Low confidence was defined as a score of 3 or lower on a five-point Likert scale. Among the 97 respondents were 35 internal medicine residents and 61 family medicine residents: 36 first-year, 45 second-year and 14 third-year residents. Confidence was lowest for localization, with a median score of 3 and an interquartile range of 2–4.

Confidence in forming a differential was also lower than elsewhere in the assessment, with a median of 4 and an interquartile range of 3–4. Confidence did not increase with additional years of postgraduate training. Residents instead reported that greater exposure to neurology and additional teaching would improve their confidence.

The survey also did not directly test whether low confidence caused residents to omit neurological examinations or established any effect on patient outcomes. Residents rated their peers as most likely to skip neurological and musculoskeletal examinations. For neurological examinations, the median perceived likelihood of skipping was 3 on the five-point scale, with an interquartile range of 2–3; for musculoskeletal examinations, the median was also 3, with an interquartile range of 2–4.

The reported likelihood of skipping part of a neurological examination increased from postgraduate year one to year two. In free-text responses, residents cited low confidence and lack of time as major contributors to examination avoidance. The findings offer a limited snapshot because the study was conducted at a single centre, only 97 of 350 eligible residents responded, and both confidence and perceived peer behavior were measured by self-report rather than direct observation.

The results therefore cannot establish actual rates of neurological examination omission in clinical practice. Within those constraints, the survey identified low confidence in localization and differential diagnosis as prominent concerns and reported lack of time as an additional practical barrier.

Subject of Research: Neurological examination confidence and exam avoidance among medical residents

Article Title: Confidence in the neurological examination and exam avoidance among medical residents

Article References: Keeling, A. B., & Daniels, V. J. (2026). Confidence in the neurological examination and exam avoidance among medical residents. PLOS One, 21(10), e0360249. https://doi.org/10.1371/journal.pone.0360249

Image Credits: AI Generated

DOI: 10.1371/journal.pone.0360249

Keywords: medical residents, neurological examination, clinical confidence, exam avoidance, localization, differential diagnosis, medical education, neurology training, Confidence, neurological, examination, exam

The review, conducted by Shakiba Moradi and Pete Evans, synthesized literature published between 2015 and 2026, identified through structured searches of PubMed, Scopus, Web of Science, and Google Scholar. Rather than estimating pooled intervention effects, the authors organized findings thematically across four domains: the nature and sources of AI anxiety, the development of AI confidence and self-efficacy, the emotional and motivational mechanisms associated with engagement, and the individual, disciplinary, cultural, and institutional conditions that alter these relationships. Self-Determination Theory, Bandura’s Self-Efficacy Theory, and research on emotion-cognition interactions provided the principal interpretive lenses.

Across the reviewed literature, AI anxiety was generally associated with avoidance, reduced perceived control, cognitive burden, and reluctance to experiment. Students who perceive AI as a threat to their intellectual autonomy or career relevance may experience lower participation and cognitive overload. Confidence and self-efficacy, by contrast, were linked to persistence, exploration, and more active use of AI-supported learning tools. Yet the review cautions that these relationships are neither uniform nor necessarily causal, and that much of the current evidence relies on cross-sectional, self-reported data, making it difficult to determine whether anxiety causes disengagement, whether disengagement intensifies anxiety, or whether both are shaped by unmeasured contextual factors.

At a conceptual level, AI anxiety resembles broader technostress but carries unique features. Traditional technostress stems from overload or complexity, whereas AI anxiety is entangled with existential and epistemic concerns: the fear of being replaced, misjudged, or outperformed by intelligent systems. It also encompasses uncertainty about ethical use, academic integrity, and data privacy. The review identifies three intertwined dimensions: cognitive, reflecting confusion and overload; emotional, manifesting as fear, stress, or avoidance; and existential, tied to meaning and identity. Cognitive psychology offers further insight, noting that anxiety thrives where predictability is low and perceived demands exceed perceived resources, conditions that opaque neural networks and probabilistic models resist intuitive comprehension.

Confidence operates through the opposite loop. When students experience mastery or even partial success, they interpret that success as evidence of capability, fueling motivation and deeper exploration. Each subsequent encounter reinforces efficacy beliefs, creating a positive spiral of growth. Bandura identified four principal sources of efficacy beliefs: mastery experiences, vicarious experiences, social persuasion, and physiological states. Early exposure to low-stakes experimentation is crucial, as is constructive feedback that highlights human judgment alongside AI suggestions. Confidence is also socially contagious; observing peers who use AI effectively normalizes engagement and reduces intimidation. Importantly, the review warns that overconfidence can lead to uncritical trust in machine output, so the optimal stance is calibrated confidence, a balanced awareness of both AI’s utility and its limitations.

Self-Determination Theory offers a complementary framework. The theory emphasizes three universal psychological needs: autonomy, competence, and relatedness. AI technologies, depending on their design, can either enhance or erode each. Adaptive platforms can personalize pace and difficulty, allowing students to pursue mastery in self-directed ways, yet opaque algorithms that make decisions invisible to the user diminish ownership. Competence involves not only task success but emotional appraisal; feedback devoid of empathy can reinforce the impression that the learner is being judged by an unfeeling machine. Relatedness may suffer if students perceive AI as replacing human contact, though partially automating routine feedback can free educators to focus on mentorship and dialogue.

The review also maps contextual factors that shape these relationships. Prior technological experience and AI literacy are among the most consistent moderators; students with greater digital competence tend to approach AI with curiosity rather than fear. Gender and identity dimensions, disciplinary culture, cultural narratives and media discourses, and institutional climate all influence how AI is interpreted, as a learning resource, an evaluative threat, or a challenge to human agency. Universities that present AI integration as a top-down mandate may inadvertently evoke resistance, while those that frame it as co-creation foster ownership. Assessment policies that penalize improper AI use without clear guidance can heighten anxiety by casting technology as a disciplinary threat rather than a learning tool.

Drawing on recurring patterns across the literature, the authors propose a provisional five-stage framework of emotional adaptation: curiosity, anxiety, experimentation, confidence, and flourishing. The framework is explicitly intended as a heuristic rather than a universal or empirically validated sequence; students may enter at different points, move between stages, remain ambivalent, or experience confidence and anxiety simultaneously. Curiosity marks the initial encounter, when novelty stimulates exploration but understanding remains limited. Anxiety intensifies as complexity becomes apparent and questions arise about correctness, judgment, and obsolescence. Experimentation follows once anxiety stabilizes, as risk-taking replaces avoidance and mistakes are reframed as data rather than failure. Confidence solidifies through accumulated mastery experiences, and flourishing represents the integration of cognitive mastery, emotional stability, and ethical awareness, a dynamic equilibrium rather than a permanent state.

For educational practice, the review supports a layered response rather than a single intervention. At the instructional level, students should receive scaffolded opportunities to use AI in low-stakes settings, compare AI outputs with human reasoning, identify errors and biases, and reflect on their own emotional and cognitive responses. At the institutional level, AI literacy should include not only operational skills but also understanding of uncertainty, hallucination, bias, privacy, authorship, and appropriate verification. Clear policies are needed to distinguish acceptable educational use from misconduct, and institutions should explain when AI is being used in assessment, advising, monitoring, or feedback and how student data are collected and interpreted. Human mediation remains essential; faculty guidance, peer support, mentorship, and psychologically safe discussion can help students interpret technological difficulty as part of learning rather than evidence of personal inadequacy.

Ultimately, the review argues that flourishing engagement should not be equated with uncritical enthusiasm for artificial intelligence. It is better understood as the capacity to engage with AI confidently, reflectively, ethically, and without surrendering human judgment. Educational success in the age of AI will depend not only on technological capability but also on whether institutions create conditions in which students can question, evaluate, and use these systems with informed agency. The authors call for longitudinal, experimental, qualitative, and mixed-method research to test the proposed mechanisms and identify which interventions most effectively support students who experience AI-related anxiety, noting that the direction and temporal ordering of the proposed relationships remain theoretically inferred rather than conclusively established.

Subject of Research: Educational Psychology

Article Title: How AI anxiety and confidence shape student engagement in higher education

Article References: Moradi, S., & Evans, P. (2026). How AI anxiety and confidence shape student engagement in higher education. Discover Education, 5(1), Article 857. https://doi.org/10.1007/s44217-026-02078-5

Image Credits: AI Generated

DOI: 10.1007/s44217-026-02078-5

Keywords: AI anxiety, student engagement, self-efficacy, higher education, emotion regulation, AI literacy, anxiety, confidence, shape, student, engagement, higher

The research, led by Juliet Aizer, Sarah P. Gottesman, Kirana Gudi, Xiaoyue Ma, Julia T. Nguyen, Stephen A. Paget, and Lisa A. Mandl at Weill Cornell Medicine and the Hospital for Special Surgery in New York, set out to answer a question that has long troubled medical educators. Previous studies had shown that providing primary care physicians with education about knee osteoarthritis improves their confidence in the knowledge and skills needed to care effectively for these patients. What remained unknown was whether educational programs could be similarly effective in medical trainees, who are at an earlier stage of professional development and whose schedules are notoriously packed.

The team developed an online program specifically designed to improve learners’ knowledge, skills, and confidence in their abilities to recognize, diagnose, and manage patients with or at risk for knee osteoarthritis. The program was offered to Internal Medicine trainees during their rheumatology rotation in the 2023–2024 academic year at a large urban residency program. This timing was deliberate: the rheumatology rotation is where trainees are most likely to encounter joint disease, making it a natural window for targeted learning about a condition they will later manage far more often in primary care settings than in specialty clinics.

To measure the program’s effects, the researchers used a validated 27-item instrument administered both before and after the educational intervention. The primary outcome was the change in confidence related to the knowledge and skills involved in caring for patients with or at risk of knee osteoarthritis. Confidence, in this context, is not a vague feeling but a measurable construct that medical education researchers track because it influences whether clinicians apply what they know in practice. Pre- and post-program scores were compared using the Wilcoxon signed rank test, a non-parametric statistical method well suited to paired data that do not follow a normal distribution, which is typical of small before-and-after studies in residency training environments.

The results were unambiguous. Of the 36 Internal Medicine residents enrolled, 24 completed the online program between October 2023 and June 2024. Confidence in all knee osteoarthritis related domains showed statistically significant improvements after program completion. Every participant agreed or strongly agreed that the program content was useful and relevant to their scope of practice. In medical education research, where interventions often produce mixed or marginal effects, uniform improvement across all measured domains is a notable technical achievement and speaks to the precision with which the curriculum was matched to the trainees’ actual learning needs.

The significance of these findings extends beyond a single residency program. Primary care physicians are frequently responsible for recognizing and managing patients’ knee osteoarthritis, a condition that affects hundreds of millions of people worldwide and is a leading cause of chronic pain and disability. Yet studies consistently show that these physicians often feel underprepared for the task. If trainees graduate with greater confidence in managing the condition, the argument runs, they may be better equipped to deliver guideline-based care from the first day of independent practice, potentially improving outcomes for an enormous patient population.

The online format itself carries important implications for how medical education may evolve. Web-based curricula can be scaled across institutions, completed at times that fit around clinical duties, and standardized so that every learner receives the same high-quality content regardless of where they train. In a large urban residency program, where trainees rotate through demanding clinical services and formal teaching time is scarce, an asynchronous online module offers a practical route to competency-based learning that traditional lecture formats struggle to match. The study’s authors note that the program was designed to improve knowledge, skills, and confidence together, reflecting a modern understanding that competence in medicine depends on all three elements working in concert.

The researchers are careful about the limits of what their study demonstrates. The instrument measured confidence, not clinical performance, and the authors explicitly state that whether the observed gains translate into improved patient care deserves further study. The completion rate of 24 out of 36 residents, while respectable, also means that a third of eligible trainees did not finish the program, and the study was conducted at a single institution during one academic year. Confidence gains measured immediately after an intervention may also fade over time, a phenomenon known in the education literature that would require follow-up assessment to rule out.

Nevertheless, the study offers a technically rigorous proof of concept that online education can move the needle for trainees, not just for practicing physicians. The work was supported by the Pfizer 2020 Osteoarthritis and Osteoarthritis Pain Virtual Learning Competitive Grant Program, the Hospital for Special Surgery Academy of Medical Educators, and the Weill Cornell Medicine Clinical and Translational Science Center. The study was conducted in accordance with the Declaration of Helsinki, approved and granted an exemption by the Hospital for Special Surgery’s Institutional Review Board, and all responses were de-identified, with no individual participation metrics shared with residency program leadership.

For a condition that will dominate the future caseload of nearly every primary care physician, the message of this research is quietly powerful. A well-targeted online program, delivered at the right moment in training, can give young doctors the confidence to recognize, diagnose, and manage knee osteoarthritis before they ever face it alone. The next step, the authors suggest, is to determine whether that confidence becomes better care for the millions of patients whose knees hurt when they walk, climb stairs, or simply get out of a chair in the morning.

Subject of Research: Medical Education

Article Title: Assessing an online program’s effects on residents’ confidence to care for patients with or at risk of painful knee osteoarthritis

Article References: Aizer, J., Gottesman, S. P., Gudi, K., Ma, X., Nguyen, J. T., Paget, S. A., & Mandl, L. A. (2026). Assessing an online program’s effects on residents’ confidence to care for patients with or at risk of painful knee osteoarthritis. BMC Medical Education. https://doi.org/10.1186/s12909-026-10209-7

Image Credits: AI Generated

DOI: 10.1186/s12909-026-10209-7

Keywords: knee osteoarthritis, medical education, online learning, Internal Medicine residents, primary care, rheumatology, confidence, BMC Medical Education, Assessing, online, program, effects

The study, conducted from June to November 2025, used a structured online questionnaire distributed through pharmacy-related social media groups and professional pages across Kosova. Participants included pharmacists and pharmacy technicians directly involved in dispensing medicines and counseling patients. The 29-item instrument covered demographic and professional background, awareness of opioid abuse and misuse, attitudes toward counseling and harm reduction, reported practices, and self-reported confidence related to naloxone administration. Most participants were female (75.8 percent), aged 19 to 30 (57.7 percent), and held a Master of Pharmacy degree (65.4 percent). More than half reported one to four years of work experience, and most worked full time in urban pharmacies.

The results paint a picture of a workforce that is well informed on paper. A striking 93.5 percent of respondents reported understanding the difference between opioid abuse and opioid misuse, terms that carry distinct clinical meanings: misuse refers to using prescription opioids in a manner other than prescribed, while abuse refers more broadly to harmful or non-medical opioid use. Familiarity with naloxone and its mechanism of action was reported by 81.5 percent of participants. Internal consistency of the awareness items was good, with a Cronbach’s alpha of 0.81, suggesting the questions reliably captured a coherent construct of self-reported knowledge.

Attitudes toward professional involvement were similarly encouraging. Three quarters of respondents (75.8 percent) agreed that pharmacies are appropriate settings for counseling individuals suspected of opioid abuse, and 75.1 percent said they were willing to provide such counseling. Support for dispensing opioids only with a valid medical prescription was nearly universal at 93.5 percent, and 94.6 percent agreed that pharmacy professionals should not sell medications to individuals who appear to be abusing them. Almost all respondents (95.0 percent) agreed that pharmacy staff should be trained to recognize individuals with substance misuse, and 98.5 percent said staff should be informed about medications with abuse potential. The attitude scale showed acceptable internal consistency with a Cronbach’s alpha of 0.76.

Yet beneath these positive surface indicators, the survey exposed significant vulnerabilities. Only 56.2 percent of participants reported feeling either confident enough or very confident in administering naloxone in their pharmacy setting, while 43.8 percent reported not feeling confident at all. This is a substantial proportion of a workforce that, in many countries, is increasingly expected to serve as a frontline response to opioid overdose. More than half of respondents (54.2 percent) also expressed concern that administering naloxone would require more time than they could provide in routine pharmacy practice, pointing to workflow pressures as an additional barrier beyond knowledge alone.

Work experience emerged as a meaningful variable across multiple measures. Chi-square tests revealed statistically significant associations between years of practice and several responses, including views on prescription-only opioid dispensing (p = 0.013), training in recognizing substance misuse (p = 0.040), naloxone-related education (p = 0.001), and time constraints around naloxone administration (p = 0.027). Most notably, self-reported confidence in administering naloxone was significantly associated with work experience (p = 0.017), and familiarity with naloxone showed an even stronger association (p = 0.001). These findings align with previous international research suggesting that more experienced pharmacists often report greater comfort when addressing opioid-related situations and engaging in patient counseling.

The regulatory context in Kosova adds a layer of complexity that the authors emphasize. Although national legislation governs medicinal products and retail pharmacy services, the current framework does not explicitly provide community pharmacists or pharmacy technicians with the authority to administer naloxone. This means that participants’ responses regarding confidence in naloxone administration should be understood as perceptions of a potential role rather than reflections of established practice. The authors note that this regulatory uncertainty is particularly relevant when assessing pharmacy professionals’ willingness and confidence to participate in opioid overdose management, and that clearer professional and regulatory guidance may be needed to address the gap.

Not all findings were uniformly positive. While 78.1 percent of respondents reported suggesting safer alternatives when misuse was suspected, only 31.9 percent reported contacting nearby pharmacies to inform them of an individual suspected of opioid misuse. This pattern, in which pharmacists feel comfortable with individual counseling but hesitate when broader professional collaboration is required, has been observed in other studies and may reflect barriers to inter-professional communication that were not directly examined in the present research. Additionally, 38.5 percent of participants agreed that the sale of controlled medications without a prescription represents an important source of pharmacy income, a finding the authors say should be interpreted cautiously as it reflects perceptions rather than dispensing practices.

The study has several limitations that shape how the results should be read. The cross-sectional design captures attitudes at a single point in time and cannot establish causality. Responses were self-reported rather than directly observed, meaning actual behavior and practical competence may differ from what participants described. Social desirability bias may have influenced answers to questions about professional responsibility and ethical practice. The voluntary online recruitment through social media may have attracted pharmacy professionals with a stronger interest in public health, potentially skewing responses toward more positive attitudes. The questionnaire was developed specifically for this study without formal pilot testing, and confidence in naloxone administration was assessed using a single item. Pharmacists and pharmacy technicians were also analyzed together without separate analysis by professional qualification.

Despite these limitations, the findings carry practical weight for a region where empirical evidence on pharmacy involvement in opioid harm reduction has been scarce. The authors conclude that strengthening the role of community pharmacy professionals in opioid abuse prevention will require more than awareness alone. Practice-oriented training, clear professional guidance, and supportive regulatory frameworks are needed to translate positive attitudes and self-reported knowledge into genuine confidence at the counter. International evidence suggests that successful pharmacist involvement in opioid harm reduction is most likely when training, regulatory clarity, and institutional support are aligned, and in settings where access to specialized addiction services is limited, community pharmacies may represent one of the most promising settings for early identification of opioid-related concerns and overdose prevention.

Subject of Research: Community pharmacists' awareness, attitudes, and confidence regarding opioid abuse and naloxone use in Kosova

Article Title: Community pharmacists’ awareness, attitudes, and confidence regarding opioid abuse and naloxone use in Kosova: A cross-sectional survey

Article References: Ibraimi, Z., Haxhiu, A., Shabani, D., Qeriqi, V., Bresa, A., & Haliti, D. (2026). Community pharmacists’ awareness, attitudes, and confidence regarding opioid abuse and naloxone use in Kosova: A cross-sectional survey. Pharmacia, 73, Article e204449. https://doi.org/10.3897/pharmacia.73.e204449

Image Credits: AI Generated

DOI: 10.3897/pharmacia.73.e204449

Keywords: opioid abuse, naloxone, community pharmacists, Kosova, harm reduction, cross-sectional survey, Community, pharmacists, awareness, attitudes, confidence, regarding

The researchers set out to test a prediction that had circulated in behavioral economics for years but had rarely been put to a direct, quantitative test. The idea is simple enough. When you are confident about a quantity, there is little reason to blur it. But when your internal estimate is noisy and unreliable, rounding to a prominent number such as 50 or 100 feels like an acceptable way to communicate a vague answer. Uncertainty, in other words, should predict rounding. Earlier experimental work, including studies by Ruud, Schunk, and Winter, had shown that people round more when they are made to feel uncertain, but the field lacked evidence that naturally occurring differences in uncertainty, measured within the same individuals across many estimates, actually forecast which responses get rounded.

To close that gap, the team turned to secondary analyses of two independent datasets, one involving numerosity estimation and the other involving temporal estimation. This dual approach matters because of a influential idea in cognitive science known as the theory of magnitude, proposed by Vincent Walsh in 2003, which holds that the brain represents time, space, number, and other quantities with a common metric code. If the round-number bias emerges from a general property of magnitude representation, it should appear in both domains, and its strength should track uncertainty in both domains as well. Testing two different quantity types with the same analytical logic gave the researchers a way to check whether the effect is domain-specific or genuinely general.

The datasets themselves came from earlier work by the same research group on what psychologists call metric error monitoring, the ability to judge not just how long something lasted or how many items were present, but how good your own estimate was. In these tasks, participants first produced an estimate and then rated their confidence in it, allowing the researchers to link each individual judgment to a metacognitive signal. That design is what makes the new analysis powerful: instead of comparing confident and anxious people, the team could ask whether the same person rounds more on the trials where they feel less sure of their answer.

The results supported the uncertainty hypothesis. Estimates that received a confidence rating indicating lower certainty were significantly more likely to be rounded to a multiple of five. The relationship held in both numerosity and temporal estimation, suggesting that the bias is not a quirk of counting objects or judging durations but a general strategy for coping with noisy internal signals. The finding dovetails with an older theoretical account from Yaniv and Foster, who argued in 1995 that judgment under uncertainty involves a trade-off between accuracy and informativeness. A coarse, round answer may be less precise, but it is easier to process, easier to remember, and easier to communicate. When the underlying evidence is weak, the mind apparently decides that the loss of precision is worth the gain in simplicity.

The study also tested a second, subtler prediction that flows from one of the best-documented laws in psychology. Magnitude estimates obey what is known as the scalar property, the human version of Weber’s law: the variability of an estimate grows in proportion to the size of the quantity being estimated. Judging whether a tone lasted two seconds or three seconds is easy; judging whether it lasted twenty seconds or thirty seconds is much harder, because the internal noise scales with duration. If uncertainty drives rounding, then larger magnitudes should naturally attract more rounding, since they are inherently more uncertain to represent. That is exactly what the data showed. Round-number bias increased at larger magnitudes in both estimation domains, providing an elegant confirmation that the bias follows the statistics of the mind’s own measurement apparatus.

Together, these two findings sketch a coherent cognitive architecture for the round-number bias. The internal representation of any magnitude comes with an uncertainty signature, a kind of built-in error bar. When that error bar is wide, either because the person lacks confidence or because the quantity is large enough that scalar variability dominates, the response system gravitates toward prominent, accessible numbers. This connects the bias to a broader family of metacognitive phenomena. Research by Fleming and Lau on how to measure metacognition, and by Kononowicz, Roger, and van Wassenhove on the neural decoding of temporal confidence, suggests that the brain continuously monitors the reliability of its own signals. The new study implies that this monitoring does not just color our feelings of confidence; it actively shapes the numbers we produce.

The practical implications stretch well beyond the laboratory. Financial behavior is riddled with round numbers: stock prices cluster at whole-dollar values, limit orders pile up at prominent levels, and studies of equity crowdfunding and German real estate transactions have documented the same attraction to round figures. Survey researchers have long worried about heaping, the tendency of respondents to report their age, income, or financial expectations in round numbers, which distorts datasets and can mask real variation. The new findings suggest that heaping is not merely laziness or satisficing, the cognitive shortcut of giving a good-enough answer. It is a systematic output of uncertainty, which means that the amount of rounding in a dataset could itself serve as a diagnostic of how uncertain respondents are, a possibility that researchers studying inflation expectations and other economic beliefs may find valuable.

There are also implications for how we interpret the numbers other people give us. Work by Jerez-Fernandez, Angulo, and Oppenheimer showed that listeners use precision as a cue to a speaker’s confidence, treating a precise estimate as a sign of expertise. The new study adds the flip side of that coin: when someone hands you a round number, it may be a signal, whether intended or not, that their internal evidence was weak. In negotiations, market forecasts, or eyewitness estimates of time and distance, the shape of the number carries information about the mind that produced it. Conversely, the finding that larger magnitudes attract more rounding offers a useful corrective, since a round estimate of a large quantity may reflect nothing more than the ordinary scaling of noise rather than any special ignorance on the part of the estimator.

The authors are careful to frame their work as a step toward understanding rather than a final verdict. Because the analysis relied on secondary datasets, future studies can manipulate uncertainty directly and track rounding trial by trial, and researchers can explore whether the same uncertainty-to-rounding link appears in other magnitude domains such as line length, weight, or distance. Still, the core message is striking in its simplicity. The round numbers that saturate human communication, from price tags to promises of five more minutes, are not arbitrary cultural conventions. They are the visible fingerprints of an internal measurement system that knows, with remarkable sophistication, exactly how unsure it is, and chooses its numbers accordingly.

Subject of Research: The role of estimation uncertainty in producing the round-number bias in numerosity and temporal magnitude judgments

Article Title: Uncertainty levels in magnitude estimations predict round number bias

Article References: Öztel, T., Ünübol, H., & Balcı, F. (2026). Uncertainty levels in magnitude estimations predict round number bias. Current Psychology, 45(17), Article 1486. https://doi.org/10.1007/s12144-026-10027-6

Image Credits: AI Generated

DOI: 10.1007/s12144-026-10027-6

Keywords: round number bias, uncertainty, magnitude estimation, temporal estimation, numerosity estimation, metacognition, confidence, scalar property, Weber's law, behavioral economics, numerical cognition, Current Psychology

The team’s experimental platform was deceptively simple. Observers viewed two coloured patches of moving dots, and after a brief delay a colour cue told them which of the two motion directions they had to reproduce by rotating an arrow. Because responses were continuous rather than categorical, the precision of each reproduction could be measured to a fraction of a degree, and the distribution of errors across hundreds of trials could be fed into a well-established mathematical framework known as the neural resource model. In that model, a visual stimulus is encoded by a population of idealized direction-selective neurons with bell-shaped tuning curves, and the total activity of the population is held fixed by divisive normalization, a canonical neural computation. Crucially, the model allows a gain parameter to be fitted freely, revealing how the fixed pool of spiking activity was divided between the two competing stimuli. A narrower hill of activity means more resources and more precise reports; a broader, heavier-tailed error distribution betrays a starved representation.

In the first experiment, the researchers replicated the classical external reward effect with a twist. Accurate recall of one colour earned observers fifteen points, while accurate recall of the other earned only five, and the accumulated points translated into a bonus payment. Reproduction errors were reliably smaller for the high-reward colour, and the fitted gain parameter showed that roughly sixty-five per cent of the neural resources were channelled towards the privileged stimulus. Yet when the team computed the allocation that would truly maximize the expected point total, a strategy that would have demanded starving the low-reward item far more aggressively, the observers fell short. They distributed resources more equally than a reward-maximizing policy required, earned fewer points per trial than the optimum predicted, and showed essentially no correlation between their individual allocations and their individual optima. Something beyond the pursuit of points, the authors reasoned, was holding the allocation in check, and the most obvious candidate was that being accurate across the board is itself satisfying.

To test that idea directly, the researchers turned to deception. In a second series of experiments, they manipulated the feedback displayed after each response, silently magnifying the apparent error for stimuli of one colour and shrinking it for the other. Observers had no idea the feedback was distorted; none suspected the manipulation when debriefed, yet eighty-four per cent came to describe the error-magnified colour as harder to remember. The consequences were striking. Reproductions of the error-minified stimulus, the one that made observers feel competent, became genuinely more precise, even though the manipulation had changed nothing about the stimulus itself or the observers’ true accuracy. A control experiment in which the two stimuli appeared sequentially, eliminating competition at the moment of encoding, abolished the effect entirely, pinpointing attentional competition during perception rather than memory maintenance as the locus of the bias. A further refinement showed that the asymmetry was driven by the pleasant side of the manipulation: shrinking errors boosted precision, but inflating them did not reliably impair it.

The third series of experiments replaced fabricated feedback with honest variation in stimulus quality. On most trials, one colour carried dots moving with eighty-five per cent coherence, an easy, low-noise motion signal, while the other carried only forty-five per cent coherence, a noisy, difficult one. On interleaved probe trials, both stimuli were shown at the same intermediate coherence, so any difference in precision had to reflect learned associations rather than the stimuli themselves. Once again, the colour previously paired with the easy, confidence-inspiring signal was reproduced more accurately even when it was objectively no easier than its rival. The fitted resource model estimated that observers allocated nearly twice as much neural activity to the high-coherence colour, and the effect survived strict fixation control in a laboratory version of the task with eye tracking, ruling out overt shifts of gaze as an explanation. When stimuli were presented sequentially, the advantage dissolved, echoing the conclusion that the bias operates through attentional competition at encoding.

Perhaps the most provocative finding is that these internally driven allocations were not merely suboptimal; they were actively counterproductive. The researchers simulated the allocation policy that would minimize overall response error in the difficulty experiments and found it was essentially equal division of resources, because pouring resources into a noisy stimulus cannot rescue an estimate drowning in perceptual noise without bankrupting the good one. Observers did the opposite, favouring the already-easy item, and their overall error exceeded the optimum as a result. In the feedback experiment, the error-minimizing strategy would have required shifting twice the resources towards the magnified, difficult item; instead observers favoured the minified one, producing feedback error significantly worse than optimal. The brain, it seems, was not solving the task it was given. It was chasing the feeling of getting things right.

To explain this seemingly irrational behaviour mechanistically, the authors augmented the neural resource model with a simple reinforcement learning rule. On each trial, three candidate reward signals are combined into a composite: the external points earned, an exponential transform of the feedback error, and an estimate of internal confidence derived from the model itself. Because spiking activity is generated by a Poisson process, the precision of the decoded estimate fluctuates from trial to trial, and the width of the likelihood function over stimulus values provides a natural, report-free measure of how confident the observer’s brain is on that particular trial. This composite reward updates a running value associated with each stimulus colour, with a leak parameter determining how quickly old value decays, and the accumulated value is mapped through an exponential function onto the neural gain that sets resource allocation on the next trial. Rewards, whether external or internal, thus become attached to the identifying features of objects, and those features inherit the priority of the rewards they have predicted.

The model’s vindication came from a comparison of two independent estimates. The freely fitted gain parameter, extracted from error distributions alone with no knowledge of rewards, correlated almost perfectly with the allocation trajectory computed purely from the history of accumulated rewards, with correlation coefficients approaching 0.98 in the external reward experiment and remaining above 0.83 in the confidence-driven experiments. Within individual observers, trials on which the model predicted below-median allocation to the probed stimulus produced reliably larger errors than trials with above-median allocation, across all four experiments. In other words, a learning rule that treats the private sensation of confidence as a reward, and nothing more, is sufficient to reconstruct how the brain divided its visual budget, trial by trial, without any explicit strategy or awareness.

The findings resonate with a growing literature suggesting that the striatum and its dopaminergic inputs, long celebrated for encoding prediction errors about money and juice, also encode prediction errors about confidence, and that perceptual learning can proceed without any external feedback at all, driven by internal confidence signals. They also connect to the economics of mental effort: tasks perceived as difficult are avoided and discounted, and the present results suggest that difficulty discounts the subjective value of a stimulus in much the same way, diverting resources towards whatever feels achievable. The authors are careful to note the limits of the work. Confidence was never measured directly but manipulated through validated proxies, and the tasks were demanding enough that even easy stimuli required genuine effort, leaving open whether extremely trivial stimuli might trigger the opposite, curiosity-driven reallocation towards harder items. Still, the broader implication is hard to escape. The allocation of perception is not a dispassionate optimization of accuracy but a market shaped by reward, and some of the most powerful currency in that market is manufactured entirely inside our own heads, rewarding us for being right and quietly reshaping what we are able to see.

Subject of Research: How intrinsic reward signals such as confidence guide the allocation of limited visual processing resources through reinforcement learning

Article Title: Intrinsic rewards guide visual resource allocation via reinforcement learning

Article References: Tomić, I., Raimundo, R., & Bays, P. M. (2026). Intrinsic rewards guide visual resource allocation via reinforcement learning. Nature Human Behaviour. https://doi.org/10.1038/s41562-026-02573-7

Image Credits: AI Generated

DOI: 10.1038/s41562-026-02573-7

Keywords: reinforcement learning, intrinsic reward, visual attention, working memory, confidence, neural resource allocation, population coding, perceptual decision-making, metacognition, reward learning, psychophysics, computational modelling

The research team, led by Sebastian Priest and colleagues, began from a familiar observation in medical education literature. Doctors entering their Foundation Year One posts, the first full year of practice after medical school in the United Kingdom, frequently report feeling under-prepared for the realities of being on call. Undergraduate curricula tend to emphasise structured ward rounds, supervised procedures and theoretical knowledge, whereas on-call work is defined by interruption, competing priorities, incomplete information and decisions made at speed with limited support. This gap between expectation and reality contributes to what educators call transition shock, a period of stress, fatigue and self-doubt that can affect not only the wellbeing of new doctors but also the safety of the patients they treat. Simulation has long been recognised as one of the most powerful tools for closing that gap, allowing trainees to make and correct mistakes in an environment where no patient is at risk.

The most established format for this kind of training is the in-situ bleep simulation. In these exercises, junior doctors carry dedicated pager devices while working through realistic on-call scenarios embedded within the actual clinical environment. Facilitators send bleeps, dispatch simulated nurses and relatives, and observe how candidates triage, prioritise and escalate. The technique has earned a strong reputation because it reproduces the sensory and cognitive texture of a real shift: the corridors, the phones, the competing demands on attention. Yet the same features that make it effective make it difficult to scale. Each scenario demands considerable facilitator input, often with several staff members required to run a single participant or small group through a session. Hospitals also report resistance from clinical teams who worry about disruption to live patient care, and the logistical complexity of staging multiple simultaneous simulations across a working hospital quickly becomes prohibitive. For an entire Foundation Year One cohort undergoing induction, these constraints are serious.

“Not the Bleep!” was designed specifically to address that scalability problem while preserving what makes on-call simulation valuable. The development process began in September 2023 with a formal learning needs assessment and a stakeholder consultation. Rather than inventing scenarios in an education department office, the team consulted the people who actually carry the bleep: junior doctors who had recently completed their own first on-call periods, along with senior clinical staff who supervise them. The tasks, dilemmas and frustrations that emerged from that consultation were then translated directly into the simulation’s scenario material. This design choice proved central to the programme’s later success, because facilitators evaluating the sessions specifically singled out the realism of the tasks, which participants recognised as drawn from the authentic texture of their future working lives rather than from textbook abstractions.

The resulting tabletop simulation was delivered at the annual Foundation Year One trust induction at University Hospitals Sussex, meaning that an entire incoming cohort of new doctors could be trained in a single coordinated event. In a tabletop format, participants work through on-call scenarios seated around tables rather than moving through hospital corridors, with facilitators delivering requests, clinical problems and competing demands in real time, and structured debriefing built into the flow of the exercise. Because the activity is station-based and table-based, multiple groups of candidates can be run simultaneously, dramatically increasing throughput compared with in-situ modalities. The approach also reduces the ratio of facilitators to participants, a critical consideration for NHS trusts facing severe workforce and financial pressures. Where a traditional bleep simulation might require a facilitator or more per small number of participants, the tabletop format allows a far larger cohort to be trained with the same staff, and crucially does so away from live clinical areas, eliminating the disruption that often generates resistance from ward teams.

The evaluation combined several complementary data streams. Eighty-six individuals took part in the tabletop simulation, and 61 candidates completed evaluative feedback. The primary outcome was a self-reported confidence score, collected on a scale out of ten both before and after the intervention. The improvement from 3.9 to 6.4 represents a substantial gain of two and a half points, achieved within the compressed timeframe of an induction programme. Alongside this, the team measured satisfaction with the induction process as a whole, both before and after the simulation was embedded within it, and recorded an increase from 3.83 to 6.08. The second finding carries practical significance for workforce planning, because induction quality is known to influence how welcomed and prepared new doctors feel, and by extension how quickly they settle into safe, effective practice and how likely they are to remain within the organisation.

Qualitative feedback added texture to the numbers. In facilitator debriefs conducted after the sessions, educators praised the strength of near-peer debriefing, in which more senior trainees, close in experience and status to the participants, guide post-scenario reflection. Near-peer teaching is well recognised in medical education because junior doctors often disclose uncertainty and ask questions more freely of someone only a year or two ahead of them than of a consultant, and the Sussex team found this dynamic particularly powerful in their format. The debriefing occurred in real time, immediately after each scenario, when the emotional and cognitive imprint of the decision-making was still fresh. Facilitators also highlighted the realism of the simulation’s tasks, a direct product of the earlier stakeholder consultation, noting that participants engaged readily with material that mirrored the bleeps, referrals and escalations they would genuinely face on their first night shifts.

The study navigated the ethical and practical requirements of educational research carefully. The Clinical Research Facility at the Royal Sussex County Hospital and the Research Standard Operating Procedures of University Hospitals Sussex determined that formal ethics approval was unnecessary, and the study was assigned project number 1931. All participant data were anonymised at the point of collection, informed consent was obtained from every participant, participants were told in advance how their data would be collected and analysed, and participation operated on an opt-in basis through voluntary completion of the data collection tool. The authors declared no competing interests, and publication costs were funded by My Charity University Hospitals Sussex, Registered Charity No. 1050864, with the customary caveat that the views expressed are those of the authors and not necessarily those of the funder. The article was received in May 2025, accepted in September 2026 and published open access on 22 September 2026, making the full methodology and supplementary materials freely available to other institutions.

The authors are explicit that the tabletop format should be understood as an alternative or adjunct to, rather than a replacement for, in-situ pager-based simulation. Each modality has distinct strengths: in-situ simulations offer unmatched environmental fidelity, while tabletop delivery offers throughput, low facilitator burden and freedom from clinical disruption, combined with immediate near-peer debriefing. The most important next step, the team argues, is direct comparative research between the two methodologies to establish whether the confidence gains observed here match or differ from those of in-situ training, and which format better transfers learning to actual on-call performance. They also identify expansion into interprofessional education as a promising frontier, running scenarios with nurses, pharmacists and other members of the multidisciplinary team alongside junior doctors, reflecting the genuinely collaborative nature of out-of-hours hospital care. For NHS trusts confronting large incoming cohorts, limited simulation staff and wards that cannot absorb repeated disruption, the message is strikingly simple: a conference room, a stack of realistic scenarios and a debrief from someone who has been there may be enough to make the first bleep feel less like a threat and more like a well-rehearsed beginning.

Subject of Research: Evaluation of a tabletop on-call simulation for training new Foundation Year One doctors

Article Title: “Not the bleep!”; a novel tabletop approach to on-call simulation

Article References: Priest, S., Clarke, B., Huszka, H., Tovell, N., Okorie, M., & Malik, S. (2026). “Not the bleep!”; a novel tabletop approach to on-call simulation. BMC Medical Education. https://doi.org/10.1186/s12909-026-10456-8

Image Credits: AI Generated

DOI: 10.1186/s12909-026-10456-8

Keywords: tabletop simulation, on-call simulation, medical education, Foundation Year One, new doctors, bleep simulation, near-peer debriefing, induction, confidence, transition to practice, NHS training, scalability

The distinction between correlation and causation sits at the heart of the research. Prior work has established that language models can report calibrated confidence estimates: when a model assigns a high probability to its own answer, that answer is more likely to be correct, and when the model expresses doubt, errors become more frequent. But such correlations leave open an ambiguity. A model might compute an internal confidence signal and then act on it, or the confidence estimate might simply be a by-product of the same computational processes that generate the answer, exerting no independent influence on behaviour. Disentangling the two requires more than observational measurement; it requires intervention.

That is precisely the approach taken in the new work. Rather than passively recording what models say about their confidence, the researchers manipulated the internal states associated with confidence and observed whether behaviour changed as a result. By intervening directly on the representations that carry confidence information within a model’s activations, the team could test whether those representations play a genuine causal role in driving the model’s subsequent outputs. The logic mirrors classic interventionist methods in neuroscience and psychology, where researchers establish that a neural signal causes behaviour by perturbing it and watching what happens, rather than merely noting that the two tend to co-occur.

The results, according to the study, show that when confidence-related internal states are amplified, models behave in ways characteristic of high-confidence systems: they commit more firmly to their answers, become less likely to hedge or express uncertainty, and shift their response patterns in measurable ways. When those same states are suppressed, the opposite occurs. Models become more tentative, more prone to qualify their statements, and more likely to distribute probability across multiple candidate answers. Crucially, these behavioural shifts occur without altering the model’s underlying knowledge or the input prompt, indicating that the confidence signal itself is doing causal work in the decision-making pipeline.

The implications extend well beyond academic curiosity. If a language model’s confidence genuinely drives its behaviour, then the way a model expresses certainty is not a cosmetic layer added on top of reasoning—it is an integral part of how the system decides what to say. This reframes long-standing concerns about overconfident AI. Models that state falsehoods with unwavering assurance may do so not because they lack a mechanism for detecting their own errors, but because the confidence signal that should temper their claims has been miscalibrated or overridden. Conversely, models that hedge excessively may be suffering from systematically deflated internal confidence, even when their answers are largely correct.

The findings also speak to a growing body of research on introspection in artificial systems. One of the most debated questions in AI safety is whether language models have meaningful access to their own internal computational states—whether, in some functional sense, they know what they know. Self-report alone is a weak guide, because models can be trained to produce plausible-sounding confidence statements without those statements reflecting anything real. The new causal evidence strengthens the case that something substantive underlies these reports: the confidence a model expresses is connected, through identifiable internal mechanisms, to the behaviour it subsequently produces. That connection is exactly what one would expect if the model is, in a limited but genuine sense, monitoring and acting on its own uncertainty.

Methodologically, the study represents a maturing of interpretability research. Early work in the field focused on locating where particular concepts are encoded in a network’s layers, producing maps of representation that were suggestive but often causally inert. The field has since shifted toward intervention-based techniques—activating, suppressing, or editing internal features and measuring the downstream consequences. The new research applies this interventionist toolkit to metacognition, the capacity of a system to represent its own cognitive states. Demonstrating that confidence representations pass causal intervention tests places them in the same evidentiary category as other well-established internal features, such as those implicated in factual recall and instruction following.

For practitioners deploying language models in high-stakes settings, the research carries practical weight. Calibration techniques, uncertainty quantification, and abstention mechanisms—methods that allow models to say when they do not know—are increasingly treated as essential safety infrastructure for medical, legal, and scientific applications. The new findings suggest these mechanisms are not merely bolted-on output filters but are entangled with the model’s core decision processes. Interventions that improve calibration at the level of internal confidence representations could therefore propagate through the entire behavioural stack, making the model more honest about its limits in a way that end-to-end fine-tuning alone may not reliably achieve. At the same time, the causal link cuts both ways: anything that distorts a model’s internal confidence—adversarial inputs, distribution shift, or aggressive post-training—could distort not just its stated confidence but its actual behaviour in hard-to-predict ways.

The study also raises questions that the authors and the broader field will now need to address. Confidence in language models is not a single scalar; it is distributed across layers, tokens, and computational pathways, and different components of a network may encode different, sometimes conflicting, estimates of reliability. Understanding which of these representations are causally potent—and which are epiphenomenal—will be essential for building systems whose self-assessments can be trusted. There are also open questions about generality: whether the causal role of confidence holds across model scales, architectures, and training regimes, and whether the same mechanisms govern other forms of self-monitoring, such as a model’s assessment of the safety or harmfulness of its own outputs.

What the research establishes, for now, is a conceptual milestone. Language models are not black boxes that merely emit text; they are systems whose internal states, including states that encode confidence, play demonstrable causal roles in shaping behaviour. That realization moves the field closer to a science of machine metacognition—one in which a model’s relationship to its own uncertainty can be measured, verified, and engineered rather than assumed. As language models are entrusted with ever more consequential decisions, knowing that they use their confidence to drive what they do is a crucial step toward ensuring that what they do remains worthy of the confidence we place in them.

Subject of Research: Causal role of internal confidence representations in driving the behaviour of large language models

Article Title: Causal evidence that language models use confidence to drive behaviour

Article References: Kumaran, D., Daw, N., Osindero, S., Veličković, P., & Patraucean, V. (2026). Causal evidence that language models use confidence to drive behaviour. Nature Machine Intelligence. https://doi.org/10.1038/s42256-026-01293-x

Image Credits: AI Generated

DOI: 10.1038/s42256-026-01293-x

Keywords: large language models, confidence, causality, interpretability, AI safety, metacognition, calibration, uncertainty, neural networks, machine learning, Nature Machine Intelligence, self-monitoring