Using generative AI in science necessitates that institutions and researchers practice epistemic reflexivity to protect intellectual diversity and resist the threat of scientific monoculture. We argue that researchers and institutions can position generative AI both as a tool for epistemic reflexivity and as an object that must be examined reflexively.
Abstract A primary objective of intensive longitudinal studies is to investigate within-person dynamics. In this context, item heterogeneity plays a critical role, as within-person processes may vary across items within a scale. A common example is the assessment of momentary affect using adjective lists (e.g., sad, angry, anxious, stressed), where each item captures different facets of positive or negative affect, providing unique and non-interchangeable information. However, standard practices often overlook item heterogeneity by aggregating item scores or assuming a single within-person factor in dynamic structural equation models. This simplification does not permit a fine-grained analysis of within-person dynamics and compromises cross-study comparability when item pools differ across studies. In this article, we reanalyze five large-scale intensive longitudinal datasets assessing momentary affect to illustrate how item heterogeneity can be explicitly modeled. We introduce a flexible modeling approach that accommodates item-specific and person-specific dynamics while improving psychometric comparability across studies, based on residual dynamic structural equation modeling with reference items. We compare this method to conventional modeling strategies and provide practical guidance for addressing item heterogeneity in the analysis of intensive longitudinal data.
Everyday life requires that we navigate dozens of interactions with strangers quickly and effectively. Here we hypothesized that representations of roles (e.g., cashier, mechanic, doctor) enable people to build rapid expectations about what others will do, what they know, and who else serves the same function. To test this, we used a self-paced reading paradigm in which we timed how long it took participants to read short passages about social interactions. Across three studies (N=300 participants per study), we show that role representations support real-time expectations about how other people might act (Study 1), the knowledge they might possess (Study 2), and whether it is appropriate to generalize across agents (Study 3). Moreover, people reported more surprise when the events deviated from role expectations and were more likely to misreport what happened in a way that conformed to role expectations. Our results suggest that roles are a powerful route for social understanding that has been previously understudied in social cognition. Using a self-paced reading paradigm, we show that, from just the mention of a role, people build rapid expectations about how other people will act, what they know, and whether they can be interchangeable with others in the same role.
Most adults have limited understanding of child development, yet their perceptions of development influence their decisions about whether to support children's autonomy-such as support for autonomy-limiting anti-transgender legislation. This work found that adult participants believe gender develops later for transgender youth than cisgender youth, predicting lower support for children's autonomy to determine their gender identity, and greater support for anti-transgender legislation (Study 1; N = 200). However, learning about similarities in the development of transgender children and cisgender children decreased this tendency. This predicted greater support for children's autonomy in determining their gender identity, which was associated with reduced support for anti-transgender legislation in a large, nationally representative sample (Study 2; N = 1552), and increased donations to an organization supporting the rights of transgender youth (Study 3; N = 174). Results highlight the importance of adults' beliefs about gender development in transgender youth to understand their support for the autonomy and rights of transgender youth.
Adversities experienced during early childhood can have powerful and long-lasting influences that may affect the development of cognitive, social, and emotional skills. Game-based interventions, played by groups of peers and focused on optimizing social cognition, may address these effects and improve resilience, positively impacting academic achievement and social well-being. However, empirical evidence for the effectiveness of such interventions remains limited, particularly in low- and middle-income countries where their impact may be especially beneficial. In this study, we developed and preliminarily evaluated a game-based curriculum designed to enhance children’s understanding of social and emotional information, critical skills for effective communication, metacognition, and learning from others. In a preregistered, double-blind cluster-randomized pilot study conducted in Brazil, we evaluated the effectiveness of the social game-based Intervention, CuriousMind, in the context of urban poverty among preschool children. Schools were randomized to one of three conditions: CuriousMind, a game-based math intervention (active control), or a no-treatment control. Children were evaluated before and after the 10-week intervention. CuriousMind significantly enhanced children’s social and emotional skills, particularly in emotion labeling, emotion recognition, and emotion situation knowledge, providing evidence that game-based interventions with peers can effectively promote social-cognitive development in early childhood. In a preregistered, doubleblind randomized pilot study in Brazil, a game-based social-cognitive intervention improved preschool children’s emotion labeling, recognition, and emotion situation knowledge, supporting peer games for early social-emotional development.
Establishing common ground between individuals with divergent opinions appears more difficult than ever. How can people with opposing points of view establish common ground when preexisting shared experiences, traits, or values are not easily uncovered? Drawing from an online sample of U.S. adults (N = 558), this preregistered randomized controlled experiment tested social perspective taking (SPT) as a promising approach for creating common ground. Paralleling many digital contexts, we focused on a hot-button issue—educational policy related to climate change—in a context where similarities and shared values were unlikely to be readily discovered. Specifically, in two treatment conditions, a fictional teacher with an opposing point of view took the perspective of participants by identifying, acknowledging, and affirming the merit of either their (a) arguments or (b) intentions. These interventions created a sense of common ground and improved multiple relationship-related outcomes as evidenced through higher levels of perceived perspective-taking, perceived similarity, perceived fairness, and anticipation of a positive relationship with the fictional teacher. Estimated effect sizes further suggested that affirming the other party’s arguments was about twice as effective as affirming their intentions. In the present historical moment when society faces multiple challenges in enabling productive conversations, our study uses a social perspective-taking intervention to provide a recipe for creating common ground between people who may not otherwise discover any. This preregistered RCT (N = 558 U.S. adults) tested perspective-taking on a divisive environmental education topic. Affirming either arguments or intentions of the opposing side built common ground and improved relations; the former twice as effective.
Everyday use of AI has introduced more AI-generated content into surveys, diluting insights and distorting findings. While the concern about agentic “bots” responding to surveys is now widely recognized by most researchers, this comment discusses the simultaneous rise in use of AI by genuine human survey respondents and provides guidance on identification and implementation of the necessary safeguards that ensure surveys reflect public opinion rather than recycled or programmed content. Surveys are compromised not only by bots, but also by the use of AI by genuine human survey respondents. How can researchers ensure that surveys reflect public opinion rather than recycled or programmed content?
Recent advancements in natural language processing and large language models (LLMs) facilitate the study of under-researched areas in mental health. Their capacity to automatically and meaningfully analyze large-scale text data makes them particularly valuable for studying highly individualized phenomena with clinical relevance, such as triggers of obsessive-compulsive symptoms (OCS), where pattern identification is often challenging. To address this gap, we surveyed 1495 individuals from the general population about contamination-related obsessive-compulsive symptoms (C-OCS), as well as their triggers and corresponding intensity. Using LLM-based embeddings, we generated a map of key trigger categories for C-OCS, revealing their diversity across ecological domains and varying degrees of semantic similarity. Monte Carlo simulations further showed that individuals frequently reported semantically similar trigger pairs that differed in intensity. These findings provide a basis for further investigations into associative learning processes at the categorical and semantic levels, which may enhance understanding of mechanisms involved in the development, maintenance and treatment of obsessive-compulsive disorders. Using LLMs, this study mapped contamination-related obsessive compulsive symptom triggers in 1’495 individuals. Distinct trigger categories across ecological domains and varying intensity were identified, providing a foundation for future translational research.
Humans prioritize learning about kinds (e.g., dogs, in general), over individuals (e.g., a specific dog), as kind-based facts provide rich information about categories that support learning. However, to build a comprehensive mental model of the world, we must also learn about individuals and their unique properties. What factors lead us to seek information about individuals over kinds? Here, we examined how social information, “special” status, and future utility shape children’s motivation to learn about individuals. Across two preregistered experiments (N = 212 5-to-9-year-olds tested in the United States and France), we found that marking an item as special (Experiment 1) or useful (Experiment 2) reliably shifted children’s baseline preference for learning about kinds to learning about that specific individual. Utility, but not specialness, significantly predicted subsequent recall, suggesting that curiosity alone does not always lead to enhanced memory. Age-related differences emerged, such that older children showed a stronger preference to learn about kinds for regular items. In Experiment 1, preference to learn about individual special items was higher with older age, while in Experiment 2, there was no evidence for age-related differences. Together, these findings demonstrate that from age 5 and across two distinct cultural contexts, children selectively prioritize what to learn: they default to kinds but shift to individual-focused learning when items are marked as special or likely to be useful. This flexible selectivity guides both what children learn and what they remember across development. How do children build comprehensive models of the world? Across two experiments with 5- to 9-year-olds in the United States and France, marking items as special or useful shifted children’s curiosity from wanting to learn about kinds to individuals, with select consequences for learning.
Cravings contribute to eating behaviours and health, yet how they emerge and translate into consumption remains unclear. Individual biases in how we learn to assign value to cues and actions, through Pavlovian conditioning (sign-tracking and goal-tracking) and reinforcement learning (model-free and model-based), provide a framework for exploring these behaviours. Cravings, emerging from associations between environmental cues and physiological responses, may be amplified by learning biases enhancing the cues’ salience (i.e., sign-tracking bias) and habitual behaviour (i.e., model-free). Learning biases may also moderate the relationship between cravings and consumption by intensifying automatic and habitual processes. Lastly, approach bias, which measures automatic processes involved in craving formation, may be associated with learning biases. To test these hypotheses, we conducted a two-week longitudinal study in 81 participants (minimum post-exclusion sample of 65) using ecological momentary assessments via a smartphone-based app to measure high-caloric food cravings, food intake, and approach-avoidance tendencies in daily life. Learning biases were assessed using Pavlovian conditioning and sequential learning tasks. Results indicate that sign-tracking bias was not related to craving intensity or variability, whereas model-free bias was associated with higher mean cravings but not lower day-to-day craving variability. The hypothesis linking cravings to consumption was not tested due to a consumption floor effect, and neither learning bias was associated with approach bias. The healthy-weight sample may explain our null findings, as they may show limited expression of high-caloric food cravings and consumption. Overall, these findings suggest that reinforcement learning tendencies may contribute to everyday craving dynamics. How individuals learn to assign value to cues and actions may help explain how eating behaviours emerge. This study investigated whether reinforcement learning biases are associated with food cravings, consumption, and approach tendencies in daily life.
Reward responsiveness plays an important role in promoting health and well-being in youth. Understanding associations between early experiences and variation in the typical developmental trajectory of reward responsiveness is essential to identifying potential pathways to health. We examined developmental trajectories of the reward positivity, a neurophysiological indicator of reward responsiveness, across 4 assessments in late childhood through late adolescence in 465 youth. We estimated growth curves and assessed parenting, observed via parent–child interactions, and stress exposure at age 3, as early environmental predictors of reward positivity trajectories. The reward positivity to win and loss feedback showed a quadratic trajectory of change across childhood and adolescence. Lower positive parenting and higher negative parenting predicted a reduced increase in the reward positivity to win feedback over time; stress exposure associations were smaller and statistically non-significant. Together, results suggest a link between early parenting and reward development across childhood through adolescence. Reward responsiveness plays an important role in promoting health in youth. We examined neural reward response from childhood through adolescence. Lower positive parenting and higher negative parenting predicted a reduced reward response over time.
Abstract Current efforts to understand Large Language Models (LLMs) are largely metaphorical. Researchers map LLMs onto familiar domains, from physics and neuroscience to psychology and sociology, each illuminating specific facets while obscuring others. We chart these metaphors across mechanistic, behavioral, and interactive scales and delineate their explanatory boundaries. Crucially, this metaphorical projection creates a recursive loop of anthropomorphism, fueling the “genuine understanding” versus “pattern matching” impasse. As an alternative approach, we propose machine experientialism, positing that LLMs build their own form of understanding from training corpora. The priority shifts from cataloging LLMs’ human-like traits to uncovering their distinct logic that emerges from this text-based world.
In context-dependent memory research, focal objects are by default ascribed to item whereas background scenes are considered the context. Questioning this assumption, it was proposed that context is reconstructed rather than encoded: any aspect of an event could become either item or context depending on the question asked. Here we provide evidence supporting this hypothesis across six experiments employing a context reinstatement paradigm. When the memory test focused on background scenes (Experiments 1-2), reinstating the original focal objects increased old responses to targets and lures. This effect remained when background scenes were task-irrelevant and focal objects task-relevant at encoding (Experiment 3). Finally, Experiments 4a-5 showed that, when objects are task-irrelevant at encoding, this effect exists but depends on the size of the objects and scenes on screen. Overall, these results support that context is reconstructed rather than encoded, although the perceptual properties of the stimuli and their relative salience at encoding can limit the effect of context on memory retrieval.
Human relationships form a complex web of affiliative and antagonistic ties. Yet how the brain represents their affective structures remains unclear. Here, using a television drama depicting intertwined friendships and rivalries, we examined how the brain encodes the valence of interpersonal relationships. Participants underwent fMRI scanning before and after watching the drama while viewing the faces of its central characters. They then rated each character pair for relationship strength and valence, and whole-brain representational similarity analysis (RSA) identified brain regions representing these relational structures. Relationship valence effects were most prominent for antagonistic (negative) relationships in the left anterior supramarginal gyrus and right medial prefrontal cortex. Univariate analyses revealed increased activation in the precuneus after drama viewing, suggesting enhanced retrieval of narrative-related person knowledge, though this region did not show statistically significant representational similarity patterns reflecting interpersonal relationships. These findings indicate that the human brain constructs a multidimensional social map from narrative experience, with antagonistic ties playing important role in shaping social relationship representations.
Collective past and future representations can be thematically linked and vary across sociopolitical identity. Here, we introduce the notion of collective mental time travel pathways: distinct connections that emerge between past and future representational networks upon reflection on a critical public event. We demonstrate that these pathways capture nuanced sociopolitical profiles shaping collective mental time travel. In Study 1 (N = 403), a representative sample remembered the 2023 Turkish Presidential Elections, and equally important past and future events. Participants rated each event's characteristics (Valence, Vividness, Agency, Importance). Using these ratings, we identified four unique representational pathways in which the elections bridged collective past and future. Each collective mental time travel pathway had distinct sociopolitical profiles: Despite losing, vividly remembering the election as very important, characterized by high national agency, and continuing post-election conversations were related to high future individual agency; remembering a positive and agentic past was related to high future political group agency. When participants remembered collective past and future events without reflecting on the election (Study 2, N = 201), event characteristics were separated by importance and agency. These two studies illustrate how collective mental time travel pathways can be activated by reviewing critical and consequential elections.
Can we detect students' hidden beliefs about fitting in at college in ways that predict academic success and inequality? Using computational language‑based measures, we analyzed 25,000 pre‑college essays from students at 23 U.S. colleges to infer two beliefs about belonging. One belief-simple optimism-reflects confidence in a smooth college transition without acknowledging potential challenges. The other-a process‑oriented perspective-acknowledges potential struggles but represents them as common and temporary. Students from socially disadvantaged backgrounds more often expressed simple optimism and less often the process‑oriented view. These patterns mattered: simple optimism predicted lower grades and rates of full‑time enrollment through the first year of college, whereas a process‑oriented view predicted higher GPA and persistence, controlling for academic preparation, among other factors. A short social‑belonging exercise shifted students' language toward the process‑oriented perspective by 25 percentage points, reducing differences by disadvantaged status. These findings suggest that latent beliefs about belonging are evident in everyday language and predict student outcomes, offering a tool for education leaders to better understand their students and support college success.
A central challenge in understanding joint action is explaining how individuals achieve successful coordination in dynamic, real-world settings. Although coordination is thought to depend on anticipating future events, including the actions of others, the precise contribution of such predictive processing remains unclear, since most existing evidence comes from simplified laboratory tasks that infer anticipation indirectly from reaction times. To address this gap, we studied 70 participants (forming 64 pairs) during solo and joint sessions of a turn-taking ball-hitting task while recording multimodal behavioral, physiological, demographic, and social measures. We found that successful coordination was most strongly associated with anticipation of one's own action outcomes, outperforming physiological synchrony, motor behavior, demographic characteristics, and social closeness. Partners whose predictions of their own action outcomes were more closely aligned coordinated better, likely because each behaved in ways the other implicitly expected. A Bayesian generative model further showed that well-coordinated pairs relied more heavily on prior expectations than on incoming sensory evidence when generating predictions during joint action. These results suggest that efficient coordination in dynamic real-world tasks emerges primarily when individuals' internal models for predicting their own actions are well aligned across partners. Our quantitative multimodal approach provides a framework for disentangling the contributions of predictive, physiological, motor, and social factors to coordination in naturalistic interactions.
Swiping-based dating apps have become a pervasive feature of contemporary social life, reshaping how individuals seek intimacy, curate self-presentation, and encounter psychological feedback. This systematic review and meta-analysis compared dating app users vs non-users across 27 studies (N = 21,263) to assess associations with mental health outcomes. Across six theoretically derived domains, meta-analytic results indicate small-to-moderate associations between dating app use and emotional distress, appearance concerns, body image disturbance, behavioral dysregulation, and interpersonal sensitivity, with the strongest effects observed for behavioral dysregulations (g = 0.44) and body-related outcomes (g = 0.32). Effects for general wellbeing were small and non-significant. Women and men who belong to a sexual minority exhibited elevated appearance- and body-related vulnerability. Although mechanisms cannot be inferred from available evidence, converging patterns across studies suggest that visually driven, evaluative interaction features may compound appearance-based concerns, and that high-volume partner choice may correlate with compulsive or dysregulated patterns of use. Considerable heterogeneity across studies underscores the influence of individual susceptibility and social context. Overall, findings indicate significant associations between dating app use and adverse psychological outcomes, while highlighting substantial gaps in longitudinal and mechanistic evidence. Future research should employ prospective and intersectional designs to clarify temporal pathways and inform digital mental health interventions.
People often wish they could reverse previous decisions. However, few sequential decision-making tasks used in the literature allow participants to undo their decisions. As a result, it remains an open question when, how, and why people reverse their decisions. We designed a sequential decision-making task in which the participant connected cities on a map using a limited road budget, either without or with the option to undo their decisions. We found that when decisions were reversible, participants spent less time on their first action, progressively improved on their solutions through decision reversals, and ultimately achieved better final performance. Furthermore, decision reversals predominantly corrected errors, and did so with greater precision when errors were larger. Finally, we found evidence that decision reversals were predicted by factors that may be associated with a subjective prior probability of an error. Taken together, our results are consistent with a view in which uncertainty about being on the right track - both retrospective and prospective - drives decision reversals. Our work opens the door to neurocognitive, computational and translational studies of decision reversals.
In this study, two new acquaintances joined an online chatroom, spent five minutes discussing the qualities of a job candidate, then a newcomer joined the chat. The three group members (Ntriads = 370; Nparticipants = 1110) engaged in the Hidden Profile Task during which they chose the best job candidate among four to hire. Before they began, all group members received common information and one group member-either one of the initial members or the newcomer-received critical, unshared information about one candidate that the team needed to make the correct decision. Teams in which the newcomer held critical information were less likely to make the correct decision than teams in which one of the initial members held it. Newcomers with critical information showed weaker linguistic coordination to group members-a measure of social alignment that captures the degree to which group members mimic each other's language style-and reported greater perceived task conflict than newcomers who did not have critical information. Lastly, we tested whether a manipulation of prestige-based status-giving false feedback on a test for skills ostensibly relevant to decision making-could overpower newcomer status; however, we did not find credible evidence that it did so. These findings illustrate the power of brief interactions for shaping biases against newcomers.
