The research took place at King Abdulaziz University, where most participants were undergraduates and 80% were women. Participants were recruited through snowball sampling, and the investigators used self-administered standardized instruments to assess AI dependency, AI-related general anxiety, digital stress and quality of life. The analysis considered a wider set of cognitive, motivational and contextual factors: cognitive offloading, availability pressure, fear of missing out, digital overload, digital vigilance and social acceptance anxiety.

Network analysis was used to examine the interrelations among these measures and identify central variables and conditional interactions within the system. AI dependency emerged as a central node, showing strong conditional associations with digital stressors such as digital vigilance and fear of missing out. Anxiety also occupied a bridging position, linking cognitive reliance to environmental pressures. The study reports that these patterns of association were related to reduced quality of life.

Social acceptance anxiety had a distinct role in the model, translating cognitive pressures into relational and identity concerns. The analysis also identified non-linear, conditional associations, indicating that the psychological structure of AI reliance involved cognitive, motivational and contextual factors rather than direct usage intensity alone. These are statistical relationships within the network; they do not establish that AI dependency caused anxiety, digital stress or lower quality of life.

The authors characterized AI dependency as a complex cognitive–motivational construct that extends beyond the amount of AI use. They noted that reliance can have potential advantages, including reduced cognitive load and increased efficiency, but can also be associated with diminished autonomy, heightened anxiety and long-term digital strain. The results provide a culturally contextualized model for understanding how AI dependency may relate to student well-being, although the evidence comes from one university and should not automatically be extended to other institutions, populations or cultural settings. The use of self-reported measures, snowball sampling and a sample that was predominantly female and undergraduate also limits the conclusions that can be drawn from the study.

Subject of Research: Network analysis of AI dependency, anxiety, digital stress and quality of life among university students

Article Title: AI dependency and its relationships with anxiety, quality of life, and digital stress

Article References: Elsayed, F. K., Al-Mutairi, J. F., & Moussa, M. A. (2026). AI dependency and its relationships with anxiety, quality of life, and digital stress. BMC Psychology. https://doi.org/10.1186/s40359-026-05496-0

Image Credits: AI Generated

DOI: 10.1186/s40359-026-05496-0

Keywords: AI dependency, artificial intelligence, student anxiety, digital stress, quality of life, network analysis, cognitive offloading, fear of missing out, university students, digital vigilance, dependency, relationships

The experimental arrangement was a simplified version of the “director task,” a well-known paradigm in which one participant knows which object their partner should move, even when the partner can see several matching objects. A shelf with 16 compartments stood between the two people, and some compartments were hidden from the director’s view. Six objects were placed on the shelf: two were unique, while the other four formed two pairs of identical items, including candles, flowers, vases and cups. When the director requested one member of a pair, the partner could identify it by noticing whether that object was hidden from the director. All object names were feminine, bisyllabic German nouns, preventing grammatical gender from providing an unintended clue to the intended referent.

The researchers analysed 26 native German speakers; seven further participants were excluded because of technical problems, including disconnected eye-tracking cables, human error and loss of wireless connectivity. The analysed participants were 14 women and 12 men, with a mean age of 24.94 years. Each person acted as the listener, or “matcher,” while a trained confederate played the director. Participants wore a 128-channel mobile EEG system and Pupil Core eye-tracking glasses, which recorded pupil position at a nominal 120 hertz alongside a 30-hertz view of the scene. Four fiducial markers around the shelf allowed the software to track all 16 compartments even as the listener moved their head.

Solving the problem in the physical world required an unusual analytical workflow. EEG samples brain voltage at hundreds of times per second, eye trackers sample gaze at a different rate, and audio equipment operates at yet another timescale. The team used the open-source LabStreamingLayer protocol to timestamp and synchronize the three streams, with the recording computer serving as the master clock. The approach allows researchers to ask not merely whether a neural response followed a word, but whether the same response was stronger when the listener happened to be looking at the relevant object at that moment.

Natural interaction also creates movement artifacts, overlapping brain responses and a lack of strict experimental timing. To clean the EEG data, the team combined artifact subspace reconstruction, which detects transient high-amplitude disturbances and reconstructs brain activity from cleaner portions of a recording, with independent component analysis, which separates statistically independent neural and non-neural sources. Their central statistical innovation was regression-based event-related potentials, or rERPs. Rather than averaging short epochs and assuming that responses do not overlap, this method models continuous EEG as the combined result of responses to spoken words, fixations and other events. Linear deconvolution then estimates the response attributable to each event while accounting for the others.

The results overturned one of the study team’s predictions. Because a hidden competitor reveals information about what the director can see, the researchers expected listeners to inspect ambiguous objects. Instead, the competitors received significantly less visual attention than other non-target objects. Between 1.4 and 3.4 seconds after the critical noun, listeners looked at the hidden competitor less often than at the other objects on the shelf. By contrast, they looked more often at the target when it had a hidden competitor than when it was a unique object, with a significant difference between 1.1 and 1.9 seconds after the noun. The target was both the object to move and the one that disclosed whether the instruction was ambiguous.

This strategy appears to be an example of cognitive offloading. Rather than retaining a detailed mental map of the entire shelf, listeners used the visual scene to obtain the one piece of information they needed. Hidden competitors remained in place throughout a 10-to-15-minute block, whereas targets changed position repeatedly, so little could be gained by looking at the competitors after their locations had been learned. Fixation-related brain potentials supported this interpretation. When a target was viewed during or after its name was spoken, the visual and linguistic information could be checked against one another, producing stronger and earlier neural effects than when a target was found before its name.

Responses time-locked to the spoken nouns also differed between the paired and unique conditions. In the N400/P300 interval, conventionally extending from 300 to 500 milliseconds after word onset, paired targets produced a pronounced positivity over posterior and occipital electrodes, peaking at about 500 milliseconds. The polarity reversed over frontal and prefrontal regions, and a later positivity appeared between roughly 600 and 850 milliseconds. The authors interpret the posterior response as a P300-like signal of attentional reorientation: hearing “candle” required listeners to detect that the referent was not unique, redirect attention and initiate a visual search. A unique target behaved more like an expected item and required less reorientation.

Fixation timing carried much of the interpretive information. In the paired condition, later target fixations were associated with a stronger posterior positivity, especially from 300 to 500 milliseconds after the noun. If listeners had already found the target before hearing its name, they could prepare for it; if they found it during the noun, visual and linguistic evidence arrived together; if they searched only after the word ended, target detection was delayed and required more attentional effort. The study’s resampling analysis showed that the competitor effect in gaze was highly robust, while the key interaction between fixation timing and condition in the fixation-related potentials stabilized at a median sample size of 16. The language-related effects were less secure: the 400-to-500-millisecond N400 effect was the most stable component, whereas several other time windows and higher-order interactions did not stabilize across reduced samples.

The authors are careful not to present the study as a definitive account of language-specific brain mechanisms. Its strongest contribution is methodological: it demonstrates that synchronized mobile EEG and eye tracking can capture language processing during a real, face-to-face exchange, even while participants move and produce movement-related noise. The workflow also preserves the complementarity of the two signals. Gaze reveals what a participant is inspecting and when, while EEG reveals how the brain responds to a spoken word or a fixation. Used together, they can distinguish whether a neural difference reflects the presence of ambiguity, the timing of visual search, or simply the number of looks at an object. All analysis scripts and two example datasets are openly available, while protected participant data may be shared on request.

Future versions could extend the approach in several directions. Recording the director’s gaze, facial expression and prosody would allow both sides of the exchange to be analysed, potentially revealing joint attention as speech is planned and interpreted. More varied utterances would also test whether the workflow survives less constrained language than the short noun phrases used here. The same regression and overlap-correction framework can also be adapted from voltage time courses to continuous EEG-derived measures such as band-limited power, opening the door to oscillatory analyses of communication. For now, the results show that even a simple instruction such as moving a candle can recruit a dynamic loop of prediction, visual search and neural reorientation—and that measuring the eyes alongside the brain can reveal that loop in remarkable detail.

Subject of Research: How eye movements and brain activity jointly support referential selection during spoken interaction

Article Title: Combined eye tracking and electroencephalography during referential selection in dyadic interaction

Article References: Brilmayer, I., Georgis, P., & Schumacher, P. B. (2026). Combined eye tracking and electroencephalography during referential selection in dyadic interaction. Behavior Research Methods, 58(11), Article 326. https://doi.org/10.3758/s13428-026-03172-0

Image Credits: AI Generated

DOI: 10.3758/s13428-026-03172-0

Keywords: mobile EEG, eye tracking, referential selection, language processing, N400, P300, dyadic interaction, naturalistic neuroscience, regression ERP, cognitive offloading, Combined, tracking

The study, published on 27 August 2026, is explicitly conceptual rather than empirical. The authors do not claim to have validated a new psychometric construct, nor do they assert that plausibility effects, fluency, authority bias, automation bias, or epistemic dependence are new phenomena invented by AI. Instead, they define what they call epistemic plausibility bias more narrowly, as a recurring validation-level distortion in which the perceived quality of a knowledge-like output is allowed to stand in for an adequately calibrated assessment of the warrant supporting it. In plain terms, a text that reads well, cites convincingly, and sounds authoritative can pass through the checkpoints that universities use to certify knowledge, even when nothing underneath has been verified.

The theoretical machinery behind this claim draws on decades of cognitive science. Research on processing fluency has shown that statements that are easier to read are judged more likely to be true, a metacognitive shortcut that Alter and Oppenheimer described as uniting the tribes of fluency into a metacognitive nation. Work by Connell and Keane on plausibility judgments showed that conceptual coherence and distributional word coherence both shape how plausible a statement feels, independently of whether it is correct. Large language models are, in effect, industrial-scale producers of both kinds of coherence: they are trained to generate text whose internal statistics match the distribution of human writing, which is precisely the property that makes them feel plausible. The authors argue that this is where the danger lies, because the very features that make AI output convincing are orthogonal to whether it is warranted.

What distinguishes the new concept from existing ones, the authors insist, is its focus on the epistemic transition from plausibility cues to attributed warrant, and, when outputs enter academic procedures, to institutional validation. Existing concepts such as confirmation bias, automation bias, and cognitive offloading explain important antecedents of the distortion, but they do not capture the specific moment at which a plausible-sounding output acquires the social status of validated knowledge. Automation bias, documented extensively in human factors research by Lee and See and by Parasuraman and Manzey, describes the tendency to over-rely on automated systems; epistemic dependence, analyzed by Hardwig and by Goldberg, describes the way knowers must rely on others they cannot fully verify. The new analysis connects these threads to show how a plausibility cue generated by a machine can be laundered into institutional warrant when a student submits it, a researcher cites it, or a reviewer endorses it.

A crucial part of the argument concerns what AI actually is in this framework. The authors treat AI as a cognitive condition only as shorthand for a socio-technical condition of academic cognition: AI reorganizes the environments in which academic search, synthesis, writing, assessment, verification, and judgment are delegated, accelerated, and institutionally normalized. They explicitly reject the idea that this expression competes with established frameworks such as technological mediation, distributed cognition, extended cognition, or cognitive offloading. The point is not that minds have literally expanded into machines, in the spirit of Clark and Chalmers’s extended mind thesis, but that the institutional scaffolding around cognition has changed so thoroughly that the default path from question to answer now runs through systems optimized for plausibility rather than truth.

The implications for teaching are direct. If students can produce fluent, well-structured, citation-rich text on demand, then the traditional proxy signals of learning, such as the polished essay, lose their diagnostic power. The authors argue that higher education should respond not through simple prohibition or uncritical adoption, but through critical AI literacy, process-oriented assessment, preservation of critical time, and institutional governance of epistemic delegation. Process-oriented assessment means evaluating how students arrive at conclusions rather than only the conclusions themselves, making the warrant visible. Preservation of critical time means deliberately protecting the slow, effortful verification work that fluency tempts us to skip, since the economics of generative tools make fast output nearly free while careful checking remains expensive.

Research writing and peer review face a parallel version of the same distortion. The authors, who have previously published on the ethical and methodological challenges that AI poses to the peer review process, note that reviewers working under time pressure are exactly the population most vulnerable to automation bias and fluency effects. A manuscript polished by a language model can read better than a correct one, and a reviewer who delegates screening to AI tools inherits the model’s plausibility judgments without its accountability. The study also engages with the broader literature on critical generative AI literacy, including work by Rapanta and colleagues on postdigital configurations and by Roe and colleagues on metaphorical frameworks for AI literacy, positioning critical literacy not as a technical skill but as an epistemic capacity: the ability to ask what warrant stands behind a fluent sentence.

The analysis extends beyond the seminar room to cognitive inequality and the public role of universities. Because the ability to detect unwarranted plausibility depends on domain expertise, students and institutions with fewer resources may be least equipped to catch the distortions, deepening existing gaps in epistemic access. At the societal level, the authors argue that universities hold a distinctive responsibility: they are society’s certified producers of warrant, and if their validation procedures can be satisfied by plausibility alone, the public epistemic commons degrades. This connects the study to philosophical work on epistemic authority by Zagzebski and on trust in experts by Goldman, and to Floridi’s characterization of large language models as agency without intelligence, systems that can act competently in language without understanding it.

The authors are careful about the boundaries of their claim. They note that the study does not generate or analyze datasets, and that the expression epistemic plausibility bias is offered as a conceptual tool for locating a specific failure point in the knowledge pipeline, not as a rival to existing constructs. The contribution is diagnostic and normative rather than psychometric: it names the transition from cue to warrant to institution, and it assigns responsibility for policing that transition. In their framing, epistemic responsibility in the age of generative AI is not primarily about detecting fabricated citations or catching plagiarism, tasks that technical tools may increasingly handle, but about redesigning the procedures by which academic communities decide that something counts as known.

What emerges is a picture of higher education at an epistemic crossroads. The technology that makes knowledge-like text abundant also makes the verification of warrant scarce, and the institutions built to certify knowledge must now decide whether their seals of approval attach to fluency or to justification. The authors’ answer is neither panic nor embrace but governance: curricula that teach the difference between plausibility and warrant, assessments that reward visible reasoning, schedules that protect the time verification requires, and institutional rules that make epistemic delegation explicit and accountable. Whether universities can execute that agenda before plausibility becomes the default currency of academic knowledge is, on the authors’ own analysis, the defining question of the coming decade.

Subject of Research: How AI-generated plausibility becomes mistaken for knowledge in higher education and what epistemic responsibility requires in response

Article Title: AI-generated plausibility and epistemic responsibility in higher education: implications for knowledge, learning, and society

Article References: Maturo, F., Micozzi, A., & Porreca, A. (2026). AI-generated plausibility and epistemic responsibility in higher education: implications for knowledge, learning, and society. SN Social Sciences, 6(9), Article 386. https://doi.org/10.1007/s43545-026-01683-3

Image Credits: AI Generated

DOI: 10.1007/s43545-026-01683-3

Keywords: artificial intelligence, higher education, epistemic plausibility bias, epistemic responsibility, critical AI literacy, peer review, automation bias, cognitive offloading, assessment, large language models, academic integrity, knowledge validation

The research, led by Dr. Georgia M. Vasilakis of the UPMC Heart & Vascular Institute alongside Dr. Kaitlyn E. McSurdy, Dr. Kathryn Berlacher, and Dr. Natalie Stokes, is a cross-sectional survey study — a snapshot in time of attitudes and behaviors across a single academic medical community. Cross-sectional designs like this one capture prevalence and perception rather than causation, but they are precisely the instrument needed when a technology has spread faster than the literature describing it. Graduate medical education, the period when medical school graduates train as residents and fellows under faculty supervision, is an especially revealing setting, because it pairs the least experienced clinicians with the most experienced ones and asks how both groups respond to the same unfamiliar technology.

The urgency of the question is hard to overstate. Generative AI systems — large language models that can draft prose, summarize evidence, answer clinical questions, and simulate patient encounters — have penetrated professional life at a pace with few precedents in medicine. Surveys of physicians conducted by the American Medical Association in 2026, cited in the study’s bibliography, document rapidly shifting sentiment toward augmented intelligence in clinical practice. Yet formal guidance has lagged well behind adoption. A 2025 report from the Association of American Medical Colleges found that artificial intelligence curricula remain unevenly implemented across U.S. and Canadian medical schools, meaning many trainees arrive in residency having never received structured instruction in what these tools can do, when they fail, or how their outputs should be verified.

That gap between use and instruction is the study’s central concern. The title’s nautical metaphor — sailing by the stars — is deliberate. Generative AI offers trainees a navigational aid: a quick answer on a busy call night, a draft of a difficult patient message, a summary of a dense guideline document. But stars can mislead. Large language models generate text by predicting plausible word sequences, not by verifying truth, and they can produce confident, fluent, entirely fabricated content — the phenomenon known as hallucination. A resident who consults an AI system at three in the morning may be consulting something that sounds authoritative and is wrong in ways that are difficult to detect without exactly the expertise the trainee is still developing.

The study’s authors situate their work within a growing but still thin body of evidence. Previous research by Cervantes and colleagues, published in the Journal of Investigative Medicine in 2024, examined medical educators’ perceptions of generative AI and found a profession divided — intrigued by the technology’s potential, wary of its risks to rigor and to the development of clinical reasoning. A 2025 review in the New England Journal of Medicine by Abdulnour, Gin, and Boscardin went further, proposing educational strategies for the clinical supervision of AI use itself, arguing that supervising how trainees engage with AI should become as routine as supervising how they prescribe medications or interpret electrocardiograms. The Pittsburgh survey adds empirical ground truth to that theoretical framework by asking what is actually happening in one large academic center.

The stakes of the question extend beyond individual competence into the architecture of medical training itself. Residency is built on graduated responsibility: interns perform supervised tasks, seniors direct teams, attendings hold ultimate accountability. Generative AI scrambles this hierarchy in an unfamiliar way, because it gives the most junior member of the team access to something resembling expert-level output without any guarantee of expert-level accuracy. A 2023 randomized vignette study published in JAMA by Jabbour and colleagues measured this directly, testing whether AI assistance improved diagnostic reasoning for hospitalized patients — and found that the benefits were neither automatic nor uniformly distributed, underscoring that how the tool is deployed matters as much as whether it is deployed at all.

There is also a subtler cognitive risk that the study’s framework acknowledges: the phenomenon of cognitive offloading. Research published in the journal Societies in 2025 by Gerlich examined how reliance on AI tools correlates with diminished engagement in critical thinking, a finding with obvious resonance for a profession whose core skill is precisely disciplined, skeptical reasoning. If a trainee learns to accept an AI-generated differential diagnosis without interrogating it, the technology may quietly erode the very habit of mind that residency exists to build. Conversely, if trainees are taught to treat AI output the way they treat a consultant’s recommendation — as an informed opinion to be weighed, checked, and integrated — the technology could sharpen rather than soften clinical judgment. Which outcome prevails depends heavily on the norms that faculty model and programs codify, which is exactly why measuring both groups’ behaviors matters.

The Pittsburgh team’s design reflects that dual focus. By surveying trainees and faculty within graduate medical education at the same institution, the researchers could capture the full supervision chain: the people most likely to reach for these tools and the people responsible for evaluating the work those tools touch. Portions of the work were presented as a poster at the University of Pittsburgh Graduate Medical Education Connect Conference in February 2026 and as a plenary session at the medical center’s Department of Research Day in May 2026, before the full paper was accepted on August 13, 2026 and published online on August 31, 2026. The authors report no funding and no competing interests, and the study involved no patients or protected health information, sparing it the privacy entanglements that often complicate clinical AI research. In a move that strengthens the study’s transparency, the datasets generated and analyzed are publicly available in the Harvard Dataverse repository, allowing other researchers to interrogate the findings independently.

What makes this study resonant beyond academic medicine is the broader economic and social evidence it aligns with. A widely discussed 2023 field experiment by Dell’Acqua, McFowland, Mollick, and colleagues described the performance of AI on knowledge work as a jagged technological frontier: tasks on one side of an invisible boundary are transformed by AI assistance, while superficially similar tasks just beyond it see little benefit or outright harm. Medicine sits squarely astride that frontier. Drafting administrative text and surfacing literature are tasks where generative models excel; nuanced diagnostic reasoning under uncertainty remains stubbornly human. The problem, as the Pittsburgh survey implicitly frames it, is that trainees and faculty may not agree on where the boundary lies — or may not know it exists at all.

The study’s ultimate contribution may be less a single statistic than a reframing of institutional responsibility. Sailing by the stars served generations of navigators well, but only because the stars’ positions were meticulously cataloged, their small motions understood, and their limitations respected. Generative AI in medical training deserves the same treatment: not prohibition, not blind adoption, but rigorous charting of where it helps, where it misleads, and what competencies trainees must retain regardless of how capable the tools become. The Pittsburgh team — clinicians embedded in the very training environment they studied — have supplied one of the careful navigational fixes the field now needs. As residency programs across the country confront trainees already carrying these tools in their pockets, the question the study poses is no longer whether graduate medical education will engage with generative AI, but whether it will do so with a map or by guesswork.

Subject of Research: Generative AI use among graduate medical education trainees and faculty physicians

Article Title: Sailing by the Stars: A Cross-Sectional Survey Study of Generative AI Use Among GME Trainees and Faculty

Article References: Vasilakis, G. M., McSurdy, K. E., Berlacher, K., & Stokes, N. (2026). Sailing by the Stars: A Cross-Sectional Survey Study of Generative AI Use Among GME Trainees and Faculty. Journal of General Internal Medicine. https://doi.org/10.1007/s11606-026-10751-6

Image Credits: AI Generated

DOI: 10.1007/s11606-026-10751-6

Keywords: generative artificial intelligence, graduate medical education, medical education, residency training, survey study, clinical supervision, physicians, large language models, cognitive offloading, academic medicine, Journal of General Internal Medicine, hallucination

The study, conducted by Bo Yuan of the School of Educational Studies at Universiti Sains Malaysia, followed 1,043 Chinese undergraduates across three waves of data collection, each separated by intervals of three to four weeks. Rather than treating AI dependency as a single monolithic construct, the research distinguished three dimensions: cognitive and behavioral dependence, which reflects how deeply students integrate AI into their thinking and routine actions; functional dependence, which captures the practical reliance on AI to complete tasks that students might otherwise manage themselves; and emotional dependence, which describes the affective attachment and comfort students derive from having the tool at hand. This dimensional approach matters because each component may develop through different psychological mechanisms and may therefore require different interventions.

The methodological centerpiece of the study is the cross-lagged panel network analysis, a technique that combines the longitudinal rigor of cross-lagged panel models with the granularity of network psychometrics. In a conventional cross-lagged design, researchers examine whether variable A at time one predicts variable B at time two after accounting for B’s own prior level, thereby offering a window into prospective directional relationships. The network extension treats each measured construct as a node in a web of interconnections, allowing researchers to see not just whether two variables influence each other but how entire constellations of psychological factors interact across time. This approach has gained traction in studies of psychopathology and, more recently, in educational psychology, where the interplay between motivation, emotion, and behavior resists simple linear explanation.

Yuan grounded the investigation in two complementary theoretical frameworks. The first is Pekrun’s control-value theory of achievement emotions, which holds that emotions arising in academic settings are shaped by students’ perceptions of control over outcomes and the subjective value they attach to those outcomes, and that these emotions in turn feed back into learning behavior and performance. The second is self-determination theory, which frames human motivation around the satisfaction of autonomy, competence, and relatedness needs. Together, these frameworks predict that the experience of using AI in writing is not emotionally neutral: each encounter with the tool generates affective responses such as enjoyment, hope, anxiety, or boredom, and those responses may either consolidate healthy engagement or nudge students toward problematic reliance.

The central finding is that AI writing experience was prospectively associated with all three dimensions of GenAI dependency, but only through dimension-specific achievement-emotion pathways. Enjoyment emerged as a key emotional conduit. Students whose prior AI writing experience predicted greater enjoyment subsequently showed higher emotional dependence, suggesting that the pleasure of effortless, AI-assisted composition can quietly deepen an affective attachment to the tool. Enjoyment, together with hope, also mediated the path to functional dependence, indicating that positive emotions about AI use may encourage students to hand over more of the practical writing workload to the machine. Hope, in this context, likely reflects students’ anticipation that AI will help them achieve better outcomes, a forward-looking emotion that makes delegation to the tool feel like a rational strategy rather than a crutch.

Anxiety told a different and more surprising story. The indirect association between AI writing experience and cognitive and behavioral dependence through anxiety was negative, meaning that students who experienced more anxiety in connection with AI writing were less likely to develop the deepest, most habit-like form of reliance on the tool. This finding complicates the intuitive assumption that negative emotions always fuel problematic use. Instead, anxiety may function as a psychological brake, signaling to students that something about their relationship with AI feels wrong and prompting them to pull back before dependence hardens into routine. All four of these supported indirect pathways produced bootstrap confidence intervals that excluded zero, a statistical standard indicating that the effects were unlikely to be artifacts of sampling variability.

Equally informative were the pathways that failed to reach that threshold. The remaining five achievement emotions examined in the study did not yield indirect effects whose bootstrap confidence intervals excluded zero, meaning the evidence for their mediating role was not statistically compelling. This pattern of specificity is scientifically valuable: it suggests that not all emotions matter equally in the formation of AI dependency, and that blanket assumptions about the emotional drivers of technology overuse may obscure the more precise mechanisms at work. Enjoyment and hope appear to open the door to dependence, while anxiety may help keep it closed, at least for the cognitive and behavioral dimension.

An important caveat accompanies these conclusions. The supported indirect associations were estimated at the between-person level, meaning they describe differences across individuals rather than fluctuations within a single person over time. In practical terms, a student whose AI writing experience is generally higher than their peers’ tends to show the predicted emotional and dependency patterns relative to those peers, but the study cannot confirm that an individual student’s day-to-day changes in AI use produce corresponding changes in their own emotions and dependence. This distinction, long emphasized in critiques of cross-lagged panel methodology, is crucial for interpreting the findings and for designing future research that can separate within-person dynamics from stable between-person differences.

The study’s implications reach beyond the statistics classroom. As generative AI becomes embedded in writing instruction worldwide, educators and policymakers face the challenge of distinguishing productive tool use from dependency that erodes independent skill. The dimensional findings suggest that interventions should be tailored: addressing emotional dependence may require fostering alternative sources of enjoyment in writing, while curbing functional dependence might involve building students’ confidence and hope in their own unassisted capabilities. Meanwhile, the protective role of anxiety hints that a degree of critical self-awareness about AI use, rather than unqualified enthusiasm, may serve students well. The research also connects to a broader scientific conversation about cognitive offloading, the well-documented tendency to delegate memory and reasoning tasks to external tools, raising the question of whether AI writing assistants represent a new frontier of that ancient cognitive strategy or a qualitatively different kind of dependence. As the technology evolves, longitudinal studies of this kind will be essential for understanding not just how much students rely on AI, but why.

Subject of Research: Longitudinal study of how AI writing experience shapes generative AI dependency among Chinese EFL student writers

Article Title: The Longitudinal Relationship between EFL Writers’ AI Writing Experience and GenAI Dependency: A Cross-Lagged Panel Network Analysis

Article References: Yuan, B. (2026). The Longitudinal Relationship between EFL Writers’ AI Writing Experience and GenAI Dependency: A Cross-Lagged Panel Network Analysis. International Journal of Mental Health and Addiction. https://doi.org/10.1007/s11469-026-01745-w

Image Credits: AI Generated

DOI: 10.1007/s11469-026-01745-w

Keywords: generative AI, AI dependency, EFL writing, achievement emotions, cross-lagged panel network, longitudinal study, self-determination theory, cognitive offloading, higher education, academic emotions, Longitudinal, Relationship

The study used a descriptive qualitative design, drawing on semi-structured interviews with students from a single western Chinese medical university. Each participant completed a brief paper-based background form and attended a one-on-one in-person interview, providing first-hand accounts of how they actually use tools such as large language models in their daily academic work. The researchers then applied inductive content analysis to the transcripts, allowing themes to emerge from the data rather than testing predetermined hypotheses. Two established frameworks—the Technology Acceptance Model, or TAM, and AI Literacy frameworks—informed the discussion of the findings, giving the team a theoretical vocabulary for describing why students accept, reject, or reflect on the technology.

What emerged was a clear split in usage patterns across academic levels. Undergraduates reported turning to generative AI primarily for coursework: drafting assignments, clarifying concepts, and completing routine academic tasks. Postgraduates, by contrast, described a far more sophisticated repertoire, deploying the technology for literature reviews, experimental design, and data analysis. This distinction matters because it suggests the technology is not simply amplifying existing academic effort—it is being woven into fundamentally different kinds of cognitive work at different stages of training. The tools that help a first-year student understand a concept are the same tools that help a doctoral candidate design an experiment, but the cognitive stakes and the potential for harm are entirely different.

Students across all levels perceived real benefits for critical thinking and knowledge acquisition. Interview accounts suggested that engaging with generative AI could push learners to articulate questions more precisely, evaluate the quality of AI-generated responses, and consolidate their understanding of complex material. For undergraduates, this often meant using the technology as a kind of Socratic partner—asking it to explain a concept, then probing the explanation for gaps. For postgraduates, the perceived benefits extended into research competence, with master’s and doctoral students reporting that generative AI helped them navigate vast literatures, refine experimental logic, and interpret datasets more efficiently than they could alone.

But the perceived gains came with a counterweight. Students described a cluster of risks that they believed could undermine the very competencies the technology was supposed to strengthen. Over-reliance emerged as a central concern: the fear that habitual use of generative AI would erode the capacity for independent reasoning rather than enhance it. Academic integrity was another recurring theme, with students wrestling with the boundary between legitimate assistance and misconduct. Perhaps most striking was the concept of cognitive offloading—the technical term for delegating mental work to an external system. Students recognized that when they offload tasks such as summarizing, synthesizing, or problem-solving to generative AI, they may be sacrificing the cognitive effort that produces durable learning.

The perceived benefits were not distributed evenly across the educational spectrum. Research competency gains, the study found, were largely confined to postgraduates. Undergraduates reported benefits primarily for knowledge acquisition and critical thinking, while master’s and doctoral students described advantages that extended into the technical machinery of research itself. This asymmetry suggests that generative AI functions differently depending on the cognitive maturity of the user: for early-stage learners, it is a comprehension aid; for advanced learners, it is a research instrument. The distinction has direct implications for how AI literacy education should be designed, because the skills needed to use the technology safely as a comprehension aid are not the same as those needed to use it rigorously as a research tool.

A pronounced demand for systemic support pervaded every academic level. Students across all stages called for teacher guidance and institutional training, indicating that they do not feel equipped to navigate generative AI on their own. This demand was not limited to technical instruction—students wanted help with the ethical, epistemic, and practical dimensions of AI use. The consistency of this finding across educational levels suggests that the need for structured support is not a transitional problem that disappears with experience, but a persistent feature of the technology’s integration into health professions education. Students at every stage recognized that the risks of unguided use—over-reliance, integrity violations, cognitive offloading—require deliberate institutional intervention rather than individual trial and error.

Based on the qualitative findings, the research team proposed an extended three-level TAM model as a preliminary conceptual framework for capturing students’ different generative AI use patterns. The model describes three distinct orientations: a perceived ease-of-use-driven orientation, a perceived usefulness-driven orientation, and a critical reflection orientation. The framework is designed to explain why students at different stages engage with the technology in fundamentally different ways—undergraduates may be drawn primarily by how easy the tools are to use, postgraduates by how useful they are for research, and advanced learners increasingly by their capacity for critical reflection on the technology’s outputs and limitations. The researchers emphasized that the model requires further validation in diverse educational contexts before it can be considered a robust theoretical tool.

The study’s implications extend beyond the classroom. As generative AI becomes embedded in clinical decision support, diagnostic reasoning, and patient communication, the competencies that medical and nursing students develop during training will shape how they use the technology in practice. The finding that perceived risks shift across educational levels suggests that AI literacy education may need to be adaptive rather than uniform—tailored to the specific cognitive work that learners are performing at each stage. For undergraduates, this might mean structured exercises that use generative AI to strengthen comprehension while guarding against over-reliance. For postgraduates, it might mean training in the rigorous evaluation of AI-assisted literature reviews, experimental designs, and data analyses. For doctoral students, it might mean developing the critical reflection needed to interrogate the technology’s outputs and limitations in the context of original research.

The researchers acknowledge that their study has constraints. It drew on students from a single western Chinese medical university, and the sample of 44 participants, while sufficient for qualitative analysis, cannot capture the full diversity of experiences across institutions, cultures, and educational systems. The proposed three-level TAM model remains preliminary and requires further validation in varied contexts. Nonetheless, the study provides empirical evidence for constructing an AI-literacy education system tailored to learners at different levels, and it highlights a question that educators worldwide are only beginning to confront: not whether generative AI belongs in health professions education, but how its integration should be calibrated to the distinct developmental needs of students at every stage of training. The answer, this study suggests, lies in recognizing that the technology’s perceived influences are not uniform—they are as differentiated as the learners themselves.

Subject of Research: Perceived influences and educational support needs of generative AI among medical and nursing students across educational levels

Article Title: Differences in perceived influences and educational support needs of generative AI among medical and nursing students across educational levels: a descriptive qualitative study

Article References: Zhou, G., Li, B., Wen, T., Wen, H., Sun, T., Zhang, H., Li, X., Wang, L., & Wang, Y. (2026). Differences in perceived influences and educational support needs of generative AI among medical and nursing students across educational levels: a descriptive qualitative study. BMC Medical Education. https://doi.org/10.1186/s12909-026-10637-5

Image Credits: AI Generated

DOI: 10.1186/s12909-026-10637-5

Keywords: generative AI, medical education, nursing education, AI literacy, qualitative research, critical thinking, academic integrity, cognitive offloading, Technology Acceptance Model, health professions education, undergraduate students, postgraduate students

The framework’s central insight rests on a distinction that cognitive load theory has refined over decades: not all mental effort is equal, and not all of it deserves to be automated. Intrinsic load refers to the inherent complexity of a task — the number of elements a writer must hold in mind simultaneously to make sense of an argument. Extraneous load, by contrast, is the friction that contributes nothing to understanding: hunting for a citation format, wrestling with a template, untangling a garbled bibliography. Germane processing is the deliberate allocation of working memory resources toward building schemas, the mental structures that make expertise possible. The paper’s argument is that AI tools are remarkably well suited to absorbing extraneous load and strikingly ill suited to absorbing germane processing, because germane processing is not a task that can be handed off at all. When a tool performs the work, the processing does not transfer to the writer; it simply does not occur.

This claim gains force from recent empirical findings on what happens when writers lean heavily on generative assistance. Studies documenting reduced neural engagement, weaker recall of one’s own writing, and a diminished sense of authorship among heavy AI users suggest that the costs of offloading are not merely pedagogical abstractions. They are measurable changes in how deeply a writer encodes their own ideas. Writing, as Janet Emig argued in 1977, is a distinct mode of learning rather than a record of thinking already completed. The act of struggling to render a messy idea in precise language is itself the mechanism by which the idea becomes clear to the writer. Remove the struggle, and the clarity never arrives.

The evidence on writing-to-learn adds an important nuance. A review of 48 school-based writing-to-learn programs by Bangert-Drowns and colleagues found a consistent positive effect on achievement, and a subsequent synthesis by Graham and colleagues reported comparable effects across 56 experiments in science, social studies, and mathematics. But what predicted a stronger effect is instructive for AI design: writing that included metacognitive prompts helped more, while assigning longer pieces helped less. The benefit of writing comes from the requirement that writers think about what they are doing, not from sheer output volume. This is precisely the condition that AI can silently remove. A student who prompts a chatbot to expand a paragraph has produced more words while doing less thinking — the exact inversion of what the writing-to-learn literature identifies as beneficial.

The nine-phase framework responds to this problem by refusing to treat research writing as a single undifferentiated activity. Instead, it maps the process from initial curiosity and inquiry through references and submission, and for each phase specifies three things: the dominant cognitive demand, the appropriate AI role ranging from minimal involvement to full collaboration, and design guidelines for how tools and interactions should behave. The logic is that the right AI role depends on the cognitive demand of the phase, not on the tool’s capabilities. In early phases, where the writer must generate questions and confront ambiguity, AI involvement should be minimal, because the struggle to formulate a researchable question is germane processing in its purest form. In later phases, where formatting conventions and citation management dominate, AI can operate at full collaboration, because those demands are largely extraneous and offloading them frees working memory for the thinking that matters.

This phase-aware approach is an application of responsive pedagogy to tool design. Responsive pedagogy holds that instruction should respond to the learner’s current understanding through recursive feedback dialog rather than proceeding on a fixed script. Transposed to AI-assisted writing, the principle becomes: the tool should respond to the cognitive demands of the task at hand, adapting its level of involvement as the writer moves through phases. A system that offers the same degree of assistance at every stage — drafting, organizing, synthesizing, polishing — is pedagogically blind, because it cannot distinguish between the friction that should be removed and the friction that is doing the teaching.

The framework’s scope is unusually broad. It addresses research writing from secondary school students in grades 9 through 12 developing their first researchable questions, through doctoral candidates building dissertations, to postdoctoral and professional researchers preparing manuscripts for peer review, across disciplines. This range matters because the temptation to offload struggle is not confined to novices. The paper opens with an anecdote about a high school student who submitted three AI-generated papers, frustrated by what the student perceived as a double standard — teachers use AI all the time, so why is it not acceptable for students? The anecdote illustrates how pervasive the temptation has become, and how the absence of clear norms leaves students improvising their own ethical boundaries. But the same dynamics apply to a seasoned researcher facing a deadline, for whom the pull toward delegation is arguably stronger because the procedural fluency to prompt effectively is already in place.

The paper’s own composition models its recommendations. Each section includes a human-AI collaboration disclaimer documenting exactly which cognitive work the researcher performed and which was delegated. The researcher conceived the paper’s focus, proposed the conceptual framework, and implemented it. Grammarly was used for editing, with sentence revision suggestions accepted or dismissed by the researcher. When Claude Pro was asked to review the initial single research question, it suggested splitting the question into two — one addressing which cognitive demands constitute extraneous load versus germane processing, the other addressing how human-in-the-loop collaboration should be designed across phases — and the researcher, after review, agreed. In building a definitions table, the researcher set the structure, identified terms, and selected sources; the AI drafted definitions grounded in those researcher-selected sources and flagged terms the researcher had overlooked. The researcher then verified each entry against the cited sources and reorganized the categories. This is the framework applied to itself: human curation and judgment retained, AI operating within boundaries the human establishes.

The transparency of these disclaimers points toward a broader implication for how AI-assisted scholarship should be evaluated. If the value of a paper lies partly in the germane processing its author performed, then disclosure of the division of cognitive labor becomes a form of methodological honesty, analogous to reporting data collection procedures. Readers can then judge for themselves whether the claims rest on understanding the author actually built or on synthesis the tool performed. The framework implies that assessment practices — in classrooms, in doctoral programs, in peer review — may need to shift from evaluating final products toward evaluating the process by which they were produced, since the product alone can no longer reveal whether the thinking occurred.

Equity considerations complicate the picture. The framework’s implications for responsive digital education include tool selection, assessment, and support for writers with cognitive differences. For some learners, certain demands that are extraneous for others may function differently, and rigid phase prescriptions could either burden or liberate them inappropriately. A student with dyslexia, for example, may experience transcription demands as genuinely extraneous friction that assistive tools should remove, while still needing to own the conceptual struggle of argument development. The framework’s emphasis on human judgment as the operational mechanism — rather than fixed rules about which tools may be used when — is what allows this flexibility. The line between offloadable and non-offloadable work is drawn by a person who understands the learner and the task, not by a policy document or a software default.

What emerges is a design philosophy rather than a prohibition. The paper does not argue that AI should be kept out of research writing; it argues that AI’s role should be calibrated to preserve the cognitive work that makes research meaningful. The productive struggle literature, drawing on work by Hiebert and Grouws and by Warshauer, positions struggle as a means of engaging in and developing critical thinking — not as an obstacle to learning but as its engine. Anne Lamott’s familiar advice to her overwhelmed brother facing an overdue report on birds — take it bird by bird — captures the framework’s practical spirit: large cognitive tasks become manageable when broken into steps, but the steps themselves must be climbed by the writer. AI can hold the ladder, hand over the field guide, and check the citations. What it cannot do is climb, because the climbing is the learning. The framework’s contribution is to specify, phase by phase, where the ladder belongs and where the writer’s feet must stay on the ground.

Subject of Research: A nine-phase human-in-the-loop framework for AI-assisted research writing based on cognitive load theory

Article Title: Designing a human-in-the-loop AI iterative process for research writing: a framework for preserving productive struggle

Article References: Chauncey, S. A. (2026). Designing a human-in-the-loop AI iterative process for research writing: a framework for preserving productive struggle. Frontiers in Computer Science, 8, Article 1947835. https://doi.org/10.3389/fcomp.2026.1947835

Image Credits: AI Generated

DOI: 10.3389/fcomp.2026.1947835

Keywords: artificial intelligence, research writing, cognitive load theory, productive struggle, human-in-the-loop, generative AI, education, academic writing, cognitive offloading, responsive pedagogy, NotebookLM, authorship

The motivation for the work lies in a genuine measurement gap. Classic instruments of self-regulated learning, such as the Motivated Strategies for Learning Questionnaire and the Learning and Study Strategies Inventory, were built for classrooms where the learner, the teacher, and the textbook were the only relevant actors. They were never designed to probe what happens when a student must decide how much to trust a chatbot, when to verify its output, or how to integrate machine-generated material with their own reasoning. Recent reviews of ChatGPT’s impact on education have documented both promising gains, including improved creative problem-solving performance in experimental settings, and worrying risks such as cognitive offloading, in which students outsource thinking they should be doing themselves. Without a validated measure of AI-integrated learning strategies, researchers and educators could not distinguish adaptive, sophisticated AI use from passive dependency.

Aldalham chose gifted upper-secondary students as the target population for both theoretical and practical reasons. Gifted education has increasingly shifted from a static conception of fixed ability toward talent development models that emphasize evolving complexity and creative productivity, frameworks associated with scholars such as Dai, Renzulli, and Subotnik and colleagues. Within these models, high-ability learners are expected to self-regulate intensely, pursue independent inquiry, and push beyond standard curricula, making them an ideal group in which to observe the most advanced forms of AI-integrated strategy. The study also aligns with Saudi Arabia’s National Strategy for Data and AI, which positions artificial intelligence capability as a national priority, giving the research a clear policy context alongside its psychological one.

Methodologically, the study followed an exploratory mixed-methods scale-development design, a sequence widely recommended in contemporary guidelines for constructing psychological measures. Item generation began with a literature-based definition of the construct, refined through expert input and then subjected to quantitative content evaluation using content validity index procedures rooted in the classical work of Lawshe and later refinements by Polit, Beck, and Owen. Only after the item pool survived this content scrutiny did the study move to psychometric testing, which was conducted with a sample of 504 gifted students. Crucially, the sample was randomly split into two halves, one for exploratory analyses and one for confirmatory analyses, a practice that guards against the common pitfall of discovering a factor structure and then validating it in the very same data.

The exploratory phase combined traditional exploratory factor analysis with exploratory graph analysis, a newer psychometric network technique developed by Golino and Epskamp that estimates the number of underlying dimensions by examining the connectivity structure among items. Both approaches converged on the same conclusion: AI-integrated adaptive learning is not one thing but four related things. The first dimension, Autonomous AI-Driven Learning Agency, reflects the degree to which students take initiative in directing their own learning with AI, setting goals and pursuing inquiry rather than waiting for instruction. The second, Human-AI Metacognitive Calibration, captures the delicate skill of monitoring and adjusting trust in AI outputs, knowing when to accept a machine’s answer and when to challenge it, a theme that echoes the adaptive trust calibration literature of Okamura and Yamada.

The third and fourth dimensions extend the construct into higher cognitive territory. Creative-Adaptive Knowledge Innovation measures the extent to which students use AI as a springboard for original thinking, transforming and recombining machine-generated content into something genuinely new rather than merely reproducing it. AI-Augmented Cognitive Orchestration, a term borrowed conceptually from Dillenbourg’s work on classroom orchestration, describes the strategic coordination of cognitive resources, in which the learner allocates tasks between their own mind and the AI system to manage cognitive load and maximize understanding. Together, these four dimensions sketch a portrait of the AI-era learner as an active conductor of a hybrid human-machine cognitive system, not a passive consumer of generated answers.

The confirmatory phase provided strong statistical support for this architecture. Confirmatory factor analyses endorsed both a correlated first-order model, in which the four dimensions relate laterally, and a higher-order model, in which they load onto a single general factor of AI-integrated adaptive learning. Reliability estimates, including modern internal-consistency coefficients such as omega alongside traditional measures, were satisfactory for all four subscales, and a two-week test-retest assessment demonstrated that scores are stable over time, a key requirement for any instrument intended to track students across semesters or interventions. Convergent and discriminant validity indices, evaluated with procedures such as the Fornell-Larcker criterion, fell within acceptable ranges, indicating that the scale measures what it claims and does not merely duplicate existing constructs.

Two additional findings strengthen the scale’s practical credentials. First, tests of measurement invariance showed configural, metric, and scalar invariance across gender, meaning the instrument functions comparably for boys and girls and that score differences between groups can be interpreted substantively rather than as measurement artifacts, following the sensitivity framework articulated by Chen. Second, AI-ALSS scores correlated positively with self-reported grade point average, offering preliminary concurrent criterion-related evidence that students who deploy more sophisticated AI-integrated strategies also tend to achieve more academically. While a self-reported correlation cannot establish causation, it suggests the scale taps something educationally meaningful rather than a stylistic tendency to answer questionnaires favorably.

The implications reach well beyond gifted education in Saudi Arabia. For researchers, the AI-ALSS provides a continuous measure that can finally connect the self-regulated learning tradition, from Zimmerman through Pintrich to contemporary work by Järvelä and Hadwin, to the emerging science of human-AI collaboration championed by Shneiderman’s human-centered AI framework. For teachers and school leaders, the four dimensions offer a diagnostic vocabulary: a student may show strong creative innovation with AI but weak metacognitive calibration, blindly trusting outputs, while another may orchestrate tools expertly yet rarely initiate independent inquiry. Such profiles could guide targeted coaching instead of blanket policies that either ban or uncritically embrace AI. For policymakers navigating the tension between AI literacy goals and fears of cognitive offloading, the scale offers a way to monitor whether integration strategies are actually adaptive.

The author is careful to frame the results as initial rather than definitive. The validation was conducted in a single cultural context with a specific population, and the study itself calls for future work on predictive validity, sensitivity to intervention-related change, and cross-cultural replication. Whether the four-dimension structure holds for mainstream students, for younger learners, or in classrooms with different levels of AI access remains an open empirical question. Yet the study arrives at a moment when such tools are urgently needed, as generative AI reshapes homework, assessment, and the very definition of studying. By transforming a vague intuition, that students learn differently with AI, into a measurable, multidimensional construct with documented reliability and validity, the research gives the field something it has lacked: a ruler for the age of artificial intelligence. How educators, psychologists, and students wield that ruler may shape how the next generation learns to think alongside machines.

Subject of Research: Development and validation of a scale measuring AI-integrated adaptive learning strategies in gifted students

Article Title: Adaptive learning strategies in the age of artificial intelligence: development and validation of a scale for gifted students

Article References: Aldalham, M. S. (2026). Adaptive learning strategies in the age of artificial intelligence: development and validation of a scale for gifted students. Current Psychology, 45(20), Article 1606. https://doi.org/10.1007/s12144-026-10137-1

Image Credits: AI Generated

DOI: 10.1007/s12144-026-10137-1

Keywords: artificial intelligence, adaptive learning, gifted students, scale development, self-regulated learning, metacognition, generative AI, psychometrics, education, Saudi Arabia, cognitive offloading, measurement invariance

The system targets a skill that decades of second-language acquisition research have identified as foundational: morphological awareness, the ability to slice unfamiliar words into their meaningful building blocks. A learner who recognizes that “international” decomposes into the prefix inter-, the root nat, and the suffixes -ion and -al can decode whole families of words instead of memorizing them one by one. Large language models, the authors note, are remarkably good at performing this decomposition in real time, yet the conversational interfaces through which we access them leave the resulting knowledge trapped in unstructured chat logs, fragmented and impossible to consolidate systematically.

Morphowiki’s architecture is deliberately simple and rests on three layers. A generation layer orchestrates seven Agent Skills, small callable command modules that direct the language model through specific tasks. A data layer stores everything as plain-text Markdown files organized into directories for words, prefixes, roots, and suffixes. A review layer runs a lightweight local Node.js server with a single-page web application and a JSON datastore that tracks spaced-repetition scheduling. Because the entire knowledge base lives in Markdown and JSON rather than a proprietary database, it can be read by humans, parsed by programs, diffed by version-control systems, and shared freely, in sharp contrast to the closed data silos of commercial study platforms.

The workflow follows an “initialize, ingest, expand, review” cycle. After a one-command setup that builds the directory structure and installs dependencies, the learner feeds in an unfamiliar word or an entire sentence. The system tokenizes the input, filters out stopwords, both a default list of roughly 150 English function words and a personalized exclusion list the learner can extend, and then applies what the authors call an architecturally bounded recursive decomposition. Given “international,” the pipeline isolates the root nation (itself nat plus -ion), expands to national, and resolves the target word with the prefix inter-, creating a separate Markdown page for each layer and wiring every page bidirectionally to its constituent morphemes. A visited-set mechanism prevents the recursion from looping endlessly.

Once a word is archived, the /flashcard command enrolls it into the review queue after a strict three-part check: the word’s Markdown page must exist, the word must not be on the stopword list, and no duplicate entry may exist in the JSON datastore. Scheduling follows the classic SuperMemo-2 algorithm, initialized with an easiness factor of 2.5 and a zero interval. When the learner runs the review command, a local server identifies due cards, parses the relevant Markdown on the fly, and presents a four-button evaluation, Again, Hard, Good, or Easy. The authors condensed the traditional six-grade SM-2 scale with a mathematical justification: all failed retrievals reset the interval and repetition counters without touching the easiness factor, so they can safely be merged into a single “Again” button.

Two further commands push the system beyond passive flashcards. The /new-word command scans the learner’s accumulated morpheme pool and synthesizes derivatives that are not yet recorded, for example proposing “nationalize” once the root nat has been mastered, then automatically running the ingestion and linking pipeline. Because language models can hallucinate plausible-sounding pseudowords, the skill specification embeds a validation step requiring the model to confirm that each candidate is an attested modern English word. A /lint command then audits the knowledge base in two modes: a deterministic, token-free structural check of link symmetry and schema completeness, and a model-assisted semantic inspection of individual entries. Crucially, lint never edits anything on its own; it only raises flags, and every state change requires explicit human authorization.

What elevates the paper beyond a software description is its quantitative evaluation of the decomposition step. The authors assembled 950 test items, including a main set of 600 words stratified by vocabulary level and morpheme count, drawn from the intersection of a Taiwanese senior-high-school word list and the SIGMORPHON 2022 morpheme segmentation shared task. Against the unsupervised baseline BPE, which scored an F1 of 29.13, and Morfessor 2.0, which reached 47.26, Morphowiki paired with Claude Opus 5 achieved an F1 of 83.06 and an exact-match rate of 74.83 percent. On a subset of 200 words containing four or more morphemes, the baselines managed essentially nothing, 0.00 and 3.00 percent exact match, while Morphowiki still resolved 46.50 percent correctly.

The most intellectually interesting result comes from a two-step ablation that separates the contribution of the underlying model from that of the Agent Skills scaffolding. With the model held fixed, removing the skill file and then the repository’s project instructions revealed that the two components do very different work. The project instructions, which describe the project in etymological terms but prescribe no segmentation rules, actually degraded every measure on the main set and nearly tripled over-segmentation of simple words, from 6.7 to 23.3 percent. The skill file, by contrast, added more than 52 points of exact-match accuracy on words recorded in canonical form and corrected 17 of the over-segmentation errors the instructions induced. The authors conclude that the skill does not teach the model how to segment strings, a capability it already possesses, but rather imposes canonical discipline: knowing when to restore a base to its dictionary form, when to record an assimilated prefix canonically, and where recursion should stop.

The team is candid about the system’s limits. Decomposition quality remains tethered to the reliability of the underlying model, and the deployed configuration still over-segments monomorphemic words at 12.0 percent, a trade-off traced to the etymological priming of the project instructions rather than the skill itself. The reference standard used for scoring is itself noisy, with coverage of the most elementary vocabulary at only 43.1 percent, so the reported accuracies likely underestimate true performance. Cross-model comparison covered only two models from a single provider, the pipeline is calibrated exclusively for English derivational morphology, and cost is nontrivial: roughly 9 to 23 US dollars per thousand words processed, which positions the framework for curated vocabulary construction rather than bulk corpus processing. Pedagogical efficacy, finally, has not yet been tested in classrooms.

Even with those caveats, Morphowiki speaks to a tension now rippling through education research. Recent studies on cognitive offloading and “metacognitive laziness” suggest that instantaneous AI answers can undermine the deep semantic engagement that durable memory requires, a concern rooted in the classic levels-of-processing work of Craik and Lockhart. By forcing learners through active retrieval, spaced review, and human-approved curation, Morphowiki is designed to make the AI a builder of infrastructure rather than a dispenser of answers. The authors frame it as a reproducible testbed on which researchers in computer-assisted language learning can systematically compare models, prompting strategies, and scheduling algorithms on a standardized, transparent data format. If the broader lesson holds, the future of AI-assisted learning may depend less on how cleverly a chatbot explains a word, and more on whether the explanation leaves behind something the learner can keep.

Subject of Research: An open-source Agent Skills framework that uses large language models to decompose English vocabulary into morphemes, build bidirectionally linked knowledge graphs, and schedule spaced-repetition review.

Article Title: Morphowiki: reusable agent-skill infrastructure for morphology-centered vocabulary knowledge construction

Article References: Chen, Y.-T., Yao, K.-C., Chang, S.-H., & Lin, Y.-C. (2026). Morphowiki: reusable agent-skill infrastructure for morphology-centered vocabulary knowledge construction. SoftwareX, 36, Article 103114. https://doi.org/10.1016/j.softx.2026.103114

Image Credits: AI Generated

DOI: 10.1016/j.softx.2026.103114

Keywords: Morphowiki, large language models, morphology, vocabulary acquisition, spaced repetition, Agent Skills, computer-assisted language learning, open-source software, second language acquisition, cognitive offloading, Markdown knowledge graph, SM-2 algorithm

Christian, a research fellow with the UC Berkeley Center for Human-Compatible AI, was part of a team behind a landmark study that found that spending just ten minutes relying on an AI tool significantly impairs our ability to focus and persist with difficult tasks. A preliminary draft of the work made waves earlier this year, and the team presented the updated, peer-reviewed paper this week at the Conference on Language Modeling. The collaboration included scholars from Carnegie Mellon University, the Massachusetts Institute of Technology, the University of Oxford and UCLA, and its central result is deceptively simple: a brief exposure to AI assistance measurably degrades independent performance the moment that assistance is withdrawn. “It’s a striking finding, but in a way it supplies ammunition for a story that a lot of us kind of feel in our gut,” Christian said. “I experience it too.”

To test how quickly people’s focus and persistence plummeted, the researchers designed a series of randomized controlled trials and recruited 1,222 participants online. The first experiment involved 354 participants. One group was tasked with solving 15 basic fraction problems without any help, while another group had ChatGPT open on the side to use for assistance, or even to ask directly for the answer. The group with AI access started off more accurate than the group working unaided, which is exactly what one would expect from a capable assistant. Then, after 12 problems, the researchers took the AI tool away. Almost immediately, the people in that group stopped solving questions accurately. The advantage they had enjoyed while the tool was available evaporated as soon as they had to rely on their own reasoning.

A similar pattern emerged when the team ran the experiment on a second, larger group of 667 participants. Whereas the AI users got answers wrong or gave up entirely when the assistance was withdrawn, the group that had never used AI stuck with the task and were more successful at the end. The data also revealed a telling behavioral shift: there was a significant drop-off in the solve rate for the group that had access to the AI tool after the tool was removed, and a sharp increase in the rate at which those participants skipped questions. In other words, the problem was not merely that people performed worse without the machine. It was that they became less willing to try at all, abandoning problems that the unassisted group continued to attack.

The researchers then ran a third experiment with 201 participants using an SAT reading comprehension prompt, a task far removed from arithmetic. The results echoed those from the fraction problems: when the AI tool was removed, both persistence and accuracy dropped. This consistency across two very different domains suggests that the effect is not about math anxiety or a lack of practice with fractions. It points to something more general about how quickly the human mind adapts to the presence of an always-available answer machine, and how poorly that adaptation serves us when the machine is taken away. “You almost couldn’t ask for a clearer story,” Christian said.

The findings back up what anecdotes and surveys have increasingly suggested in recent years: an overreliance on AI for answers erodes our cognitive fitness. The mechanisms the researchers describe are familiar to anyone who has studied how habits form. When AI produces answers in seconds, we get used to that speed, and anything that takes longer begins to feel inefficient. The friction that once accompanied difficult thinking, the slow and sometimes frustrating process of wrestling a problem into submission, starts to feel like a defect in the workflow rather than an essential part of learning. Relying on AI also removes what the team called “productive struggle,” the effortful engagement with difficulty that can lead people to discover their research strengths and passions in the first place.

There is a deep irony here that Christian is quick to acknowledge. These systems were built to be helpful, and by almost any benchmark they are. They answer accurately, quickly and on demand. But helpfulness in the short term and helpfulness in the long term can pull in opposite directions. “We’ve built these systems to be helpful, but are they being helpful to us in a way that’s actually helping us?” Christian said. “We showed, ironically, that they are often helping us in ways that are kind of unhelpful.” A tool that hands over the answer saves time on this problem and quietly costs capacity on the next one, leaving the user dependent on the very assistance that made them feel more capable.

The study attracted wide attention as its findings matured. Outlets from WIRED to Psychology Today covered the research when early results were released in the spring, and The New York Times published a story last week about the paper and similar work on how chatbots may affect thinking and brain development. Notably, frontier AI companies themselves have acknowledged how offloading tasks to chatbots can reduce a person’s baseline understanding and mastery of a subject. That admission from within the industry gives the academic findings added weight, because it suggests the builders of these systems recognize that the default interaction pattern, in which the model simply supplies the answer, may not be the one that best serves human learning over time.

Christian argues that the design of AI systems could change in response. Instead of defaulting to quick answers, AI responses could default to being more instructive, behaving more like a tutor that guides a learner toward understanding than like an oracle that dispenses solutions. The distinction matters because a tutor’s goal is to make itself unnecessary, transferring skill to the student, while an answer machine’s goal is to minimize the time to a correct response, which can leave the underlying skill untouched or even atrophied. “We need to work toward a future in which human abilities are, as much as possible, augmented rather than supplanted,” Christian said. The ten-minute finding suggests that the window in which that choice gets made is startlingly short, and that the direction of the effect, augmentation or replacement, is set almost as soon as the habit of asking takes hold.

The stakes, Christian believes, extend well beyond classrooms and test questions, and they are especially acute in research and academia. He worries about the way overreliance on AI tools in science could jeopardize what makes for a good scientist in the first place: staying close to data, being immersed in the material, and working as part of a team. Something significant could be lost, he suggests, when the team of ten developing the ideas that lead to a breakthrough is whittled down to two or three researchers and a chatbot. The informal exchange of half-formed hypotheses, the patient debugging of a stubborn analysis, the shared struggle that builds both expertise and judgment, all of these are the substrate of discovery, and none of them can be outsourced without cost. “This is not a story about fractions and SAT problems,” Christian said. “This is something that is happening to human knowledge and human expertise from top to bottom and from grade-school students all the way to the leading experts in the world.”

Subject of Research: The effect of brief AI assistance on human persistence and independent problem-solving performance

Article Title: Using AI for just 10 minutes erodes your ability to persist at hard things

Article References: Using AI for just 10 minutes erodes your ability to persist at hard things. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: artificial intelligence, cognitive persistence, ChatGPT, randomized controlled trial, productive struggle, human-compatible AI, education, cognitive offloading, Conference on Language Modeling, problem-solving, academic research, UC Berkeley