Collaborative problem-solving skills have been widely recognized as crucial and necessary skills in the 21st century (Andrews-Todd et al., 2023; Li et al., 2023; Partnership for 21st Century Skills, 2019). These skills have been defined in terms of the ability to solve problems effectively and obtain solutions through interaction with several agents (OECD, 2017). High-level collaborative problem-solving skills can help foster critical thinking skills (Xu et al., 2023), thereby promoting learning achievements (Hwang & Chen, 2023) and learning engagement (Unal & Cakir, 2021). However, previous studies have indicated that the collaborative problem-solving skills of learners require further improvement (Lu et al., 2023; OECD, 2017). According to employers, many college students are not equipped with sufficient collaborative problem-solving skills (D’Mello et al., 2024). Global social challenges also necessitate fostering and promoting collaborative problem-solving skills (Cheruiyot & Molnár, 2025). In addition, learners face certain difficulties in solving problems effectively, including a lack of personalized feedback (Aslan et al., 2025) and monitoring (Haataja et al., 2022) as well as a failure to generate solutions (Kwon et al., 2019).

Previous studies have attempted to use diverse technologies to improve individuals’ collaborative problem-solving skills, including conversational artificial intelligence (AI)-mediated learning environments (Aslan et al., 2025), web-based learning, and game-based learning tools (Dehghan-Chaleshtori, 2025). Recently, intelligent agents have emerged as a dominant trend in computer-mediated learning environments, given their ability to facilitate dynamic engagement that mirrors human-like interactions (Lang et al., 2022). Intelligent agents have increasingly begun to incorporate affective and conversational attributes while taking on social roles in the context of learner interactions (Sikström et al., 2024). In addition, researchers have revealed that multiagent systems (MASs) are superior to single-agent systems because of their ability to facilitate parallel cooperative operations, autonomous agent coordination, and distributed task execution (Viswanathan et al., 2022). However, the use of intelligent agents in education has a dual impact. Namely, although this approach can benefit cognitive and emotional learning outcomes, it simultaneously raises concerns regarding psychological effects, efficiency, and usability; furthermore, the findings of research on this approach have remained contradictory or inconclusive regarding its overall educational value (Dolata et al., 2023).

In addition, generative artificial intelligence (GenAI) has been viewed as a valuable resource for facilitating collaborative problem solving in educational contexts (Wei et al., 2025). As the initial form of GenAI and a current subset of such technologies, large language models (LLMs) focus on processing and generating human-like content, allowing them to be applied widely in many fields (Bewersdorff et al., 2025). LLM-based MASs have distinct abilities to facilitate interactions and complete complex problem-solving and world simulation tasks (Guo et al., 2024). Nevertheless, very few empirical studies have thoroughly examined the effects of a GenAI-enhanced multiagent approach in collaborative problem-solving scenarios. To address these research gaps, this study proposes and validates a GenAI-enhanced multiagent approach to empowering collaborative problem solving.

The proposed GenAI-enhanced multiagent approach is grounded in social agency theory and social constructivism theory. Social agency theory posits that social factors such as social presence and social cues are important to promote learning performance (Mayer, 2014). Social constructivism theory states that cognitive growth occurs first at the social level and later at the individual level and that the active construction of knowledge occurs as a dynamic process of social interaction (Vygotsky, 1978). The use of multiple GenAI-enhanced agents is intended to create social presence, social cues, and social interaction to facilitate the elaboration of knowledge and collaborative problem-solving. Knowledge elaboration refers to a constructive cognitive process in which learners need to integrate new information with prior knowledge (Kalyuga, 2009). Thus, the following research questions are proposed.

  • 1.

    To what extent does a GenAI-enhanced multiagent approach promote students’ learning achievements, knowledge elaboration, and collaborative problem-solving performance and skills?

  • 2.

    What perceptions do learners have when using a GenAI-enhanced multiagent approach?

The results of the current study suggest that this GenAI-enhanced multiagent approach substantially affects learning achievements, knowledge elaboration, and collaborative problem-solving performance and skills. This study elucidates how to improve collaborative problem-solving skills based on GenAI technologies. The findings of this study contribute to the construction of a theoretical model for collaborative problem solving, enabling us to better understand the mechanism of collaborative problem solving and how collaborative problem solving evolves over time. This study provides insight into the implementation of massive amounts of personalized learning and precise interventions as well as the development of policies concerning the use of multiple GenAI-enhanced agents in practice.

Source link