human factors research. The original instrument comprises six subscales — Mental Demand, Physical Demand, Temporal Demand, Effort, Performance, and Frustration — each rated on a 21-point scale. For this study, the team selected three subscales tailored to the OSPDE context: mental demand as the primary indicator of cognitive processing, temporal demand given the timed, bell-ringer station format of the examinations, and frustration to probe whether AI models show any sensitivity to task-related affective load. Physical demand was excluded as inapplicable to both the examinations, which involve no psychomotor component, and to AI; performance was excluded to avoid overlap with measured accuracy; and effort was excluded because, in a task with no physical component, it reduces largely to mental effort. Subscales were analyzed individually rather than as a composite, consistent with the diagnostic approach recommended for identifying discrete workload sources.

The experimental setup was meticulous. Data were sourced from OSPDEs administered to twenty fourth-year dental students during a preclinical orthodontics and pediatric dentistry course in 2025. The question pool comprised seventy items across four formats: ten plain multiple-choice questions without visual support, and twenty short-answer questions, twenty multiple-choice questions, and twenty True/False questions, all supported by visuals such as clinical photographs, radiographs, illustrations, and tables. By clinical discipline, the pool included thirty-three pediatric dentistry items (47 percent) and thirty-seven orthodontics items (53 percent). Correct answers were validated against three standard reference texts, and all visuals passed a three-component quality checklist scored by a single faculty member whose intra-rater reliability exceeded Cohen’s kappa of 0.8.

Ten multimodal AI models were evaluated from three providers: four from Anthropic (Claude Sonnet 3.5, Claude Sonnet 4.5, Claude Haiku 4.5, and Claude Opus 4.6), three from OpenAI (GPT-5.2 Pro, GPT-5.1, and GPT-4o), and three from Google Gemini (Gemini 3 Pro, Gemini 2.5 Pro, and Gemini 2 Flash). All models were accessed via a paid subscription to POE between January 25 and March 5, 2026. A single investigator, blinded to study objectives and student performance data, inputted each question into the models via a new chat session to prevent context contamination, transcribing questions verbatim with visuals uploaded as JPEGs at 300 dpi using a zero-shot prompting approach. The team employed a two-step elicitation protocol: after providing its answer, the complete question–response pair was resubmitted to the model with a prompt requesting ratings on the modified NASA-TLX, each subscale rated on a seven-point scale from one (very low) to seven (very high). To assess intra-model response stability, each question was tested in three independent sessions, with final responses determined by majority rule and NASA-TLX ratings by median value across trials. Students completed the same modified NASA-TLX in print immediately following the examination, with a 100 percent response rate.

The accuracy results set the stage for the workload findings. Overall accuracy differed significantly across responders (chi-square = 95.309, p < .001): students achieved the highest score at 77.9 percent, followed by GPT-5.2 Pro at 74.3 percent and Gemini 3 Pro at 72.9 percent, while Claude Haiku 4.5 scored lowest at 44.3 percent. Only GPT-5.2 Pro, Gemini 3 Pro, Gemini 2.5 Pro, and Claude Opus 4.6 did not differ significantly from students. Performance gaps were largest on visual short-answer questions (chi-square = 44.676, p < .001), where several Claude models scored at or below 25 percent, and smallest on plain multiple-choice questions, where the difference was not statistically significant. Intra-model reliability, measured by intraclass correlation coefficients, ranged from 0.677 to 0.955 across all models and dimensions (all p < .001), with mental demand showing the highest consistency and frustration the most variability — a gradient the authors attribute to frustration capturing an emotional reaction that is inherently more state-dependent than ratings referencing concrete task properties.

The workload comparisons produced the study’s central contribution. Seven of ten models reported significantly higher mental demand than students, six reported significantly higher temporal demand, and nine reported significantly higher frustration. The greatest human–AI divergence occurred on visual short-answer questions, where nine of ten models reported significantly higher mental demand and all ten reported significantly higher temporal demand. Yet frustration told a different story on this format: only three models (Gemini 2.5 Pro, Gemini 3 Pro, and GPT-5.1) reported significantly higher frustration than students, compared with seven to nine models on other formats, despite visual short-answer questions producing the lowest AI accuracy. On plain multiple-choice questions, models performed comparably to students on both accuracy and mental demand, indicating that human–AI convergence is greatest when visual processing is absent and recognition-based options constrain response demands.

This dissociation between demand ratings and frustration on the hardest format is not paradoxical, the authors argue, once the cue sources underlying each dimension are distinguished. Mental and temporal demand are stimulus-facing judgments: their cues derive from observable task properties such as inferential steps, visual integration requirements, and time constraints. Frustration, by contrast, is a self-evaluative judgment that requires comparing output quality against some performance standard. In multiple-choice and True/False formats, fixed alternatives provide an implicit benchmark against which the model can gauge correctness; short-answer questions strip away this external scaffold entirely, leaving the model with impoverished evaluative cues. Koriat’s cue-utilization framework supports this interpretation: metacognitive judgments lose diagnostic sensitivity when available cues are restricted. Additionally, the higher processing demands on visual short-answer items may consume resources that would otherwise support self-evaluative reflection, analogous to Sweller’s finding that heavy cognitive load diminishes capacity for secondary processes. Hence, the models had ample cues to detect that the task was difficult, but insufficient cues to evaluate how poorly they performed.

The authors also caution against anthropomorphic readings of the frustration data. The NASA-TLX frustration item asks how insecure, discouraged, irritated, or stressed the respondent felt — wording that presupposes an affective state current models do not possess. A model’s response is therefore generated text about frustration rather than a report of an experienced emotion, and the researchers treat the affective subscale strictly as a behavioral output rather than as evidence of internal experience. A parallel caveat applies to temporal demand: language models generate output without an internal clock or representation of elapsed time, so the instruction that two minutes were available was processed as semantic content rather than experienced as time pressure. Elevated temporal demand ratings are therefore more plausibly an inference from task characteristics — item length, information density, reasoning steps, and visual parsing load — than a report of temporal stress. This mirrors the broader pattern: models tracked content-derived indices of difficulty while showing attenuated responses on the affective dimension, which carries no comparable task signal.

Correlational analyses between workload and accuracy revealed a predominantly inverse relationship among both students and AI models, though among students significant correlations were weak and confined to frustration (rs = −0.08, p < .01). Among AI models, the pattern was architecture-dependent: Claude Sonnet 4.5 and Claude Opus 4.6 showed significant negative correlations across all three dimensions overall, while GPT-4o, Claude Sonnet 3.5, and the Gemini models showed none. Claude Haiku 4.5 showed significance for temporal demand (rs = −0.35) and frustration (rs = −0.40), and GPT-5.2 Pro for temporal demand (rs = −0.30) and frustration (rs = −0.33). At the format level, significant negative correlations were most concentrated on visual multiple-choice and visual short-answer items and least on visual True/False items and plain multiple-choice questions, suggesting that the diagnostic sensitivity of AI workload ratings scales with format complexity. This variability has practical significance: models with significant negative load–accuracy associations and acceptable accuracy — notably Claude Opus 4.6 and GPT-5.2 Pro — could serve as automated item difficulty estimators, flagging disproportionately demanding questions before student administration during examination development.

The findings carry implications for educators designing assessments intended to differentiate human from AI performance. Because human–AI convergence was greatest on plain multiple-choice questions and diverged most on visually integrated open-ended formats, educators seeking to distinguish human from AI capability should favor visually integrated open-ended question types. The study also establishes workload profiling as a complementary dimension of AI evaluation that extends beyond accuracy, offering insight not only into whether AI can answer correctly but into how it represents the difficulty of doing so — and where that representation breaks down.

Several limitations warrant acknowledgment. AI answers were determined by majority rule across three sessions whereas students had a single attempt, an asymmetry that, for accuracy, can only maintain or improve an AI’s score relative to a single attempt — meaning the finding that students outperformed most models is conservative. The student sample comprised twenty participants from a single institution, representing the entire eligible cohort but limiting generalizability, and the seventy-item question pool was smaller than licensing examinations. Potential data contamination cannot be fully excluded, as the training corpora of commercial models are proprietary, though the authors note that the study’s central finding — a systematic cross-item pattern in which workload tracked question format and visual complexity with a consistent demand–affect dissociation — would not be expected from memorization of individual images. Whether AI workload ratings reflect genuine metacognitive self-monitoring or statistically patterned output remains open; the reliability and correlational data favor a structured interpretation but cannot establish introspective capacity. Future work, the authors suggest, should compare both parties under identical single-attempt conditions and explore whether format-specific time allocations reduce the human–AI gap on open-ended items.

Subject of Research: Science News

Article Title: NASA TLX workload profiles of multimodal artificial intelligence models and dental students during objective structured practical dental examinations

Article References: Al-Haj Ali, S. N., & Farah, R. I. (2026). NASA TLX workload profiles of multimodal artificial intelligence models and dental students during objective structured practical dental examinations. Discover Education, 5(1), Article 835. https://doi.org/10.1007/s44217-026-02116-2

Image Credits: AI Generated

DOI: 10.1007/s44217-026-02116-2

Keywords: artificial intelligence, NASA-TLX, workload, dental education, multimodal AI, large language models, metacognition, cognitive load, objective structured practical dental examinations, AI evaluation, NASA, profiles

The motivation behind the work lies in a long-standing gap between what language-learning software can measure and what actually happens in a learner’s mind and body. Most computer-assisted pronunciation and grammar tools evaluate linguistic output alone: phoneme accuracy, error counts, fluency scores. Yet second-language acquisition is a deeply embodied process. Speech rate fluctuates under stress, hesitation patterns reveal working-memory strain, gestures and posture shifts signal engagement or confusion, and emotions such as anxiety and enjoyment measurably influence fluency and persistence. A system that hears only the words misses most of the story. Tang’s framework is built on the premise that linguistic output, bodily behavior, and affective responses form a single, intertwined signal that must be modeled together.

Technically, ACG-CMT begins with three parallel encoding branches. The speech encoder does not receive polished pronunciation scores; instead it learns from raw acoustic-prosodic material—Mel-spectral patterns, energy contours, pitch and prosodic rhythm, and temporal features such as pauses—using a Mel-spectrum transform, one-dimensional convolutions, and bidirectional gated recurrent units. The gesture branch applies a spatiotemporal graph convolutional network to body-keypoint trajectories extracted from video, treating joints as graph nodes and their connections as edges, so that gesture amplitude, posture stability, and movement continuity become learnable features. The affective encoder combines facial action units captured by OpenFace with vocal-emotion cues, mapping both into a shared emotional latent space. Crucially, the study enforces a strict separation between predictor inputs and the variables used to construct supervision labels, reducing the risk that the model simply reconstructs its own targets.

Because speech, motion, and facial signals arrive at different sampling rates and evolve on different timescales, the framework next performs temporal alignment. All three encoded streams are resampled onto a fixed-length time window with positional encoding, preserving each modality’s internal order while placing corresponding behaviors on the same temporal axis. A cross-modal transformer then applies attention between modalities: the speech stream can query gesture and affect streams and vice versa, allowing a pause in speech to be linked with a simultaneous drop in gesture activity or a spike in negative vocal affect. Attention analysis of the trained model revealed non-uniform patterns—speech-driven attention to gesture and affect was locally concentrated, while gesture-to-affect attention spread across wider temporal ranges—suggesting the model captures genuinely structured cross-modal relationships rather than averaging everything together.

The architectural centerpiece is the affective-cognitive gate. Conventional multimodal fusion typically assigns fixed weights to each modality after training, but the importance of speech versus gesture versus emotion changes from moment to moment. When a learner mispronounces a word, acoustic features matter most; when comprehension breaks down, facial and vocal affect carry more diagnostic weight; when engagement collapses, posture and gesture become the key evidence. The gating unit generates normalized weights for the three modalities conditioned on latent affective and cognitive context and on pre-decision task context—all computed from information available at inference time, with no access to ground-truth labels. The resulting fused representation feeds a multi-task decision head that simultaneously performs learning-state recognition, affective-cognitive load estimation, and feedback-policy matching.

The empirical evaluation involved 96 non-English-major undergraduates aged 18 to 23, each with at least six years of English study. In a controlled indoor setting, a headset microphone recorded speech at 16 kHz while a front-facing camera captured video at 30 frames per second; interaction logs recorded task types, response times, and feedback events. From 3,840 task segments, the researchers generated 12,864 synchronized five-second analysis windows with 50 percent overlap. Human annotators labeled five learning states—normal, pronunciation deviation, hesitation, comprehension difficulty, and low participation—with strong inter-rater agreement (Cohen’s kappa of 0.824). English teachers independently assigned feedback labels across five categories: error correction, prompting, pacing adjustment, difficulty adjustment, and emotional support. Data were split at the learner level into training, validation, and test sets in a 7:1:2 ratio to prevent overlapping windows from leaking between sets.

The results are striking. ACG-CMT achieved 0.892 accuracy, 0.884 macro-F1, 0.941 AUC, and 0.877 difficult-state F1 for learning-state recognition, outperforming traditional machine-learning baselines such as XGBoost (0.768 accuracy), single-modal deep models, and even recent multimodal transformers (0.876 accuracy for the strongest competitor). For the operational affective-cognitive load reference index—a composite built from NASA-TLX workload ratings, reaction time, pause ratio, and task difficulty—the model reached an R-squared of 0.836 with RMSE of 0.084 and MAE of 0.068. Feedback-policy matching hit 0.861 with an average latency of just 438 milliseconds, fast enough for genuinely interactive tutoring. Ablation experiments underscored the design: removing cross-modal attention cost the most performance (macro-F1 dropping to 0.801), followed by removing the affective-cognitive gate (0.812), while removing the feedback module mainly degraded policy matching rather than state recognition.

Perhaps the most intriguing findings concern interpretability. The gating weights shifted systematically across learning states: under normal expression the three modalities stayed balanced between roughly 0.32 and 0.35; pronunciation-deviation samples shifted weight toward speech; comprehension-difficulty samples shifted toward affect; and low-engagement samples leaned on gesture and emotional information. Precursor signals also emerged before feedback events: in the five seconds preceding a feedback trigger, the pause ratio climbed from 0.214 to 0.414 and the load-risk estimate rose from 0.341 to 0.721. Feedback delivered within a 300-to-500-millisecond window produced the highest state-recovery gain (0.231), compared with 0.148 for faster responses and a meager 0.079 when feedback arrived after 900 milliseconds—evidence that timing, not just content, shapes the effectiveness of adaptive tutoring.

Tang is careful about what these patterns do and do not prove. The gating trajectories and attention maps are described as model-level associations, not causal explanations of human cognition, and the composite load index is explicitly framed as an operational reference rather than a validated physiological measure of cognitive load—no neurophysiological signals such as pupillometry were collected. The overlapping windows mean the 12,864 samples are not statistically independent, and the participant pool of Chinese undergraduates limits generalizability. Still, the study marks a meaningful shift from outcome-oriented assessment toward process-level perception: a tutoring system that reads the whole learner, not just the transcript. Future work, the author notes, should test the framework with more diverse populations, in open-ended and longitudinal learning environments, and against independent physiological measures—steps that will determine whether emotionally and bodily aware AI tutors can move from the laboratory into the classroom.

Subject of Research: Embodied multimodal human-computer interaction using speech, gesture, and affective signals for adaptive English language learning

Article Title: Embodied multimodal human-computer interaction for English language learning: integrating speech, gestures, emotion, and adaptive feedback

Article References: Tang, Y. (2026). Embodied multimodal human-computer interaction for English language learning: integrating speech, gestures, emotion, and adaptive feedback. Frontiers in Neurorobotics, 20, Article 1920739. https://doi.org/10.3389/fnbot.2026.1920739

Image Credits: AI Generated

DOI: 10.3389/fnbot.2026.1920739

Keywords: multimodal AI, affective computing, cross-modal transformer, English language learning, embodied cognition, adaptive feedback, cognitive load, gesture recognition, speech prosody, learning-state recognition, human-computer interaction, neurorobotics

The research team, led by Bartłomiej Król-Józaga of AGH University of Krakow together with colleagues at Dartmouth College and Jagiellonian University, tackled a long-standing gap in computational neuroscience. Most existing models of decision making simulate only one of the many parallel operations the brain performs—reward processing, evidence accumulation, strategy selection, or option choice—rather than the whole coordinated process. An earlier spiking neural network model had attempted a more holistic simulation, but it ignored a key biological feature: the powerful role that inhibitory neural connections play in shaping cognition.

Inhibition is far more than a neural brake. Inhibitory interneurons synchronize the brain oscillations that coordinate the networks involved in choice, filter out irrelevant information, and refine neural responses so that behavior can adapt to varying cognitive demands. This mechanism supports the interplay between brain regions central to decision making: the orbitofrontal cortex (OFC), which binds sensory input with value signals; the dorsolateral prefrontal cortex (dlPFC), which accumulates and integrates information over time; and the basal ganglia, which select the best option. Without inhibition, neural circuits risk runaway excitation—the kind of pathological over-activation seen in conditions such as epilepsy and schizophrenia.

To test whether inhibition could explain how people choose decision strategies, the team used a well-studied experimental paradigm called the sequential probabilistic inference task. Twenty-four participants, aged 21 to 27, memorized the importance weights of six attributes and then viewed the binary values of two choice options, one attribute at a time. On each trial they could either pick option A, pick option B, or request more information. Crucially, the task came in two flavors: a Non-Compensatory condition, where one attribute dominates and a simple heuristic called Take The Best (TTB) works well, and a Compensatory condition, where all attributes carry similar weight and the rational Weighted Additive rule (WADD)—which sums weighted evidence across everything—is the smarter strategy.

The model itself was built within the Neural Engineering Framework (NEF), a mathematical methodology for constructing large-scale neural models with realistic spiking neurons. In the NEF, each neuron is characterized by an encoder, a gain, and a bias that together define its tuning curve—how strongly it fires in response to an input signal. Filtered spike trains are decoded with optimized weights to reconstruct the represented signals, and transformations such as value-weight multiplication are implemented through learned connection strengths. Dynamics, the third NEF principle, captures how neural systems evolve over time, enabling the model to reproduce mechanisms like working memory, evidence accumulation, and choice competition.

The architecture mirrors known brain anatomy. Cue values flow through perception and memory components into the model’s OFC population, where they are combined with stored importance weights. The weighted values then feed into a dlPFC evidence population whose recurrent connections act as an integrator, accumulating information across time steps. A dlPFC contrast population computes the difference between the integrated values of the two options, sharpening the distinction before signals reach the basal ganglia. There, a Winner-Take-All mechanism—implemented through the striatum, globus pallidus, and subthalamic nucleus—selects the option with the highest accumulated evidence, triggering a motor response once the decision threshold is reached.

The critical innovation was a neuromodulatory pathway inspired by the locus coeruleus-norepinephrine (LC-NE) system, a brainstem nucleus famous for regulating attention, arousal, and cognitive effort. In the model, a neuromodulatory population receives input from the accumulating dlPFC evidence—interpreted as a marker of cognitive load—and sends inhibitory connections to the OFC. As accumulated evidence grows, this pathway progressively suppresses the incoming cue-value and weight products, simulating how effort-related arousal sharpens the selectivity of pre-decisional information processing. The strength of this inhibition is captured by a single participant-specific parameter, fitted to each individual’s behavioral data alongside parameters for decision threshold, urgency scaling, and lateral-inhibition gain.

The validation strategy was deliberately demanding. The researchers fitted both model variants—the standard version and the inhibition-enhanced version—only to the Non-Compensatory condition, then transferred the frozen parameters to the structurally different Compensatory condition as an out-of-distribution test. This design served a dual purpose: it directly probed whether the load-sensitive inhibitory mechanism, calibrated under low cognitive load, could predict behavior under higher load, and it acted as a built-in control for model complexity. A model that merely exploited an extra free parameter to overfit the training data should generalize worse, not better, to unseen conditions.

The results were striking. On the training condition, the two models performed comparably, with no statistically significant difference. But on the held-out Compensatory data, the inhibition model achieved approximately 17 percent lower mean prediction error than the no-inhibition model—0.55 versus 0.67—a difference confirmed by a Wilcoxon signed-rank test (p = 0.009, rank-biserial correlation r = 0.55), with the inhibition model winning in 16 of 24 participants. Because the advantage appeared precisely where added flexibility should have hurt rather than helped, the finding argues strongly against a mere complexity artifact. Parameter importance analysis reinforced this picture: the inhibition strength ranked second on average, but with high inter-individual variability, being decisive for some participants and negligible for others—hinting that arousal-based inhibition shapes decision behavior meaningfully in only a subset of people.

Intriguingly, the inhibition advantage showed up in the continuous pattern of information sampling rather than in categorical strategy shifts. Both model variants reproduced the characteristic bimodal shape of human cue acquisition—people tend to either sample very few cues or exhaust all six—and both predicted similar WADD-versus-TTB strategy distributions across conditions. What inhibition bought was a quantitatively more accurate replication of how many attributes people consulted under demanding conditions, consistent with the idea that neuromodulation coupled with inhibition grades the selectivity of information gathering rather than flipping the brain between discrete strategies. The authors note that future work should refine the cognitive-load signal to produce realistic trial-by-trial buildup, validate neural trajectories against pupillometry and fMRI data, and explore the role of other neuromodulators such as dopamine. For now, the study offers compelling evidence that the brain’s inhibitory circuitry is not just noise suppression—it is an active, load-sensitive mechanism that helps us decide how to decide.

Subject of Research: The role of neural inhibition in decision strategy selection modeled with spiking neural networks

Article Title: The role of inhibition in modeling decision making with spiking neural networks

Article References: Król-Józaga, B., Duggins, P., Broniec-Wójcik, A., & Wichary, S. (2026). The role of inhibition in modeling decision making with spiking neural networks. Frontiers in Neuroinformatics, 20, Article 1854811. https://doi.org/10.3389/fninf.2026.1854811

Image Credits: AI Generated

DOI: 10.3389/fninf.2026.1854811

Keywords: spiking neural networks, decision making, neural inhibition, orbitofrontal cortex, locus coeruleus, norepinephrine, cognitive load, heuristics, Neural Engineering Framework, basal ganglia, evidence accumulation, computational neuroscience

The problem, the authors contend, begins with how immersion is conceptualized. VR studies across memory, attention, and executive functions have reported effects ranging from robust improvements to complete null findings. Differences in participant characteristics, intervention intensity, and task design explain part of this inconsistency, but they do not address a deeper issue: the same immersive feature that provides task-relevant support in one intervention can impose unnecessary demands in another. Navigation, spatial orientation, interface management, and even physical discomfort can consume cognitive resources that would otherwise serve the primary task. Cognitive benefit, the authors argue, cannot be inferred from engagement alone; it depends on whether task-relevant support outweighs task-irrelevant processing demands.

At the heart of the argument are two experiential constructs central to immersive VR: presence and agency. Presence is the subjective sense of being there in a mediated environment, while agency is the capacity for meaningful action with contingent feedback. Together, they can strengthen task-relevant processing by embedding cognition within interactive, spatially structured contexts. But the authors emphasize that these constructs must be measured separately, because presence without agency may produce absorption without task-relevant action, whereas agency without presence resembles little more than a conventional interactive display. Neither alone guarantees that the cognitive process being targeted is actually being exercised.

The researchers organize their argument around three propositions. First, immersion is conditional: it supports performance only when its task-relevant contribution exceeds the extraneous load it adds. Second, immersion is directional: a feature’s effect depends on its alignment with the targeted process. Spatial organization, for example, may support spatial encoding but divert resources when incidental to the mechanism being trained. Third, immersion can either scaffold or substitute for cognition, thereby shaping transfer. Cues, prompts, and feedback support learning when they preserve task-relevant challenge, but they can produce nontransferable gains when they perform the learner’s cognitive work rather than strengthen internal control.

To operationalize these ideas, the team proposes a two-dimensional decision matrix defined by immersive engagement and resource cost. Engagement reflects the joint contribution of presence and agency, while resource cost includes extraneous cognitive load and physiological discomfort such as cybersickness. Crucially, the effort required to exercise the target process is not counted as a cost—it is part of the intended training demand. Navigation, interface management, or irrelevant environmental complexity becomes costly only when it competes with that demand. The matrix yields four quadrants with distinct, testable predictions. High engagement with low cost provides the strongest case for benefit, with the best prospects for mechanism-relevant performance, transfer, and maintenance. High engagement with high cost may support immediate gains but is expected to yield weaker or less consistent transfer, along with greater fatigue, discomfort, or attrition. Low engagement with low cost should add little beyond a matched nonimmersive condition, while low engagement with high cost is the least favorable combination, adding burden without task-relevant immersive support.

The framework’s implications become concrete when applied to specific cognitive domains. In memory research, the comparison condition and the outcome measured matter enormously. One study found better recall with an immersive virtual memory palace than with a desktop version, yet a randomized controlled trial in older adults with subjective memory complaints found no added transfer benefit of immersive VR practice after method-of-loci training relative to an active VR control. These findings are not contradictory, the authors note: VR may outperform a less immersive format yet add little beyond an effective mnemonic strategy already in place. What counts as benefit depends on what VR is compared with and whether the outcome reflects trained-task performance, transfer, or long-term maintenance.

Mechanistically, immersive VR is most likely to support memory when its spatial and sensorimotor features strengthen context-dependent encoding. A virtual memory palace may improve recall by linking items to distinct locations, and presence may enhance encoding when those cues are salient and task-relevant. But the same spatial features can become costly when navigation or environmental complexity is incidental to encoding and consumes working memory without supporting retention. Because such encoding is context-dependent, near transfer—improvement on tasks resembling the trained task—is the most plausible outcome. Ecological transfer, meaning improvement in real-world functioning, becomes plausible only when training cues map onto real-world contexts, while far transfer and long-term maintenance require direct evidence.

The attentional and executive control domain illustrates similar tensions. Immersive tasks can embed planning, inhibition, updating, and reorientation within naturalistic, multi-step scenarios—a virtual grocery task, for instance, can require route planning, inhibition of distractions, goal updating, and reorientation as conditions change. Some evidence supports the potential: one study found greater gains in inhibition and task switching among adults aged 50 and older when the same exercise game was delivered in an immersive rather than nonimmersive format, with presence mediating these effects. An exploratory study of young adults reported improvements in selective attention and inhibition after VR exergaming, whereas video exercise produced no such gains. In contrast, another study found that immersive 360-degree presentation increased presence and arousal but actually impaired focused attention and recall relative to screen presentation. Together, these findings suggest that immersion can support selected control processes when task demands engage them, but effects are process-specific and depend heavily on the comparison condition; greater experiential intensity alone does not establish stronger attentional-executive control.

The authors also draw on established theory to sharpen their framework. The Cognitive Affective Model of Immersive Learning, or CAMIL, explains how immersive affordances influence learning, but training performance can improve without durable learning or capacity change. For capacity training, this dissociation is critical: a feature that makes a task easier may raise performance while removing the very cognitive work the intervention is meant to strengthen. The researchers therefore extend CAMIL with three intervention-specific requirements: immersive features should align with the target mechanism, require users to exercise rather than offload that process, and produce benefits beyond VR-supported performance. Embodied cognition helps identify when sensorimotor engagement is mechanism-relevant, while cognitive load theory helps identify when immersive demands are extraneous.

Ultimately, the authors call for a new standard of evidence in immersive cognitive research. Future studies should manipulate specific immersive features using active comparisons matched for task content and expectancy; measure experiential engagement, resource costs, mechanism-specific cognition, and outcomes within the same study; preregister moderator hypotheses concerning age, baseline capacity, task alignment, and cybersickness susceptibility; and distinguish immediate performance from near, ecological, and far transfer as well as delayed maintenance, with transparent attrition reporting. These standards move the field from asking whether VR is broadly beneficial to testing when, for whom, and through which mechanisms immersion produces durable cognitive change. They also make negative or mixed findings more informative, because they help identify whether a weak effect reflects poor mechanism alignment, excessive cost, population-specific vulnerability, or inadequate transfer assessment rather than a generic failure of VR. On these terms, immersion is not a guarantee of benefit but a design parameter whose value must be demonstrated—mechanism by mechanism, population by population.

Subject of Research: Distinguishing cognitive change from VR-supported performance in immersive cognitive training

Article Title: Beyond immersion: distinguishing cognitive change from VR-supported performance

Article References: Yang, H., Kho, H., Tay, C. P. E., & Eun, Z. K. Y. (2026). Beyond immersion: distinguishing cognitive change from VR-supported performance. Frontiers in Virtual Reality, 7, Article 1815501. https://doi.org/10.3389/frvir.2026.1815501

Image Credits: AI Generated

DOI: 10.3389/frvir.2026.1815501

Keywords: virtual reality, cognitive training, presence, agency, cognitive load, transfer of learning, memory, executive function, cybersickness, immersive learning, CAMIL, neuroscience

The research team asked a deceptively simple question: if you strip away all the text from picture-book versions of the same story, which illustrations do young children judge as making the story easiest to follow, and which version do they choose? To find out, the researchers sourced illustration versions of Little Red Riding Hood and The Three Little Pigs from four different publishing markets. For each story, they selected three key plot segments and presented them to children in a repeated-measures forced-choice experimental design, with all textual information removed so that only the pictures carried the narrative.

The method was carefully controlled. Each child saw the market-sourced illustration versions side by side and made forced-choice judgments about which version made the story most comprehensible. Because every participant evaluated the same set of versions across the same plot segments, the design allowed the researchers to compare choices directly. Version-choice distributions were then examined using descriptive statistics and chi-square tests, with supplementary mixed ANOVAs used to explore how judgments varied by story segment, age, and gender.

The central finding was striking: children’s selections clustered around particular versions rather than being evenly distributed across the options. If illustration style, color, or cultural origin alone drove preference, choices would have spread more uniformly. Instead, the concentration of choices indicates that specific visual organization strategies—how clearly illustrations arrange action cues, emotional signals, and relations between events—systematically influenced children’s perceived ease of understanding.

This result aligns with cognitive load theory, which holds that working memory can be overwhelmed when information is presented in a cluttered or poorly structured way. Illustrations that organize narrative information efficiently reduce the mental effort needed to reconstruct a story. In this view, preschoolers’ judgments reflect less the sheer visual richness of an illustration than how well it structures the narrative—clearly depicting who did what, what they felt, and how one event leads to the next.

The study also probed developmental variation. Exploratory analyses of gender differences were generally non-significant, suggesting that boys and girls judged comprehensibility similarly. Age-related variation, however, was selective and segment-dependent: differences between younger and older preschoolers appeared only for certain plot segments, not across the board. This pattern suggests that as children develop, their sensitivity to how illustrations convey narrative information sharpens, but only for story moments that demand particular interpretive work—such as understanding a character’s intention or linking cause to effect.

Importantly, the authors urge caution about overinterpreting the findings culturally. Because the task measured perceived ease of understanding rather than direct cultural interpretation, claims about children noticing or responding to cultural features in illustrations from different publishing markets should be treated as provisional. The study demonstrates that visual presentation strategies shape comprehensibility judgments; it does not establish that children were consciously evaluating cultural authenticity or origin.

The work builds on a substantial research literature. Previous studies have shown that illustrations support children’s inferential comprehension, that eye-tracking experiments reveal how pictures in picture storybooks scaffold story comprehension, and that visual narrative structure itself follows a kind of grammar that the brain processes. The new study adds a comparative dimension by sourcing versions from four publishing markets and showing that preschoolers’ judgments converge on versions whose illustrations organize narrative information most clearly.

For educators, parents, and publishers, the implications are practical. When selecting picture books for young children, the sheer amount of visual detail matters less than whether the illustrations clearly organize action, emotion, and event relations. A visually rich but narratively confusing illustration may actually hinder a preschooler’s sense of story understanding, while a simpler version that clearly depicts who did what and how events connect may make the same story feel far easier to follow.

Future research could extend these findings by examining whether children’s judgments of comprehensibility predict actual story comprehension, by testing children from different cultural backgrounds, and by investigating which specific visual organization strategies—such as spatial layout, character positioning, or emotional expression—carry the greatest weight. For now, the study offers a clear message: preschoolers are discerning critics of visual storytelling, and their judgments hinge on how well illustrations organize the narrative, not on how much visual detail they contain.

Subject of Research: Developmental psychology and early childhood education

Article Title: How Illustration Presentation Strategies Across Picture-Book Versions from Four Publishing Markets Shape Chinese Preschoolers’ Judgments of Story Comprehensibility and Version Choice

Article References: Xu, C., & Fu, Y. (2026). How Illustration Presentation Strategies Across Picture-Book Versions from Four Publishing Markets Shape Chinese Preschoolers’ Judgments of Story Comprehensibility and Version Choice. Early Childhood Education Journal. https://doi.org/10.1007/s10643-026-02356-9

Image Credits: AI Generated

DOI: 10.1007/s10643-026-02356-9

Keywords: picture books, preschoolers, illustrations, story comprehension, visual narrative, cognitive load, child development, early childhood education, China, reading development, Illustration, Presentation

The starting point of the review is a mismatch between two research traditions. Cognitive load theory, developed since the 1980s, describes how working memory limits shape learning and problem solving, while classical stress research from Hans Selye onward describes how organisms respond to threats through generalized physiological activation. Information overload research, in turn, has documented performance breakdowns under excessive input, but has largely treated these as information-processing failures rather than stress phenomena. Asriyan and Gevorgyan argue that this separation has obscured a key possibility: that sustained informational demands act on the body through the same allostatic machinery that governs responses to other chronic stressors.

Central to the framework is the concept of allostasis, the active process by which the body maintains stability through change. Building on work by Bruce McEwen and colleagues, the authors distinguish acute adaptive responses from allostatic load, the cumulative wear and tear that results when adaptive responses are repeatedly engaged or poorly regulated. They propose that informational stress operates in this same register: when informational demands are persistent, unpredictable, or perceived as uncontrollable, the regulatory effort they require can accumulate into a physiological burden comparable to that produced by other chronic stressors.

A distinctive feature of the proposal is its insistence that informational stress is not simply proportional to the amount of information received. Drawing on information theory, the authors argue that properties such as uncertainty, entropy, and the rate of change of incoming signals determine how much regulatory work a given stream of information demands. Two environments with the same volume of data can impose very different loads depending on their structure, and the authors suggest that these structural properties can in principle be quantified using tools from information theory rather than relying solely on subjective reports of overload.

The framework also incorporates appraisal, the evaluative process that determines whether a demand is experienced as threatening or manageable. Drawing on the transactional model of stress developed by Richard Lazarus and Susan Folkman, Asriyan and Gevorgyan propose that the same informational stream can be experienced as stimulating or as stressful depending on perceived control, relevance, and available coping resources. This appraisal step explains why informational stress is not a uniform response to volume alone, and why individual and contextual differences matter so much in how overload is experienced.

To connect these ideas to measurable outcomes, the review outlines a set of potential indicators spanning multiple levels. At the cognitive level, these include attentional narrowing, working memory strain, and decision fatigue. At the affective level, they include anxiety, irritability, and the feeling of being overwhelmed. At the physiological level, the authors point to established stress markers such as cortisol, heart rate variability, and inflammatory indicators, alongside more specific measures of cognitive effort such as pupil dilation and neural markers of cognitive control. The framework proposes that these levels are linked through the allostatic process, so that sustained informational demands leave traces across all of them.

The authors are careful to distinguish informational stress from adjacent concepts. It is not the same as cognitive load, which describes momentary demands on working memory during specific tasks. It is not the same as information overload, which describes a condition of excessive input. And it is not the same as digital stress or technostress, which are typically tied to specific technologies. Instead, informational stress is defined by its process character: it emerges when informational demands are sustained, appraised as taxing, and met with regulatory responses that themselves carry costs. This positioning is intended to make the concept useful across contexts, from workplace settings to education to the design of information systems.

The review also addresses recovery, a dimension the authors argue has been neglected in overload research. If informational stress operates through allostatic mechanisms, then recovery periods are not merely pauses but active parts of the stress process, during which physiological systems return toward baseline. The framework suggests that insufficient recovery between informational demands may be a key mechanism by which overload translates into lasting strain, and it calls for research designs that track these dynamics over time rather than capturing overload only in single snapshots.

Asriyan and Gevorgyan frame their contribution as a bridge rather than a replacement. They do not claim that informational stress is a wholly new phenomenon, but rather that existing findings from cognitive load research, stress physiology, and information theory can be integrated into a single process-level account that generates testable predictions. Among these are predictions about when informational demands will produce physiological strain, which structural properties of information streams are most likely to drive that strain, and how individual differences in appraisal and coping shape vulnerability.

The practical implications of the framework are potentially wide-ranging. In workplace contexts, it suggests that interventions should target not only the volume of information but also its structure, pacing, and the sense of control workers have over it. In education, it points to the importance of managing informational demands in ways that support rather than overwhelm allostatic regulation. And in the design of digital environments, it offers a rationale for building systems that respect the recovery needs of their users. The authors emphasize that the framework is intended to be tested, refined, and potentially revised, and that its success will be measured by whether it generates research questions that existing models have not.

Subject of Research: A conceptual review proposing informational stress as a framework connecting information overload to allostatic physiological strain

Article Title: Informational stress in cognitive systems: a review and conceptual synthesis

Article References: Asriyan, E., & Gevorgyan, R. (2026). Informational stress in cognitive systems: a review and conceptual synthesis. Frontiers in Network Physiology, 6, Article 1913737. https://doi.org/10.3389/fnetp.2026.1913737

Image Credits: AI Generated

DOI: 10.3389/fnetp.2026.1913737

Keywords: informational stress, information overload, allostatic load, cognitive load, stress physiology, information theory, Frontiers in Network Physiology, Informational, stress, cognitive, systems, review

The research team, led by Sidney Afonso Sobrinho Jr. and corresponding author Gustavo Christofoletti, recruited 164 participants, split evenly into 82 young adults and 82 older adults. Each volunteer completed a standardized walking protocol under three carefully controlled conditions: a single-task walk with no added demands, a motor dual-task walk in which participants carried a weight, and a cognitive dual-task walk in which they performed a semantic verbal fluency exercise—naming words that belong to a given category—while walking. The design mirrors situations that fill ordinary life, from hauling groceries down a hallway to chatting and thinking while crossing a street.

What sets this study apart technically is its measurement approach. Rather than relying on stopwatches or observation, the researchers fitted participants with wearable motion sensors capable of capturing the fine architecture of each stride. Four parameters formed the backbone of the analysis: stride length, the distance covered by a single full gait cycle; cadence, the number of steps taken per minute; gait speed, the overall velocity of travel; and foot-strike angle, the angle at which the foot meets the ground, which reflects how a person positions the ankle and foot at the moment of contact. Together, these variables provide a sensitive fingerprint of walking quality that can reveal subtle compensations invisible to the naked eye.

The statistical framework was equally deliberate. Between-group comparisons, testing whether older and young adults differed at each condition, were performed using the Mann–Whitney U test, a non-parametric method suited to data that may not follow a normal distribution. Within-group comparisons, tracking how each participant’s gait changed across the three conditions, used the Friedman test, the non-parametric equivalent of a repeated-measures analysis. Significance was set at the conventional 5 percent threshold. The result was unambiguous: gait parameters differed between older and young adults across all experimental conditions, with p-values below 0.001 for every comparison. Aging, in other words, left its mark on walking regardless of whether the walk was simple or complicated.

The most striking finding emerged when the two dual-task conditions were pitted against each other. The semantic verbal fluency condition produced larger changes in gait performance than the weight-carrying condition in both age groups. This is a meaningful departure from the intuition that physical burdens should dominate the biomechanics of walking. Instead, the data indicate that when the brain must divide its attention between locomotion and language retrieval, the cost to the walking pattern is greater than when the body must simply bear an extra load. Cognitive load, it appears, competes more fiercely with the neural resources that keep gait smooth and efficient.

The numbers tell the story with precision. The cognitive dual-task cost—the percentage decline in gait performance when thinking was added to walking—was 4.3 percent in young adults but 7.1 percent in older adults. That gap of nearly three percentage points may sound modest, but in gait science such differences are clinically meaningful, because even small reductions in walking efficiency compound over thousands of steps and can translate into slower transit, greater fatigue, and, in vulnerable populations, elevated fall risk. The study demonstrates that the aging nervous system pays a steeper price for divided attention, even when the walking surface is safe and predictable.

The motor dual-task results were, if anything, more surprising. In young adults, carrying a weight did not affect gait performance at all—their walking patterns remained stable under the added physical demand. Older adults, however, did not slow down or shorten their strides as a protective strategy; instead, they completed the walking course in less time, yielding a negative dual-task cost of −4.6 percent. A negative cost means performance under the dual task was, by this measure, better than during the single task. The authors characterize the response to the motor dual-task as stable gait performance in young adults and shorter course completion time in older adults, a pattern that may reflect compensatory pacing or heightened engagement when a physical task demands attention.

Interpreting that negative cost requires care, and the researchers are explicit that their conclusions apply under the specific experimental conditions they tested. The weight-carrying task and the verbal fluency task differ not only in modality—motor versus cognitive—but in how they capture attention and how participants prioritize them. An older adult hurrying while carrying a load may be adopting a strategy that preserves overall course time at the expense of other unmeasured aspects of gait quality. Still, the direction of the effect challenges the simple assumption that any added burden degrades older adults’ mobility uniformly, and it highlights how differently the young and aging systems allocate resources when the challenge is physical rather than mental.

The public health implications are central to the paper’s purpose. Dual-task walking is not a laboratory curiosity; it is the default condition of daily life. Crossing a road while conversing, navigating a crowded market while planning a shopping list, or carrying laundry while thinking about the day ahead all require the concurrent allocation of motor and cognitive resources. Because the study shows that cognitive demands extract a larger and more age-sensitive toll on gait than physical ones, routine gait assessments that test only single-task walking may substantially overestimate real-world mobility in older adults. The authors argue that incorporating dual-task conditions into gait assessments may help identify age-related functional vulnerabilities before they manifest as falls or loss of independence.

Such screening could reshape preventive care. If a simple wearable-sensor protocol, paired with a verbal fluency challenge, can flag older adults whose walking deteriorates disproportionately under cognitive load, clinicians could target those individuals with tailored interventions—balance training, cognitive-motor exercise programs, or environmental modifications—long before a crisis. The study, funded by the Scientific Foundation of Mato Grosso do Sul, the National Council for Scientific and Technological Development, CAPES, and the Federal University of Mato Grosso do Sul, and approved by the Research Ethics Committee of the Federal University of Mato Grosso do Sul under approval number 7.465.300, was conducted in accordance with the Declaration of Helsinki, with all participants providing written informed consent. Its open-access publication means the evidence is freely available to the practitioners who need it most. As populations worldwide age, understanding that the mind burdens the aging stride more than the muscles do may prove one of the simplest, most actionable insights in modern mobility science.

Subject of Research: Effects of motor and cognitive dual tasks on gait performance in young versus older adults

Article Title: Motor and cognitive dual tasks differentially affect gait performance across age groups: implications for public health

Article References: Afonso Sobrinho, S., Jr., Oliveira-Junior, S. A., Moretti Luchesi, B., da Silva Alexandre, T., Sebastião, E., & Christofoletti, G. (2026). Motor and cognitive dual tasks differentially affect gait performance across age groups: implications for public health. BMC Public Health. https://doi.org/10.1186/s12889-026-29856-y

Image Credits: AI Generated

DOI: 10.1186/s12889-026-29856-y

Keywords: dual-task walking, gait performance, aging, cognitive load, verbal fluency, wearable sensors, cadence, stride length, public health, geriatrics, mobility, fall risk

The experiment enrolled 334 eligible college students, who were systematically allocated in chronological order according to a repeating A-B-C sequence rather than random assignment. The completer analysis ultimately included 193 participants: 58 students in Group A, who received the audio-dominant nonverbal audiovisual intervention; 57 students in Group B, who received the same audiovisual materials with guided narration; and 78 students in Group C, a waitlist control. Attrition was substantial and roughly similar across the two active arms: 47.7 percent of Group A and 48.6 percent of Group B were not included in the final analysis, along with 30.4 percent of the control group. Baseline records were available for all 334 allocated participants, but post-intervention scores on the outcome scales were available only for the 193 completers, and the reasons for non-completion were not recorded. The authors were transparent about these limitations, noting that no imputation-based intention-to-treat analysis or missing-outcome sensitivity analysis was performed.

The intervention itself was delivered through a WeChat mini-program, a lightweight application embedded in the messaging platform that dominates daily digital life in China. For four weeks, participants in the active groups engaged daily with 15-minute nature-themed sessions built around four scenarios: Ocean, Forest, Flower Garden, and Candlelight. Group A experienced these scenes in an audio-dominant, nonverbal form, in which sound carries the therapeutic payload and visuals support it without spoken guidance. Group B received the identical materials with guided narration layered on top. This pairing of conditions constitutes what the researchers call a subtraction design: by removing the narration from one arm, they could isolate the contribution of verbal guidance and test whether the nonverbal core of the program stands on its own.

Outcomes were measured with two well-established self-report instruments. The primary outcome was the Self-Rating Anxiety Scale, or SAS, and the secondary outcome was the Self-Rating Depression Scale, or SDS. Both instruments ask respondents to rate a series of symptoms and yield scores that rise with symptom severity. The authors note an important methodological caveat: the hierarchy that placed SAS above SDS was established post hoc, after data collection, rather than being prospectively prespecified. This kind of transparency is increasingly expected in behavioral intervention research, where outcome choices made after the fact can inflate the apparent strength of findings. Here, the consistency of results across both scales offers some reassurance, but the post hoc designation tempers how decisively the hierarchy can be interpreted.

The statistical picture was striking. In the completer analysis, the researchers found significant group-by-time interactions for both SAS and SDS, each with p values below 0.001, indicating that the trajectory of symptom change differed across the three groups. Within-group analyses showed significant post-intervention reductions in both SAS and SDS scores across all three groups, including the waitlist, a reminder that simply being observed and assessed can coincide with symptom improvement. Crucially, both Group A and Group B showed greater symptom reductions and lower adjusted post-test scores than the waitlist control, with all Holm-adjusted p values below 0.001. The Holm procedure is a sequential correction for multiple comparisons, designed to keep the overall risk of false positives in check when many statistical tests are run at once, which strengthens the credibility of the between-group contrasts.

Effect sizes told an equally compelling story. Large within-group pre-post effect sizes were observed in the two active groups, with partial eta-squared values ranging from 0.36 to 0.40. In behavioral science, partial eta-squared values above 0.14 are conventionally considered large, so figures in this range describe changes that are not merely statistically detectable but substantial in magnitude. However, the comparison between the two active arms came back null: no statistically significant difference was detected between the nonverbal and narrated versions for either the anxiety or the depression scale, with both Holm-adjusted p values at 0.434. The authors are careful to state that this null result does not establish equivalence between the two formats. Demonstrating equivalence requires dedicated equivalence testing with pre-specified margins, which this study did not perform. What the result does show is that the wordless version performed at least as well as the narrated one in this sample.

Why would sound-led, nonverbal content help at all? The researchers frame their approach around the hypothesis that nonverbal audiovisual material may impose a lower cognitive load than text-heavy interventions. Reading demands active decoding, working memory, and language proficiency, all of which compete for the same mental resources that are already strained in anxious or depressed states. By shifting the informational burden to ambient natural sound and imagery, the program may allow users to settle into the experience without the friction of comprehension. The team also points to potential cross-cultural generalizability as a motivation: a forest soundscape or the rhythm of ocean waves requires no translation, whereas guided text must be localized for every language and cultural context it reaches. The study’s abbreviation list hints at the acoustic toolkit involved, referencing binaural beats and head-related transfer functions, technologies that shape how spatialized audio is perceived by listeners through headphones.

The institutional collaboration behind the work is itself notable. The author team spans the Department of Music Engineering at the Shanghai Conservatory of Music, psychological education and counseling centers at the University of Shanghai for Science and Technology and the Shanghai Technical Institute of Electronics and Information, the medical psychology teaching and research department of Shanghai Jiao Tong University School of Medicine, and a cultural communication company. This blend of audio engineering expertise and clinical psychology practice reflects a growing recognition that effective digital mental health tools are as much a design and engineering challenge as a clinical one. The work was supported by the Shanghai Education Science Research Project, and the protocol was approved by the relevant institutional bodies with all participants providing informed consent in accordance with the Declaration of Helsinki.

Still, the study’s limitations deserve as much attention as its results. The trial was not prospectively registered, the allocation was systematic rather than random, and nearly half of the participants in the active arms did not complete the post-intervention assessment, with reasons unrecorded. Because the analysis was restricted to completers, the reported effects apply only to students who finished the program, and it is impossible to know from these data whether dropouts differed systematically in ways that inflated the apparent benefit. A waitlist control, moreover, cannot separate the specific effects of the intervention from expectation, attention, or the passage of time. The authors themselves conclude that further confirmatory studies are needed to establish the intervention’s clinical value and to evaluate its proposed cognitive load advantages and cross-cultural generalizability, rather than claiming therapeutic efficacy outright.

Even with those caveats, the study opens an intriguing frontier. If a wordless, 15-minute daily encounter with simulated nature, delivered through an ordinary smartphone, can coincide with large symptom reductions in students who complete it, then the design space for digital mental health support may be far broader than the text-based paradigm that currently dominates the field. The subtraction design used here offers a template for other teams: build the sensory core of an intervention, then systematically add or remove verbal scaffolding to learn what actually does the work. For the millions of students worldwide who report anxiety and depressive symptoms, and for the counselors and technologists trying to reach them, the message of this research is quietly radical. Sometimes the most effective thing an app can say is nothing at all, and simply let the ocean speak.

Subject of Research: A nonverbal, audio-dominant audiovisual intervention for reducing anxiety and depressive symptoms in college students

Article Title: Effects of an audio-dominant nonverbal audiovisual intervention on anxiety and depressive symptoms in college students: a three-arm systematically allocated controlled intervention study

Article References: Qin, Y., Chen, F., Luo, C., Chen, F., Zhang, H., Zhang, Z., Yang, J., Xie, K., Liang, X., & Wu, Q. (2026). Effects of an audio-dominant nonverbal audiovisual intervention on anxiety and depressive symptoms in college students: a three-arm systematically allocated controlled intervention study. BMC Public Health. https://doi.org/10.1186/s12889-026-29647-5

Image Credits: AI Generated

DOI: 10.1186/s12889-026-29647-5

Keywords: digital mental health, college students, anxiety, depression, audiovisual intervention, nonverbal, nature sounds, Self-Rating Anxiety Scale, Self-Rating Depression Scale, WeChat mini-program, controlled trial, cognitive load

Working memory, the mental scratchpad that lets us hold and manipulate a handful of items over seconds, is famously limited. Most people can maintain only about three or four items at once, and cognitive scientists have long argued that the brain must therefore impose structure on information to make the most of that narrow bandwidth. How exactly it does so has been difficult to observe, because internal organization leaves few direct traces in behavior. Huang and Doeller suspected that one such trace might be hiding in plain sight: in the eyes.

Eye movements provide a natural mechanism for sequential sampling of the world. When we look at a scene, we do not take it in all at once; we visit locations one after another, stitching together a representation piece by piece. The researchers reasoned that if the brain strategically organizes remembered content, that organization might be expressed in the order in which people look at items they need to remember. To test this, they designed a working memory task with a hidden architecture built into it.

The key innovation was the use of color-frequency pairings that created a latent ordinal structure. Participants saw items whose colors and spatial frequencies followed an underlying sequence, meaning that the memory load could in principle be reduced if a person noticed and exploited the pattern. Rather than storing every item independently, an observant participant could use the sequence as a scaffold, predicting where to direct attention next. Crucially, the experimenters never told participants about this structure. The question was whether their eyes would find it on their own.

Across two experiments, the answer was yes. Analysis of gaze patterns revealed that participants spontaneously sampled the items in an order that followed the latent sequence, both when the items were first presented and later, during the maintenance interval when the display was blank and participants had to hold the information in mind. During encoding, gaze followed the embedded structure as participants took in the material. During maintenance, they revisited the remembered locations in structured patterns, as if running through the sequence again from memory. The organizational scheme, in other words, was not just an artifact of how the stimuli were displayed; it was an internally generated strategy that expressed itself overtly in the eyes.

The most revealing results came when the researchers manipulated cognitive demand. In a version of the task with three items, a load that most people can handle, participants who performed well showed robust sequential sampling during the encoding phase itself. Their eyes organized the information as it came in, essentially filing it into the optimal structure at the moment of exposure. Lower performers, however, told a different story. Rather than organizing during encoding, they compensated later, showing replay-like revisitation during the maintenance period, as if they needed a second pass to impose the structure they had failed to build the first time around.

When the load increased to four items, pushing the limits of working memory, the pattern changed again in an instructive way. The encoding-based organization that high performers had shown at lower loads broke down. Under this heavier demand, structured revisitation emerged primarily during maintenance instead. The interpretation offered by the authors is that when encoding resources are strained, the brain cannot afford to organize information on the fly, so it shifts the organizational work to the delay period, using sequential eye movements to reconstruct and restructure the memory after the fact.

This flexibility is what makes the study more than a curiosity. The eye movements, the authors argue, provide an overt expression of an internally generated organizational process, one that adapts dynamically to cognitive demands rather than running as a fixed routine. That adaptiveness also connects the findings to a broader body of neuroscience. Sequential replay, in which the brain reactivates memory representations in a structured order during quiet periods, has been observed in neural firing patterns in animal studies and in human imaging work. The revisitation patterns seen here during maintenance look, the researchers suggest, like a behavioral analogue of that neural replay, visible at the level of the eyes when encoding resources run short.

The practical implications are potentially wide ranging. Because gaze sequences can be recorded noninvasively and with high temporal precision, they could serve as a real-time readout of how a person is organizing remembered content, and possibly of how efficiently they are doing it. Individual differences in when and how structure emerges, during encoding versus during maintenance, tracked performance differences in this study, hinting that eye-tracking-based measures might one day help assess working memory health or detect early changes in cognitive function.

For now, the study’s central lesson is conceptual: a process long considered hidden, the strategic organization of limited memory resources, leaves a measurable signature in one of the body’s most observable behaviors. The eyes, it appears, do not merely follow attention; they enact the architecture of thought. As Huang and Doeller’s work shows, when the brain needs to squeeze more out of less, it rehearses, restructures, and replays, and it does so in gaze sequences that any camera can capture. Working memory’s secret filing system, it turns out, has been staring back at us all along.

Subject of Research: Capacity-dependent organization of working memory expressed through eye movements

Article Title: Eye movements reflect capacity-dependent organization of working memory

Article References: Huang, Q., & Doeller, C. F. (2026). Eye movements reflect capacity-dependent organization of working memory. PLOS Biology, 24(10), e3004047. https://doi.org/10.1371/journal.pbio.3004047

Image Credits: AI Generated

DOI: 10.1371/journal.pbio.3004047

Keywords: working memory, eye movements, cognitive load, sequential sampling, memory organization, neural replay, PLOS Biology, gaze patterns, encoding, maintenance, attention, cognitive neuroscience

Inner speech, the act of speaking silently in one’s mind, is a cornerstone of human cognition. We use it to plan our day, rehearse conversations, regulate our emotions, and wrestle with difficult decisions. For decades, neuroimaging studies have shown that inner speech draws on a fronto-temporal speech-production network that partly overlaps with, but is not identical to, the network engaged during overt speech, including regions such as the left inferior frontal cortex, the supplementary motor area, the insula, and superior temporal areas. Depending on the region and task examined, inner speech can produce weaker, comparable, or even greater cortical activation than spoken speech. Yet nearly all of this work has focused on the cortex itself, treating speech-production networks as if they operated in isolation from the systems that regulate their functional state.

The Japanese team, led by Ryuta Kuwamizu, took a different approach. Rather than asking which cortical regions light up during inner versus overt speech, they asked whether those regions operate in the same neuromodulatory state during the two forms of speech. To answer this, they turned to pupillometry, the continuous, non-invasive measurement of pupil diameter. The pupil is a remarkably faithful proxy for brain arousal: it tracks the coordinated activity of the locus coeruleus-noradrenergic system, the cholinergic basal forebrain, the hypothalamic orexin system, and the large-scale cortical network states these systems govern. Crucially, unlike EEG, fMRI, or facial electromyography, pupillometry remains robust during overt articulation and requires no head-mounted apparatus, making it uniquely suited for directly comparing silent and spoken speech within a single paradigm.

The researchers designed an elegant counting task. Across three experiments, a total of 80 healthy, native Japanese-speaking participants sat before a screen-based eye tracker and completed blocks of three conditions: counting aloud, counting silently in their heads, and an instructed no-thought control in which they attempted to think of nothing at all. Each block consisted of a 30-second rest period followed by a 45-second task period, with pupil diameter recorded continuously at 60 Hz. In the first experiment, participants simply counted upward from one. In the second, they counted using grammatically structured numeral-noun phrases in Japanese, such as one cat, two cats, adding phonological and morphological complexity. In the third, they performed step counting in increments of two, three, or four, sharply increasing demands on working memory and cognitive control.

The results upended the team’s initial hypothesis. They had expected inner speech to produce pupil dilation comparable to overt speech, reflecting the cognitive effort of covert verbal production. Instead, in the simple counting experiment, inner speech produced significant pupil constriction relative to baseline, while overt speech produced significant dilation. Inner speech even drove the pupil smaller than the no-thought condition did. In the second experiment, the constriction vanished: inner speech no longer differed from baseline, while overt speech again produced robust dilation. Only in the third experiment, under the highest cognitive load, did inner speech finally produce clear pupil dilation, converging toward, but never reaching, the level seen during spoken counting.

The most robust finding across all three experiments was the consistent dissociation between the two speech modes. Overt speech reliably elicited pupil dilation in every condition, whereas inner speech was always accompanied by lower pupil-linked arousal, even as its direction relative to baseline shifted from constriction to no change to dilation as task demands grew. The researchers verified that participants genuinely engaged in inner counting by comparing final counts across conditions: counting speed correlated strongly between overt and inner speech in every experiment, with Spearman correlations ranging from 0.85 to 0.92. In the noun-phrase experiment, participants actually counted to higher numbers during inner speech, consistent with evidence that silent speech can proceed faster than spoken speech.

The magnitude of the overt-speech effect is itself remarkable. In the step-counting experiment, the sustained pupil dilation reached approximately 0.4 millimeters for more than 10 seconds, a response comparable in size to those reported during high-intensity cycling at roughly 80 percent of maximal oxygen uptake. In other words, speaking aloud under cognitive load drives pupil-linked arousal on a scale similar to vigorous physical exertion. This likely reflects the integrated demands of motor execution, respiratory variation, sensorimotor feedback, and attention that vocalization entails, each of which engages brainstem arousal systems and cortical network states.

The authors propose a state-dependent account of inner speech. On this view, inner and overt speech share much of the same linguistic machinery, but they run that machinery in different neuromodulatory states. Overt speech requires transforming an internal linguistic representation into an externally executable sensorimotor format, complete with articulatory execution, respiratory coordination, and auditory feedback monitoring against an efference copy of the motor command. These demands may recruit stronger noradrenergic gain modulation of speech-related auditory-motor networks, whereas inner speech can often remain in a compressed representational format and operate at lower gain. Notably, inner speech arousal scaled with cognitive-control demand rather than linguistic complexity per se: enriching the verbal content alone did not raise arousal, but forcing participants to update and maintain running sums did. This may explain why neuroimaging studies have reported such mixed cortical differences between covert and overt speech; the shared networks may simply be operating at different arousal states across studies and tasks.

The implications stretch well beyond the laboratory. Because pupillometry is contactless and silent, it could be used to study how inner speech becomes internalized over childhood development, and in populations such as young children or certain clinical groups for whom overt reporting or sensor-based recording is difficult. The capacity of inner speech to sustain verbal processing at reduced arousal may be advantageous where mental calm and efficient cognition must coexist, as in cognitive control, sport performance, or the routinized inner counting many people use as a sleep aid. Yet the same low-arousal efficiency might also enable maladaptive cognitive loops that contribute to rumination. Conversely, the elevated arousal accompanying overt speech may carry functional value; prior work has found that pupil dilation during reading aloud better tracks the memory advantage for spoken words than for silent ones.

The researchers are careful to note the limits of their work. The study operationalized inner speech as silent numerical counting, an externally instructed, repetitive task that cannot capture the full diversity of naturally occurring inner speech, from dialogic self-conversation to spontaneous verbal thought. The no-thought condition was an instructed control rather than verified mental silence, and pupillometry provides only an indirect index of arousal, not a direct readout of inner speech itself. The sex ratios also differed across the three experiments, warranting caution in cross-experiment comparisons. Still, the authors frame the work as a focused first step toward a neuromodulatory science of verbal thought. If the pattern holds in more naturalistic settings, the little voice in our heads may turn out to be defined not just by what it says, but by the quiet, energy-efficient state in which it says it, a whisper that runs the brain’s arousal machinery on a fraction of the power demanded by the spoken word.

Subject of Research: Pupil-linked arousal differences between inner speech and overt speech

Article Title: Dissociable pupil-linked arousal during overt and inner speech

Article References: Kuwamizu, R., Watanabe, R., Yamamoto, N., Otani, K., & Moriguchi, Y. (2026). Dissociable pupil-linked arousal during overt and inner speech. iScience, 29(11), Article 117825. https://doi.org/10.1016/j.isci.2026.117825

Image Credits: AI Generated

DOI: 10.1016/j.isci.2026.117825

Keywords: inner speech, overt speech, pupillometry, arousal, locus coeruleus, noradrenergic system, cognitive load, counting task, neuromodulation, working memory, self-talk, brain state