
ABSTRACT Background The development of AI technology is transforming the production of videos. In particular, voice‐over video lectures can be easily produced with AI voice, which has been found to generate learning comparable to that of human voice. Further, verbal cues, which are social cues—critical factors impacting video‐based learning—expressed with words, can be readily implemented with AI voice. However, the influence of verbal cues in video lectures has been relatively underinvestigated, limiting the optimization of (AI) voice‐over educational videos. Objectives This study aims to explore the effects of motivating language and personalization, both of which are verbal cues—a common but underexplored type of social cues, on learning satisfaction, actual learning outcomes, perceived learning, and rapport in AI voice‐over video lectures. Methods A three (motivating language: autonomy‐supportive vs. controlling vs. neutral) × two (personalization: personalized vs. nonpersonalized) between‐subjects experiment was carried out among 182 university students. Results and Conclusions Motivating language impacted learning satisfaction, retention, perceived learning, and rapport, and the impacts on learning satisfaction and rapport varied with personalization; but collectively, autonomy‐supportive language was more effective than neutral language, whereas controlling language was ineffective, irrespective of personalization; besides, personalization showed benefits only under autonomy‐supportive language in rapport, and instead caused decreased learning satisfaction and rapport under neutral language. The findings have several theoretical implications, including deepening knowledge about verbal cues, extending theories about effects of social cues in multimedia learning, and broadening understanding of motivating teaching. Practical implications for the design of AI voice‐over video lectures were discussed.
ABSTRACT Background Computer‐supported collaborative learning (CSCL) is widely used in higher education, yet participation remains uneven and discourse often stays at a low cognitive level. Asynchronous online discussion (AOD) and mobile instant messaging (MIM) are two common communication environments, but how social participatory roles are expressed through discourse across these environments remains unclear. Objectives This study examined conversational‐function and cognitive‐engagement profiles associated with social participatory roles in recorded AOD and MIM interactions. Methods Thirty first‐year undergraduates in eight groups completed two collaborative tasks using Blackboard forums (AOD) and Lark (MIM) in a counterbalanced cross‐over design. Social participatory roles were inferred via social network analysis; conversational functions were coded with the ASIMeC‐F scheme; and cognitive engagement was coded at four levels (operation, way‐finding, sense‐making, innovation). Data were analysed using paired role‐distribution comparisons, binary mixed‐effects logistic regression, descriptive role‐specific summaries, and epistemic network analysis (ENA). Results and Conclusions No difference in marginal role distributions was detected, although individual role transitions were common. Adjusted models associated AOD with a higher prevalence of Analysing and MIM with a higher prevalence of Respond and Commenting without elaboration. Role‐specific profiles illustrated different discourse patterns under similar role labels, while ENA described co‐occurrence patterns within the conversational‐function and cognitive‐engagement coding schemes. These findings inform a context‐sensitive account of role enactment and suggest role‐based scaffolding for further testing.
ABSTRACT Background Visual programming has experienced a remarkable surge in popularity recently. Real‐time and delayed rewards are two important incentive mechanisms in designing effective learning activities. However, the examination of their effects on students' learning performance remains limited. Additionally, the comparative analysis of these reward mechanisms in the context of learning visual programming has received insufficient attention. Objectives We aimed to address this research gap. Methods A quasi‐experimental study was conducted involving two classes. One class ( n = 32) received real‐time rewards, while the other class ( n = 40) received delayed rewards. The study employed independent samples t ‐tests to assess the effects of real‐time versus delayed rewards on students' learning performance and to identify differences between the groups. Additionally, semi‐structured interviews were conducted to gain insights into the underlying reasons behind the findings. Results The results showed that students in the real‐time reward group generally outperformed their counterparts in terms of intrinsic motivation, extrinsic motivation, flow experience and algorithmic thinking. However, only the differences in extrinsic motivation and algorithmic thinking were statistically significant. Conclusions The implications of this study are valuable for educators and researchers, providing insights into how to design and implement visual programming learning activities using diverse reward systems as incentive mechanisms. These findings can help enhance students' learning performance more efficiently.
ABSTRACT Background With the rapid advancement of artificial intelligence (AI), research in the cross‐disciplinary field of science, technology, engineering, and mathematics (STEM) education has increasingly focused on how to leverage AI to foster computational thinking (CT) among K‐12 students, given that CT, recognized as a core 21st‐century competency, plays a pivotal role in cultivating students’ literacy and equipping them with the ability to tackle complex problems. Objective Grounded in activity theory, this study seeks to identify the key factors shaping K‐12 students’ CT development in AI‐enhanced STEM education. It systematically analyzes: (a) the subjects involved and the types and roles of AI tools employed, (b) the purposes, periods, and CT dimensions within the object, (c) the methods, strategies, and evaluation performance applied within rules, (d) the learning domains within communities and the interactions within the division of labor, and (e) the developmental trends in outcomes. Method This study employed the PRISMA framework to systematically review 57 SSCI journal publications published between 2017 and 2025. In addition, Epistemic Network Analysis (ENA) was utilized to examine the current state of CT evaluation. Results and Conclusions Findings indicate that: (1) the primary subjects of the studies were elementary school students; expert systems and robotics were the most prevalent types of AI, which were frequently employed as learning tools; (2) the primary objective was to evaluate the effectiveness of AI in CT–K12–STEM, with periods typically ranging from 2 to 6 months, focusing on conceptual knowledge and practical skills; (3) the rules focused on exploratory learning and gamification strategies, employing quantitative research methods; (4) communities spanned cross‐disciplinary domains, with division of labor tending towards mixed interaction types; and (5) outcomes focused on the learning cognitive perspective, encompassing aspects such as learning performance and problem solving. Based on the research results, practical recommendations are provided for educational researchers and policymakers.
ABSTRACT Background Integrating artificial intelligence (AI) tools into self‐regulated learning (SRL) is complex and multifaceted. Configurational approaches are based on the concept that combinations (or configurations) of causally relevant factors should be connected to the desired outcome. Objectives Self‐determination theory (SDT), the unified theory of acceptance and use of technology (UTAUT) and perceived affordances focus on basic needs, technology adoptions and technological designs, respectively. Taking a configurational approach, this study investigates how the combinations of SDT, UTAUT and perceived affordances affect SRL with AI. Methods We employed fuzzy‐set qualitative comparative analysis (fsQCA) to uncover the complex interplay of factors that facilitate effective SRL. Data from 248 students reveal that a combination of autonomy satisfaction, relatedness satisfaction and performance expectancy is the most influential factor affecting the use of AI in SRL. Results and Conclusions Our results suggest that both innate psychological needs and external learning environments should be considered together for understanding the adoption of AI in SRL. They also further highlight the critical role of perceived affordances as mediators and moderators, enhancing the relationships between innate psychological needs and external learning environments. This integrative approach provides valuable insights into the design and implementation of AI‐supported SRL environments. We recommend designing the environments to foster autonomy and relatedness while also ensuring ease of use for learners.