Lisa Bardach

The rapid emergence of ChatGPT in education was accompanied by an equally rapid proliferation of meta-analyses synthesizing its effects on educational outcomes. We conducted a systematic epistemological audit of the first wave of these meta-analyses, evaluating both their methodological and reporting quality and the primary evidence they synthesized. We included meta-analyses published after the release of ChatGPT that quantitatively synthesized its effects on educational outcomes. Five meta-analyses comprising 87 unique primary studies met our inclusion criteria. We also reconstructed their quantitative evidence base and conducted exploratory three-level random-effects meta-analyses to examine effect magnitude, heterogeneity, and robustness, and how these varied with theoretically and methodologically relevant characteristics. The five meta-analyses generally searched multiple databases, specified eligibility criteria, and examined heterogeneity and potential publication bias, but showed limitations in independent screening and coding, preregistration, open data and code, and transparent reporting. The underlying primary evidence likewise showed methodological and reporting limitations and substantial conceptual heterogeneity. Among 69 classifiable primary studies, 79.7% assessed performance exclusively, whereas only 11.6% assessed learning exclusively. The reconstructed meta-analysis yielded a positive average effect (g = 0.83, 95% CI [0.68, 0.97]), but heterogeneity was substantial (I² = 94.06%; 95% prediction interval [−0.75, 2.41]), and estimates were markedly attenuated in sensitivity analyses addressing small-study and selection effects. Although positive effects remained among stronger studies assessing learning, increasingly stringent quality criteria rapidly reduced the evidence base, limiting the certainty and generalizability of conclusions about learning effects. Overall, our findings caution against interpreting a single pooled estimate as a stable general effect of ChatGPT on learning and highlight the need for more rigorous, conceptually precise, transparent, and reproducible research.

Educational inequalities—systematic differences in achievement between social groups—persist worldwide. Yet long-term evidence on how such inequalities unfold at intersections of multiple social group categories (e.g., gender, socioeconomic status, immigrant background) remains limited. Using multilevel intersectional models adapted from epidemiology, we analyzed two decades of assessment data in reading, mathematics, and science (over 537,000 students from Australia, Brazil, Finland, Germany, Hong Kong, Italy, Israel, and the United States). In most countries, additive effects of gender, socioeconomic status, and immigrant background accounted for most between-group structuring in intersectional inequality. Intersection-specific components were mostly small but persistent. Large country-specific gaps separated the highest- and lowest-performing intersectional groups, with the largest gaps exceeding two standard deviation units. We highlighted intersections with pronounced (dis)advantages, underscoring the need to address compounded disparities. Overall, these findings clarify the structure of intersectional educational inequality across countries and over time.

Academic Children and adolescents are growing up in increasingly ethnically, racially, and culturally diverse and often unequal worlds. Therefore, how schools, as key developmental contexts, approach diversity has important implications at individual, group, and societal levels. Here, we present a review, integration, and expansion of research on cultural diversity climate in schools, focusing on five dimensions of diversity climate (the cultural diversity climate “Big Five”) and their facets. We begin by distinguishing cultural diversity climate from other related terms (e.g., diversity ideologies, diversity policies). Then, we discuss the five core dimensions of diversity climate (intergroup contact theory’s optimal contact conditions, multiculturalism climate, colorblind climate, critical consciousness climate, polyculturalism climate). We examine facets subsumed under the dimensions, propose new theory-based facets, and outline differences and commonalities between them. Next, we review measurement approaches and highlight future research directions. Overall, this integrative, multidimensional, and multifaceted framework advances psychological theory on diversity climate. Public Schools play a vital role in how children and adolescents experience and understand diversity. This article reviews the concept of cultural diversity climate in schools, focusing on five key aspects that shape how cultural diversity is approached (i.e., the cultural diversity climate “Big Five”). Our integration of the cultural diversity climate “Big Five” encompasses five main dimensions, such as multiculturalism climate (acknowledging and valuing diversity at school), critical consciousness climate (critically reflecting on and challenging racism and inequities), and polyculturalism climate (raising students’ awareness of connections between groups and cultures). To move towards a more fine-grained understanding of diversity climate, we describe existing facets (i.e., sub-components of the different dimensions) and develop new facets based on theory. We also discuss different ways of measuring cultural diversity climate. This work is important for creating more inclusive, equitable, and critically conscious school environments, thereby supporting all students, especially those from marginalized communities.

Background: Digital games are promising tools for supporting knowledge acquisition, situational interest, and academic emotions in digital learning environments. However, evidence remains mixed regarding whether video games lead to stronger learning outcomes than text-based formats, and little is known about how instructional framing shapes learners’ engagement with such environments.Objectives: This study investigated the effects of content format and instructional framing on factual knowledge, situational interest, and academic emotions. It compared a historically accurate video game (Limes) with a structurally aligned interactive text application; each paired with instructions to focus on either learning or fun.Methods: In a 2 (format: game versus text) × 2 (instruction: learning versus fun) between-subjects design, 143 university students engaged with either the video game or the interactive text application. Both formats presented the same content through main narratives, optional missions, and supplemental information in the form of Codex entries. Outcomes included knowledge acquisition, situational interest, enjoyment, and boredom. Additional analyses examined behavioural data on Codex engagement.Results and Conclusions: Participants in the game condition reported higher enjoyment and greater triggered situational interest, whereas knowledge scores were higher in the text condition. No statistically significant effects emerged for instructional framing or for the interaction between format and framing; however, text participants engaged more extensively with Codex entries, and Codex engagement strongly predicted knowledge outcomes across both formats. These findings suggest that games can enhance positive affective experiences, but learning outcomes may depend less on the format itself than on how supplemental learning materials are integrated.

Artificial intelligence (AI) is becoming increasingly integrated into education. However, fundamental questions about its effects remain unresolved, with existing meta-analyses differing substantially in reported effect sizes, moderators that amplify or attenuate these effects, and methodological quality. To synthesize the fragmented evidence base and examine sources of these variations, we conducted a preregistered second-order meta-analysis aggregating 45 meta-analyses (129 effect sizes) based on approximately 197,000 learners from prekindergarten through university. Three-level second-order meta-analyses showed that AI applications in education had a medium effect across educational outcomes. Effects were larger in more recent meta-analyses, in meta-analyses including tertiary-education students, and for general academic outcomes than for STEM performance outcomes. We did not find significant differences between AI applications. Methodological quality of meta-analyses did not explain variance in effects, but our assessment identified systematic quality deficits (e.g., regarding reproducibility and open science). Insufficient statistical power was common among the included meta-analyses. Overall, these findings advance a more differentiated understanding of the current state of research on AI in education, highlighting both the potential of AI and the limitations of the evidence base.

In this conceptual paper, we argue that motivation science provides a valuable yet underutilized framework for understanding and designing digital and AI-supported learning environments. To illustrate this, we first review studies included in a recent meta-analysis on generative AI and motivation to examine the extent to which motivation science informs current research. We show that the vast majority of studies do not use motivational theories to explain how such systems may support student motivation. Next, drawing primarily on achievement goal theory and the TARGET (task, autonomy, grouping, recognition, time) framework, we outline key motivational principles for structuring learning environments and discuss how these principles may be enacted in digital and AI-supported contexts. We also examine risks and tensions associated with these environments, acknowledging that digital and AI systems may create novel motivational challenges not fully addressed by existing motivational theories developed in more traditional educational settings. Overall, the paper aims to stimulate further research on rethinking digital and AI-supported learning environments through the lens of motivation science while also reconsidering motivation science in light of these emerging environments.

Curiosity manifests in everyday information-seeking; however, most empirical research on curiosity relies on laboratory paradigms, limiting insight into how curiosity operates in naturalistic settings. Here, we study curiosity ‘in the wild’ using large-scale data from the online question-and-answer platform StackExchange (N = 60,154 questions). Drawing on curiosity theories, we identified theoretically motivated characteristics of questions, translated into behavioral measures, that were hypothesized to predict curiosity (conceptualized as the number of views a question received). For Science Fiction questions, question concreteness and familiarity contributed to higher curiosity. Semantic distance differentiated domains: less utilitarian topics (e.g., Science Fiction, history) favored semantic coherence in terms of proximity, whereas utilitarian topics (e.g., programming, web development) favored greater distance. Across topics, curiosity peaked at moderate uncertainty and increased with question importance (upvotes), suggesting domain-general patterns. Negative sentiment predicted higher curiosity only in non-utilitarian topics. Overall, naturalistic information seeking reveals organizing principles that extend existing curiosity theories.