Pub Date : 2026-01-14DOI: 10.1016/j.compedu.2025.105553
Jonas De Bruyne , Charlotte Larmuseau , Lieven De Marez , Durk Talsma , Klaas Bombeke
‘Learning by doing’, or experiential learning, is increasingly implemented through immersive media such as virtual reality (VR) across domains like education and professional training. Immersive technologies enable dynamic instruction and guidance, but this potential remains underexplored. To support learning, cognitive load theory promotes signaling to reduce cognitive load by guiding attention to essential content, while discovery learning encourages minimal guidance to foster exploration. While the temporal aspect of the signaling principle is underrepresented in literature, this study suggests that striking a balance between the theoretical approaches is possible by delaying additional guidance. This work therefore investigates the impact of delayed signaling on experiential learning in VR, using a VR training module on electrofusion welding that is currently used in industry. When comparing performance after training either with immediate or delayed signaling, the data suggested improved procedural learning when signaling was delayed, with an average improvement of 8% in task completion time ( .05, d = .76). Furthermore, the method with delayed signaling did not increase cognitive load, as measured by self-reports, suggesting that discovery learning in combination with (delayed) guidance does not place undue cognitive demand on participants. The findings stress the – currently underexposed – importance of timing of visual aids through signaling and how they can be used to optimize training effectiveness. The results are interpreted in light of existing learning literature with future directions for adaptive training systems highlighted.
“边做边学”或体验式学习越来越多地通过虚拟现实（VR）等沉浸式媒体在教育和专业培训等领域实现。沉浸式技术可以实现动态教学和指导，但这种潜力仍未得到充分开发。为了支持学习，认知负荷理论提倡通过引导注意力到基本内容来减少认知负荷，而发现学习则鼓励最少的指导来促进探索。虽然信号传导原理的时间方面在文献中代表性不足，但本研究表明，通过延迟额外的指导，可以在理论方法之间取得平衡。因此，本研究利用目前在工业中使用的电熔焊接VR培训模块，研究延迟信号对VR体验学习的影响。当比较即时或延迟信号训练后的表现时，数据表明延迟信号时程序学习得到改善，任务完成时间平均提高8% （p < 0.05, d = 0.76）。此外，根据自我报告的测量，延迟信号的方法并没有增加认知负荷，这表明发现学习与（延迟）指导相结合不会对参与者产生不适当的认知需求。研究结果强调了——目前尚未充分暴露的——通过信号提供视觉辅助的时机的重要性，以及如何利用它们来优化训练效果。根据现有的学习文献对结果进行了解释，并强调了自适应训练系统的未来方向。
{"title":"Timing matters! Using delayed signaling to improve experiential learning in procedural VR training","authors":"Jonas De Bruyne , Charlotte Larmuseau , Lieven De Marez , Durk Talsma , Klaas Bombeke","doi":"10.1016/j.compedu.2025.105553","DOIUrl":"10.1016/j.compedu.2025.105553","url":null,"abstract":"<div><div>‘Learning by doing’, or experiential learning, is increasingly implemented through immersive media such as virtual reality (VR) across domains like education and professional training. Immersive technologies enable dynamic instruction and guidance, but this potential remains underexplored. To support learning, cognitive load theory promotes signaling to reduce cognitive load by guiding attention to essential content, while discovery learning encourages minimal guidance to foster exploration. While the temporal aspect of the signaling principle is underrepresented in literature, this study suggests that striking a balance between the theoretical approaches is possible by delaying additional guidance. This work therefore investigates the impact of delayed signaling on experiential learning in VR, using a VR training module on electrofusion welding that is currently used in industry. When comparing performance after training either with immediate or delayed signaling, the data suggested improved procedural learning when signaling was delayed, with an average improvement of 8% in task completion time (<span><math><mi>p</mi></math></span> <span><math><mo><</mo></math></span> .05, <em>d</em> = .76). Furthermore, the method with delayed signaling did not increase cognitive load, as measured by self-reports, suggesting that discovery learning in combination with (delayed) guidance does not place undue cognitive demand on participants. The findings stress the – currently underexposed – importance of timing of visual aids through signaling and how they can be used to optimize training effectiveness. The results are interpreted in light of existing learning literature with future directions for adaptive training systems highlighted.</div></div>","PeriodicalId":10568,"journal":{"name":"Computers & Education","volume":"246 ","pages":"Article 105553"},"PeriodicalIF":10.5,"publicationDate":"2026-01-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145975569","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"教育学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-14DOI: 10.1016/j.compedu.2026.105575
Joseph G. Guerriero, Pejman Sajjadi, Janet K. Swim, Alexander Klippel, Jamie DeCoster, Mahda M. Bagher
{"title":"Virtual reality serious games for promoting environmental systems thinking and pro-environmental policy support","authors":"Joseph G. Guerriero, Pejman Sajjadi, Janet K. Swim, Alexander Klippel, Jamie DeCoster, Mahda M. Bagher","doi":"10.1016/j.compedu.2026.105575","DOIUrl":"https://doi.org/10.1016/j.compedu.2026.105575","url":null,"abstract":"","PeriodicalId":10568,"journal":{"name":"Computers & Education","volume":"32 1","pages":""},"PeriodicalIF":12.0,"publicationDate":"2026-01-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145962598","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"教育学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-13DOI: 10.1016/j.compedu.2026.105572
Yimin Ning , Wee Tiong Seah , Jihe Chen , Jinhai Liu , Peifen Tan
Feedback is crucial for teacher professional development, yet most studies focus on a single source (e.g., experts or peers). Longitudinal comparisons across sources and analyses of underlying mechanisms are rare. This study therefore investigates how different feedback sources influence teacher learning outcomes. A total of 522 teachers participated in a 13-week professional learning program, including three weeks of reflective practice. Participants were divided into three groups: NF (no external feedback), EF (expert feedback), and AF (AI feedback). Teaching videos, instructional design artifacts, and reflective journals were analyzed to assess outcomes. Teachers in the EF group achieved significantly higher learning outcomes than those in the AF group, who, in turn, outperformed the NF group. These findings underscore the distinctive value of expert feedback in providing depth of insight and contextual sensitivity, while also indicating that AI feedback, although timely and adaptive, cannot fully replace expert judgment. Cross-Lagged Panel Network (CLPN) analysis identified key behaviors and sequences across groups and revealed detailed temporal patterns within behavioral clusters. Qualitative interviews further demonstrated that feedback effectiveness is shaped by the interaction of three dimensions—Time, Object, and Level (TOL)—which form the basis of a framework with 16 elements. The study highlights the complementary strengths of AI adaptability and expert insight, suggesting that multi-source feedback enhances teacher professional development by integrating process support with higher-order regulation.
{"title":"A comparative study of expert, AI, and no external feedback on mathematics teacher learning outcomes in reflective practice","authors":"Yimin Ning , Wee Tiong Seah , Jihe Chen , Jinhai Liu , Peifen Tan","doi":"10.1016/j.compedu.2026.105572","DOIUrl":"10.1016/j.compedu.2026.105572","url":null,"abstract":"<div><div>Feedback is crucial for teacher professional development, yet most studies focus on a single source (e.g., experts or peers). Longitudinal comparisons across sources and analyses of underlying mechanisms are rare. This study therefore investigates how different feedback sources influence teacher learning outcomes. A total of 522 teachers participated in a 13-week professional learning program, including three weeks of reflective practice. Participants were divided into three groups: NF (no external feedback), EF (expert feedback), and AF (AI feedback). Teaching videos, instructional design artifacts, and reflective journals were analyzed to assess outcomes. Teachers in the EF group achieved significantly higher learning outcomes than those in the AF group, who, in turn, outperformed the NF group. These findings underscore the distinctive value of expert feedback in providing depth of insight and contextual sensitivity, while also indicating that AI feedback, although timely and adaptive, cannot fully replace expert judgment. Cross-Lagged Panel Network (CLPN) analysis identified key behaviors and sequences across groups and revealed detailed temporal patterns within behavioral clusters. Qualitative interviews further demonstrated that feedback effectiveness is shaped by the interaction of three dimensions—Time, Object, and Level (TOL)—which form the basis of a framework with 16 elements. The study highlights the complementary strengths of AI adaptability and expert insight, suggesting that multi-source feedback enhances teacher professional development by integrating process support with higher-order regulation.</div></div>","PeriodicalId":10568,"journal":{"name":"Computers & Education","volume":"246 ","pages":"Article 105572"},"PeriodicalIF":10.5,"publicationDate":"2026-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145962558","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"教育学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
As artificial intelligence (AI) and generative AI (GenAI) technologies become increasingly integrated into everyday life, the need for validated tools that measure people’s knowledge about AI grows. Here, we present the development and validation of a theoretically driven, performance-based scale for assessing AI and GenAI knowledge. The scale is grounded in a two-axial framework. One axis captures three knowledge types: content knowledge (what AI is and where it is encountered), procedural knowledge (how AI systems operate and are used), and epistemic knowledge (what features and construction processes characterize AI outputs). The other axis encompasses three knowledge domains: technology-related knowledge (AI systems), user-related knowledge (users’ interaction with AI), and society-related knowledge (the social and ethical implications of AI). Based on an online survey of 800 internet-using adults from [Country], the 26-item scale was evaluated using confirmatory factor analysis, which demonstrated an acceptable model fit. It was further validated through two-stage structural equation modeling and group comparisons. Overall, the scale was found to be both valid and practically insightful: while it reproduces the expected relationships with additional constructs (e.g., trust in GenAI, attitudes toward AI) and expected differences between demographic groups, it also provides nuanced insights on the intricacies of AI knowledge. For example, the scale indicates that the relationship between trust in GenAI and knowledge about AI is grounded in both epistemic and societal knowledge. Thus, this novel tool affords more precise investigations into how different types and domains of AI knowledge relate to perceptions, behaviors, and decision-making in an AI-mediated world.
{"title":"Measuring different types and domains of AI knowledge: Developing and Validating a Performance-Based Scale","authors":"Inbal Klein-Avraham, Ruth Savir, Osnat Atias, Ido Roll, Ayelet Baram-Tsabari","doi":"10.1016/j.compedu.2026.105573","DOIUrl":"https://doi.org/10.1016/j.compedu.2026.105573","url":null,"abstract":"As artificial intelligence (AI) and generative AI (GenAI) technologies become increasingly integrated into everyday life, the need for validated tools that measure people’s knowledge about AI grows. Here, we present the development and validation of a theoretically driven, performance-based scale for assessing AI and GenAI knowledge. The scale is grounded in a two-axial framework. One axis captures three knowledge types: content knowledge (what AI is and where it is encountered), procedural knowledge (how AI systems operate and are used), and epistemic knowledge (what features and construction processes characterize AI outputs). The other axis encompasses three knowledge domains: technology-related knowledge (AI systems), user-related knowledge (users’ interaction with AI), and society-related knowledge (the social and ethical implications of AI). Based on an online survey of 800 internet-using adults from [Country], the 26-item scale was evaluated using confirmatory factor analysis, which demonstrated an acceptable model fit. It was further validated through two-stage structural equation modeling and group comparisons. Overall, the scale was found to be both valid and practically insightful: while it reproduces the expected relationships with additional constructs (e.g., trust in GenAI, attitudes toward AI) and expected differences between demographic groups, it also provides nuanced insights on the intricacies of AI knowledge. For example, the scale indicates that the relationship between trust in GenAI and knowledge about AI is grounded in both epistemic and societal knowledge. Thus, this novel tool affords more precise investigations into how different types and domains of AI knowledge relate to perceptions, behaviors, and decision-making in an AI-mediated world.","PeriodicalId":10568,"journal":{"name":"Computers & Education","volume":"2 1","pages":""},"PeriodicalIF":12.0,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145957052","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"教育学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-12DOI: 10.1016/j.compedu.2025.105556
Chia-Mei Lu
Scientific reasoning in simulation-based learning environments(SBLEs)is a time-structured process, not a terminal outcome. We advance and test a trait → mechanism → behavior model explaining how self-regulated learning(SRL)and critical thinking disposition(CTD)become consequential via a translation layer: dispositions first shape a reasoning-aligned configuration mechanism(SIM; planned contrasts, disciplined retesting), which enables process-level behavior(SIB; semantic precision, revision cadence). Eleventh graders(N = 168)completed a closed-loop aquatic simulation under Guided→Open or Open→Open sequences. Consistent Partial Least Squares for Reflective Constructs(PLSc)estimated reflective blocks; SIB was formative(Mode B). Measurement quality and cross-condition invariance were established; Cluster-Robust Variance Estimator, Type 2(CR2), and PLSpredict supported stability and utility. Process analytics(K-means profiles; three-state HMM)complemented the SEM. Findings: SRL and CTD had moderate positive effects on SIM; SIM had a vast, robust effect on SIB. Mediation showed trait effects reach behavior primarily through SIM. Format moderation was small/uncertain; temporally, lower-SRL learners dwelled longer in low-efficiency states, and editing cadence marked transitions to efficiency; early scaffolds modestly shortened dwell. Design principles: instrument platforms to monitor SIM/SIB as live control points; route support by profiles; and time-minimal, load-aware prompts to the revision window to restore Control of Variables Strategy(CVS)discipline and semantic alignment. The contribution is a validated, mechanism-aware account that yields diagnostic, feedback-ready, and scalable specifications for precision scaffolding and evaluation in SBLEs.
{"title":"Tracing Scientific Reasoning as Process: A Trait–Behavior–Performance Model with Learning Analytics in Simulated Environments","authors":"Chia-Mei Lu","doi":"10.1016/j.compedu.2025.105556","DOIUrl":"https://doi.org/10.1016/j.compedu.2025.105556","url":null,"abstract":"Scientific reasoning in simulation-based learning environments(SBLEs)is a time-structured process, not a terminal outcome. We advance and test a trait → mechanism → behavior model explaining how self-regulated learning(SRL)and critical thinking disposition(CTD)become consequential via a translation layer: dispositions first shape a reasoning-aligned configuration mechanism(SIM; planned contrasts, disciplined retesting), which enables process-level behavior(SIB; semantic precision, revision cadence). Eleventh graders(N = 168)completed a closed-loop aquatic simulation under Guided→Open or Open→Open sequences. Consistent Partial Least Squares for Reflective Constructs(PLSc)estimated reflective blocks; SIB was formative(Mode B). Measurement quality and cross-condition invariance were established; Cluster-Robust Variance Estimator, Type 2(CR2), and PLSpredict supported stability and utility. Process analytics(K-means profiles; three-state HMM)complemented the SEM. Findings: SRL and CTD had moderate positive effects on SIM; SIM had a vast, robust effect on SIB. Mediation showed trait effects reach behavior primarily through SIM. Format moderation was small/uncertain; temporally, lower-SRL learners dwelled longer in low-efficiency states, and editing cadence marked transitions to efficiency; early scaffolds modestly shortened dwell. Design principles: instrument platforms to monitor SIM/SIB as live control points; route support by profiles; and time-minimal, load-aware prompts to the revision window to restore Control of Variables Strategy(CVS)discipline and semantic alignment. The contribution is a validated, mechanism-aware account that yields diagnostic, feedback-ready, and scalable specifications for precision scaffolding and evaluation in SBLEs.","PeriodicalId":10568,"journal":{"name":"Computers & Education","volume":"29 1","pages":""},"PeriodicalIF":12.0,"publicationDate":"2026-01-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145957053","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"教育学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-08DOI: 10.1016/j.compedu.2026.105564
Yi-Chen Juan, Yuan-Hsuan Lee, Jiun-Yu Wu
Generative Artificial Intelligence (GenAI) functions not merely as a tool but an active collaborator in human knowledge construction; however, the Human-GenAI interaction dynamics is still underexplored. This study investigates Human-GenAI interaction profiles, the network interactivity and profile differences within a statistics learning community, as well as the underlying mechanisms linking Human-GenAI interaction to learning performance. We designed the Human–GenAI Inquiry and Problem-Solving Scaffold to foster shared agency between twenty-eight graduate students and GenAI across seven homework assignments in a sixteen-week advanced statistics course. Analytical approaches included k-modes clustering, social network analysis, and Partial Least Squares Structural Equation Modeling, complemented by case studies of interaction profiles. Three distinct Human-GenAI interaction profiles were identified: Human-GenAI collaborators, Peer collaborators with GenAI assistance, and Individual learners with late GenAI adoption. The network interactivity becomes cohesive with GenAI occupying the central hub role within the learning community. The models then demonstrate unique pathways through which Human-GenAI interaction influences learning performance, via degree centrality (number of direct connections) and peer nomination as helpers. The case studies highlight GenAI’s capability to augment human roles, encouraging deeper inquiry, expanding the depth of peer discussion, or promoting the exploration of diverse problem-solving strategies. These findings add value to theory and practice by providing empirical evidence for the framework of a scaffolded Human-GenAI shared agency, providing pedagogical implications to foster active student participation and harness GenAI’s potential to cultivate learner agency and symbiotic Human–GenAI knowledge construction.
{"title":"Generative Artificial Intelligence Augments Social Interactivity and Learning Outcomes: Advancing the Framework of a Scaffolded Human–GenAI Shared Agency","authors":"Yi-Chen Juan, Yuan-Hsuan Lee, Jiun-Yu Wu","doi":"10.1016/j.compedu.2026.105564","DOIUrl":"https://doi.org/10.1016/j.compedu.2026.105564","url":null,"abstract":"Generative Artificial Intelligence (GenAI) functions not merely as a tool but an active collaborator in human knowledge construction; however, the Human-GenAI interaction dynamics is still underexplored. This study investigates Human-GenAI interaction profiles, the network interactivity and profile differences within a statistics learning community, as well as the underlying mechanisms linking Human-GenAI interaction to learning performance. We designed the Human–GenAI Inquiry and Problem-Solving Scaffold to foster shared agency between twenty-eight graduate students and GenAI across seven homework assignments in a sixteen-week advanced statistics course. Analytical approaches included <ce:italic>k</ce:italic>-modes clustering, social network analysis, and Partial Least Squares Structural Equation Modeling, complemented by case studies of interaction profiles. Three distinct Human-GenAI interaction profiles were identified: Human-GenAI collaborators, Peer collaborators with GenAI assistance, and Individual learners with late GenAI adoption. The network interactivity becomes cohesive with GenAI occupying the central hub role within the learning community. The models then demonstrate unique pathways through which Human-GenAI interaction influences learning performance, via degree centrality (number of direct connections) and peer nomination as helpers. The case studies highlight GenAI’s capability to augment human roles, encouraging deeper inquiry, expanding the depth of peer discussion, or promoting the exploration of diverse problem-solving strategies. These findings add value to theory and practice by providing empirical evidence for the framework of a scaffolded Human-GenAI shared agency, providing pedagogical implications to foster active student participation and harness GenAI’s potential to cultivate learner agency and symbiotic Human–GenAI knowledge construction.","PeriodicalId":10568,"journal":{"name":"Computers & Education","volume":"94 1","pages":""},"PeriodicalIF":12.0,"publicationDate":"2026-01-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145957068","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"教育学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
As AI technologies permeate daily life, adolescents’ distinctive cognitive profiles make their decision-making highly sensitive to AI explanation features. The study aimed to examine the underlying mechanisms by which AI’s explanatory features and time frame impact adolescents’ decision-making. We created an online platform where adolescents interacted with an explainable AI. A preliminary survey identified 10 mathematics-related factors. Experiment 1 involved 158 students (Mage = 13.7) and used a 3 (explanation type: prediction, causal, counterfactual) × 2 (perceived control: high, low) × 2 (perceived reliability: reliable, unreliable) mixed design. Experiment 2 recruited 225 students (Mage = 13.7) and employed a 3 (explanation type) × 2 (time frame: short-term, long-term) mixed design. Decision-making and expectation (expected impact of each factor on math achievement) were the outcomes in both experiments. In Experiment 1, perceived unreliable counterfactual explanations for low-control factors produced the lowest expectation and decision-making probability, whereas predictions and causal explanations did not differ. For high-control factors, perceived reliable counterfactual explanations similarly reduced decision-making probability, although expectation remained constant across explanations. In Experiment 2, predictions and causal explanations led to higher decision-making probability for short-term events than long-term ones, while counterfactuals reversed this pattern. While counterfactual explanations help restore trust and motivate change in distant, uncertain contexts, they can trigger reactance and reduce action when events feel controllable or imminent. Although adolescents cognitively understand causality and time frames, they still struggle to effectively regulate their decisions. AI model explanations should therefore account for the developmental characteristics of adolescents and recognize the dual effects inherent in counterfactual explanations.
{"title":"How explanatory features of AI and time frame reshape adolescents’ decision-making","authors":"Zhuo Shen, Yinghe Chen, Jingyi Zhang, Hengrun Chen","doi":"10.1016/j.compedu.2026.105563","DOIUrl":"https://doi.org/10.1016/j.compedu.2026.105563","url":null,"abstract":"As AI technologies permeate daily life, adolescents’ distinctive cognitive profiles make their decision-making highly sensitive to AI explanation features. The study aimed to examine the underlying mechanisms by which AI’s explanatory features and time frame impact adolescents’ decision-making. We created an online platform where adolescents interacted with an explainable AI. A preliminary survey identified 10 mathematics-related factors. Experiment 1 involved 158 students (<ce:italic>M</ce:italic><ce:inf loc=\"post\">age</ce:inf> = 13.7) and used a 3 (explanation type: prediction, causal, counterfactual) × 2 (perceived control: high, low) × 2 (perceived reliability: reliable, unreliable) mixed design. Experiment 2 recruited 225 students (<ce:italic>M</ce:italic><ce:inf loc=\"post\">age</ce:inf> = 13.7) and employed a 3 (explanation type) × 2 (time frame: short-term, long-term) mixed design. Decision-making and expectation (expected impact of each factor on math achievement) were the outcomes in both experiments. In Experiment 1, perceived unreliable counterfactual explanations for low-control factors produced the lowest expectation and decision-making probability, whereas predictions and causal explanations did not differ. For high-control factors, perceived reliable counterfactual explanations similarly reduced decision-making probability, although expectation remained constant across explanations. In Experiment 2, predictions and causal explanations led to higher decision-making probability for short-term events than long-term ones, while counterfactuals reversed this pattern. While counterfactual explanations help restore trust and motivate change in distant, uncertain contexts, they can trigger reactance and reduce action when events feel controllable or imminent. Although adolescents cognitively understand causality and time frames, they still struggle to effectively regulate their decisions. AI model explanations should therefore account for the developmental characteristics of adolescents and recognize the dual effects inherent in counterfactual explanations.","PeriodicalId":10568,"journal":{"name":"Computers & Education","volume":"7 1","pages":""},"PeriodicalIF":12.0,"publicationDate":"2026-01-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145957069","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"教育学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-30DOI: 10.1016/j.compedu.2025.105555
Tinghui Wu , Xuesong Zhai , Yanjie Song
This study examines the effects of generative artificial intelligence (GAI)-enhanced pedagogical agents in the metaverse (GPAiM) on elementary school students' conceptual understanding of traditional festival customs and on the students' cognitive engagement patterns. The participants included 116 students from three intact classes. These classes were randomly assigned to two experimental groups (with 2D-GPAiM and 3D-GPAiM, respectively) and one control group (without GPAiM but with a real-person teacher). All the participants learned in the metaverse, and students in different groups were allowed to interact with 2D-GPAiM, 3D-GPAiM, and the real-person teacher during their learning, respectively. This study was conducted under a three-week AI literacy project with the learning topic of traditional festival customs. The results showed that the experimental groups (both 2D-GPAiM and 3D-GPAiM) had a positive impact on the students' conceptual understanding of traditional festivals, while the control group did not. More importantly, the 2D-GPAiM group showed a significantly positive difference in the participants’ conceptual understanding compared with the control group. In addition, regarding cognitive engagement, the 2D-GPAiM group showed a highly interactive, low-fluctuating, and high-level cognitive engagement pattern; The 3D-GPAiM group demonstrated a highly interactive, highly fluctuating, medium-level cognitive engagement pattern, while the control group exhibited a low-interactive, low-fluctuating, low-level cognitive engagement pattern. These findings provide valuable insights into future GAI-assisted pedagogical designs.
{"title":"The effects of GAI-enhanced pedagogical agents in the metaverse (GPAiM) on elementary school students’ conceptual understanding and cognitive engagement patterns","authors":"Tinghui Wu , Xuesong Zhai , Yanjie Song","doi":"10.1016/j.compedu.2025.105555","DOIUrl":"10.1016/j.compedu.2025.105555","url":null,"abstract":"<div><div>This study examines the effects of generative artificial intelligence (GAI)-enhanced pedagogical agents in the metaverse (GPAiM) on elementary school students' conceptual understanding of traditional festival customs and on the students' cognitive engagement patterns. The participants included 116 students from three intact classes. These classes were randomly assigned to two experimental groups (with 2D-GPAiM and 3D-GPAiM, respectively) and one control group (without GPAiM but with a real-person teacher). All the participants learned in the metaverse, and students in different groups were allowed to interact with 2D-GPAiM, 3D-GPAiM, and the real-person teacher during their learning, respectively. This study was conducted under a three-week AI literacy project with the learning topic of traditional festival customs. The results showed that the experimental groups (both 2D-GPAiM and 3D-GPAiM) had a positive impact on the students' conceptual understanding of traditional festivals, while the control group did not. More importantly, the 2D-GPAiM group showed a significantly positive difference in the participants’ conceptual understanding compared with the control group. In addition, regarding cognitive engagement, the 2D-GPAiM group showed a <em>highly interactive, low-fluctuating, and high-level cognitive engagement</em> pattern; The 3D-GPAiM group demonstrated a <em>highly interactive, highly fluctuating, medium-level cognitive engagement</em> pattern, while the control group exhibited a <em>low-interactive, low-fluctuating, low-level cognitive engagement pattern</em>. These findings provide valuable insights into future GAI-assisted pedagogical designs.</div></div>","PeriodicalId":10568,"journal":{"name":"Computers & Education","volume":"245 ","pages":"Article 105555"},"PeriodicalIF":10.5,"publicationDate":"2025-12-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145882238","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"教育学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-27DOI: 10.1016/j.compedu.2025.105552
Siyu Wu , Xiaodan Yu , Wei Wei
Parents can mitigate screens’ negative effects on school-aged children and adolescents by monitoring their screen use and improving their screen use skills. However, many preschoolers spend more time on screens than recommended. It remains unclear whether parental mediation is associated with preschoolers’ development through children’s screen time, and whether the first part of this pathway is moderated by problematic parental screen use. This analysis utilized parent-reported data collected in June 2019 about 57,827 children aged 4–5 years. Data included children’s developmental outcomes, children’s screen time, parental mediation (restrictive mediation, instructive mediation, and co-use) frequency, problematic parental screen use level, family income, and parental education. A significant negative correlation was found between children’s screen time and developmental outcomes (r = −0.07, 95 % confidence interval (CI) = [−0.08, −0.06]). Children’s screen time mediated the relations between parental mediation strategies and developmental outcomes. Restrictive mediation frequency was positively associated with developmental outcomes through children’s screen time (β = 0.016, 95 % CI = [0.013, 0.018]). Instructive mediation (β = −0.005, 95 % CI = [−0.006, −0.005]) and co-use (β = −0.004, 95 % CI = [−0.005, −0.003]) were indirectly, negatively associated with developmental outcomes through children’s screen time. Problematic parental screen use moderated the relations between parental mediation and children’s screen time. Higher problematic parental screen use strengthened restrictive mediation’s negative (β = −0.023, 95 % CI = [−0.032, −0.011]) and instructive mediation’s positive (β = 0.047, 95 % CI = [0.037, 0.057]) effects. Despite the modest effect sizes, the statistically robust results suggest that population-level adoption of combined parental strategies—reducing problematic parental screen use alongside implementing restrictive mediation—could translate into public health benefits for early childhood development.
家长可以通过监测学龄儿童和青少年的屏幕使用情况和提高他们的屏幕使用技能来减轻屏幕对他们的负面影响。然而，许多学龄前儿童花在屏幕上的时间超过了建议的时间。目前尚不清楚父母的调解是否通过儿童的屏幕时间与学龄前儿童的发展有关，以及这一途径的第一部分是否被有问题的父母屏幕使用所缓和。该分析利用了2019年6月收集的约57,827名4-5岁儿童的家长报告数据。数据包括儿童发育结果、儿童屏幕时间、父母干预（限制性干预、指导性干预和共同使用）频率、父母有问题的屏幕使用水平、家庭收入和父母受教育程度。儿童屏幕时间与发育结果呈显著负相关（r = - 0.07, 95%可信区间(CI) =[- 0.08, - 0.06]）。儿童屏幕时间在父母调解策略与发展结果之间起中介作用。限制性中介频率通过儿童屏幕时间与发育结果呈正相关（β = 0.016, 95% CI =[0.013, 0.018]）。指导性中介（β = - 0.005, 95% CI =[- 0.006, - 0.005]）和共同使用（β = - 0.004, 95% CI =[- 0.005, - 0.003]）与儿童屏幕时间的发展结果呈间接负相关。有问题的父母屏幕使用调节了父母调解与儿童屏幕时间之间的关系。较高的问题父母筛选率强化了限制性中介的负作用（β = - 0.023, 95% CI =[- 0.032, - 0.011]）和指导性中介的正作用（β = 0.047, 95% CI =[0.037, 0.057]）。尽管效果不大，但统计结果表明，在人口水平上采用联合父母策略——减少有问题的父母屏幕使用，同时实施限制性调解——可以转化为儿童早期发展的公共卫生效益。
{"title":"The indirect role of children’s screen time and the moderating role of problematic parental screen use on the relationships between different parental mediation strategies and preschoolers’ developmental outcomes","authors":"Siyu Wu , Xiaodan Yu , Wei Wei","doi":"10.1016/j.compedu.2025.105552","DOIUrl":"10.1016/j.compedu.2025.105552","url":null,"abstract":"<div><div>Parents can mitigate screens’ negative effects on school-aged children and adolescents by monitoring their screen use and improving their screen use skills. However, many preschoolers spend more time on screens than recommended. It remains unclear whether parental mediation is associated with preschoolers’ development through children’s screen time, and whether the first part of this pathway is moderated by problematic parental screen use. This analysis utilized parent-reported data collected in June 2019 about 57,827 children aged 4–5 years. Data included children’s developmental outcomes, children’s screen time, parental mediation (restrictive mediation, instructive mediation, and co-use) frequency, problematic parental screen use level, family income, and parental education. A significant negative correlation was found between children’s screen time and developmental outcomes (<em>r</em> = −0.07, 95 % confidence interval (CI) = [−0.08, −0.06]). Children’s screen time mediated the relations between parental mediation strategies and developmental outcomes. Restrictive mediation frequency was positively associated with developmental outcomes through children’s screen time (β = 0.016, 95 % CI = [0.013, 0.018]). Instructive mediation (β = −0.005, 95 % CI = [−0.006, −0.005]) and co-use (β = −0.004, 95 % CI = [−0.005, −0.003]) were indirectly, negatively associated with developmental outcomes through children’s screen time. Problematic parental screen use moderated the relations between parental mediation and children’s screen time. Higher problematic parental screen use strengthened restrictive mediation’s negative (β = −0.023, 95 % CI = [−0.032, −0.011]) and instructive mediation’s positive (β = 0.047, 95 % CI = [0.037, 0.057]) effects. Despite the modest effect sizes, the statistically robust results suggest that population-level adoption of combined parental strategies—reducing problematic parental screen use alongside implementing restrictive mediation—could translate into public health benefits for early childhood development.</div></div>","PeriodicalId":10568,"journal":{"name":"Computers & Education","volume":"245 ","pages":"Article 105552"},"PeriodicalIF":10.5,"publicationDate":"2025-12-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145845117","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"教育学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号