The continuous progress in large language models (LLMs) and agent technologies have made LLM-based agents a promising new tool for enhancing the quality and efficiency of teaching and learning in the educational domain. Based on the differences in functional positioning of LLM-based agents, single-agent systems (SASs) are capable of supporting various teaching and learning tasks, such as content generation, intelligent feedback, and assessment. However, due to the homogeneous nature of their interaction characteristics and functional attributes, SASs are confronted with limitations in promoting deeper cognitive development. In contrast, multi-agent systems (MASs) simulate various educational roles, enhancing the diversity and depth of learning interactions and thereby providing more personalized and profound learning experiences. Considering that the application of agents in the learning process primarily relies on learners’ initiative, this study designs motivational learning activities for both SASs and MASs based on the attention, relevance, confidence, and satisfaction (ARCS) model of motivation. To ensure the effective implementation of these activities, a quasi-experimental study is subsequently conducted in the context of English reading. The experimental results reveal that MAS-based motivational learning activities significantly outperform SASs in improving students’ academic achievement in reasoning, evaluation, and application tasks. Furthermore, MASs exhibit a stronger boost in the learning motivation of students and effectively promote deeper cognitive development, particularly in abstraction and generalization skills. This study underscores the value of MASs in supporting in-depth learning and provides insights for further exploration of their applications in education.

Amid the rapid evolution of generative artificial intelligence (GenAI), how to effectively leverage its affordances to advance students’ creative thinking development has emerged as a frontier issue in educational research. However, existing studies generally encounterthe dilemma of unclear coupling pathways between GenAI and thinking development, and a lack of pedagogical implementation for technological applications. To tackle these issues, this study constructs a multi-agent collaborative system oriented toward creative problem solving. The system is designed based on the dual logics of cognitive processes and psychological drivers, incorporating six core agents, a dynamic scheduling mechanism, and a three-layer task–knowledge–thinking knowledge graph to support the development of students’ creative thinking. A quasi-experimental study with four task rounds is conducted to compare the effectiveness of this system with that of a single agent model in supporting undergraduates’ creative problem solving. The results indicate that students’ overall creative thinking ability improved significantly, with the experimental group outperforming the control group and the performance gap widening progressively across tasks. Post hoc comparison further reveals that advancing creative thinking occurs primarily in the middle-to-late stages of the tasks. These findings demonstrate that the support of the multi-agent collaborative system can effectively activate and reorganize students’ cognitive representations through sustained interaction, thereby facilitating a transition in creative thinking from gradual development to stage-based advancement.

Based on a survey of 12,678 undergraduates across 20 higher education institutions (HEIs) in China, this study shows that most students exhibit a critical use tendency when engaging in higher-order thinking tasks, such as critical and creative tasks with generative artificial intelligence (GenAI), and believe that GenAI-assisted learning positively contributes to their problem-solving ability. Furthermore, a latent profile analysis and regression analysis reveal four distinct types of GenAI-assisted learning users, reflecting the interaction between students’ level of technology acceptance and their self-regulated learning abilities. Firstly, cautious experiencers (42%) exhibit low acceptance, high critical use, and low expectations for skill development. Secondly, labor substituters (24%) demonstrate low acceptance, low critical use, and moderately high expectations for skill development. Thirdly, balanced explorers (21.2%) show high acceptance, low critical use, and moderately high expectations for skill development. Fourthly, deep users (12.8%) exhibit high acceptance, high critical use, and high expectations for skill development. Based on these findings, this study underscores the necessity for HEIs to guide undergraduates in the appropriate use of GenAI tools and proposes effective strategies, such as targeted educational interventions and the cultivation of artificial intelligence literacy. These measures will advance students’ transition toward the ideal profile of high acceptance, high critical use, and high expectations for skill development.

Human–machine collaboration is a development trend in the future of education. Generative artificial intelligence (GenAI), represented by ChatGPT, has redefined the interaction modes in human–machine collaborative learning. However, existing research is deficient in a systematic discussion on the interaction modes between learning groups and GenAI. Based on the analysis of dialogue texts from human–machine collaborative learning supported by GenAI, three common types of questions asked by students to GenAI are application, analysis, and examples. In student–machine interaction, three primary types of interactive behaviors emerge: exploratory questioning, optimization questioning, and information integration, each of which shows a clear pattern of cognitive engagement shift. In student–student interaction, three major interactive behaviors emerge: task comprehension, theme discussion, and non-task communication, with atypical cognitive engagement shift patterns. Based on the co-occurrence of question types and human–machine interaction behaviors, three interaction modes are observed when the learning group collaborates with GenAI to complete learning tasks: the dominant-analytical mode, the integrative-thinking mode, and the answer-oriented mode. The dominant-analytical mode focuses on in-depth exploration and critical thinking, the integrative-thinking mode highlights the key role of in-group communication for integrating diverse information and reflective thinking, and the answer-oriented mode centers around human–machine interaction based on the information provided by GenAI. To enhance the effectiveness of human–machine collaborative learning supported by GenAI, it is necessary to promote high-cognitive learning with an emphasis on thinking skills and innovative awareness, foster human–machine dialogic learning centered on questioning, and adjust instructional design according to the demands of learning tasks.

The accelerating transformation of the education system toward generative artificial intelligence (AI) imposes an urgent need to enhance teachers’ AI educational literacy. It is crucial to systematically clarify its influencing factors and action mechanisms to support the profound transformation of intelligent education. This study identifies the core factors influencing teachers’ AI educational literacy. A structural equation model is developed and validated to examine such factors across three dimensions: the internal structure of AI educational literacy, teachers’ individual characteristics, and external school environment. This study finds that for internal structure, AI educational mindset positively affects AI educational ethics, foundations and applications, and pedagogy. AI educational foundations and applications positively affect AI pedagogy. Both factors contribute to AI for professional development, with AI pedagogy serving as a mediating variable. Among individual characteristic factors, AI self-efficacy and educational trust impact AI educational literacy via AI educational mindsets, while AI application behavioral intention directly affects AI for professional development. In terms of the school environment factors, AI leadership of principals positively affects AI educational ethics and pedagogy. Additionally, AI infrastructure and school organizational climate interact synergistically to build an external environment that supports the development of teachers’ AI educational literacy. Accordingly, this study proposes systematic strategies to promote coordinated development of various dimensions of AI education literacy for primary and secondary school teachers to optimize literacy structure, to empower individual teachers, and to enhance the supportive school environment.

The Smart Education of China platform for primary and secondary education (SECPSE) serves as a crucial support for achieving the digital transformation of classroom teaching. Focusing on the practical efficacy of this platform in empowering classroom teaching, this study employs a mixed-methods approach. It conducts a questionnaire survey involving 69,304 primary and secondary school teachers nationwide, supplemented by an in-depth analysis of teaching behaviors and epistemic networks of 20 typical cases, as well as a grounded theory analysis of 845 demand texts. This comprehensive investigation systematically explores the platform’s application status, typical models, and optimization pathways. The findings reveal that, firstly, teachers exhibit relatively high overall satisfaction with the platform, and its application has preliminarily formed a pattern covering the entire teaching process. Secondly, the application of the SECPSE in classroom teaching primarily manifests in three typical models. The tool-enabled model deeply embeds technology into the teaching process through on-demand utilization of platform resources and tools. The dual-teacher model leverages expert teacher video lectures from the platform to implement collaborative teaching between online experts and offline teachers, effectively expanding the coverage of high-quality resources. The self-directed inquiry model relies on the platform’s learning resources, task-pushing capabilities, and learning analytics to promote students’ autonomous knowledge construction. Thirdly, based on the grounded theory, a model encompassing content, tool, service, and mechanism needs is constructed. Accordingly, targeted optimization strategies are proposed. These include optimizing content supply to build a precise resource ecosystem, upgrading tool functionalities to enhance teaching adaptability, improving service support to establish a full-coverage guarantee system, and innovating mechanism development to stimulate momentum for sustained application. The theoretical model and methodological strategies constructed in this study may provide theoretical support and practical references for the promotion and application of the SECPSE.