Despite the established prominence of collaborative learning in higher education (Scager et al., 2016), the nuances of group performance and success in computer-supported collaborative learning environments remain under-researched, particularly with the introduction of artificial intelligence (AI) as a co-author. Existing literature on collaborative learning suggests that the composition of a group plays a significant role in shaping group interactions and influencing learning outcomes (Wang et al., 2007). However, the process of organizing students into effective groups remains complex and requires careful consideration. Prior research demonstrated mixed findings about the relationship between the level of prior knowledge, the homogeneity of the group and the group size and group performance on problem-solving tasks. Moreover, the influence of group composition on academic performance has become increasingly intricate with the rapid integration of AI technologies in higher education (Msambwa et al., 2025). While some studies have investigated human-AI collaboration in learning environments (Martínez et al., 2021; Xia et al., 2023), there remains a gap in research regarding how specific group characteristics impact the quality of collaboratively generated solutions when AI functions as a co-author (Ait Baha et al., 2024; Kumar, 2021).

This study explores the impact of group composition on the quality of collaboratively generated solutions when artificial intelligence (AI) is integrated as a co-author in problem-solving activities. The research focuses on how specific group characteristics—such as prior knowledge, heterogeneity, and group size—mediate the effectiveness of human-AI collaboration in higher education settings.

Three major research questions are addressed:

RQ1) What is the relationship between the level of prior knowledge within a group and the quality of AI-assisted problem-solving solutions?

RQ2) What is the relationship between group heterogeneity in prior knowledge and the quality of AI-assisted problem-solving solutions?

RQ3) What is the relationship between the group size and the quality of AI-assisted problem-solving solutions? 

The primary aim of this study is to examine the extent to which group composition relates to the quality of collaboratively generated solutions when AI is utilized as an active participant in problem-solving. 

This research is grounded in Collaborative Learning Theory and Cognitive Load Theory. Collaborative Learning Theory posits that interaction among group members fosters deeper understanding and knowledge construction, particularly when individuals bring diverse perspectives to the discussion. Conversely, Cognitive Load Theory highlights the potential cognitive strain that group members may experience when working together and with AI or when navigating group-based problem-solving dynamics. 

Methodological Approach:

The study adopts a quantitative research design, employing multiple regression analysis to assess the relationship between group composition and solution quality. The research was conducted with 196 bachelor's students enrolled in a macroeconomics course, who engaged in group problem-solving activities with the assistance of the ChatGPT language model. 

This study employed a quantitative research design to examine the relationship between group composition and the quality of AI-assisted collaborative problem-solving. Data were collected from 196 second-year undergraduate students enrolled in an International Economics and Finance program at a large university. Participants were assigned to groups of five to eight members using a stratified randomization technique based on their academic rankings. During the first half of the semester, students engaged in collaborative problem-solving without AI assistance. In the latter half, AI was introduced as a co-participant, with students utilizing ChatGPT (version 3.5) to support their discussions and generate solutions for problem sets. The study analyzed data from four seminar sessions, each incorporating AI-assisted collaboration. To assess the quality of AI-generated solutions, a standardized rubric was developed by an expert in macroeconomics education and later refined by a second expert in the field. The rubric evaluated student responses based on accuracy, application of economic models, and depth of analysis. Scores were converted into a percentage scale ranging from 0 to 100. Group prior knowledge was measured using students' previous scores in a prerequisite mathematical analysis course, also converted into a percentage scale. Group heterogeneity was determined by calculating the standard deviation of prior knowledge scores among group members. Group size was recorded as the number of students in each team. The study relied on primary data collected through student assessments and performance records. Statistical analyses, including multiple regression models, were conducted using R software to explore the relationships between group characteristics and the quality of AI-assisted solutions. Assumptions for linear regression, including normality and multicollinearity, were tested to ensure the robustness of findings.

It was found that groups with higher prior knowledge produced solutions of higher quality with the help of AI. However, heterogeneous groups were found to perform worse than homogenous groups when collaborating with AI. Notably, larger groups demonstrated more effective AI-supported problem-solving than smaller groups. These findings highlight the importance of carefully considering group composition when designing AI-supported collaborative activities, as it plays a significant role in determining the quality of outcomes.

Ait Baha, T., El Hajji, M., Es-Saady, Y., & Fadili, H. (2024). The impact of educational chatbot on student learning experience. Education and Information Technologies, 29(8), 10153–10176. https://doi.org/10.1007/s10639-023-12166-w An, S., Zhang, S., Guo, T., Lu, S., Zhang, W., & Cai, Z. (2025). Impacts of generative AI on student teachers’ task performance and collaborative knowledge construction process in mind mapping-based collaborative environment. Computers & Education, 227, 105227. https://doi.org/10.1016/j.compedu.2024.105227 Bringula, R. P., Basa, R. S., Dela Cruz, C., & Rodrigo, Ma. M. T. (2016). Effects of Prior Knowledge in Mathematics on Learner-Interface Interactions in a Learning-by-Teaching Intelligent Tutoring System. Journal of Educational Computing Research, 54(4), 462–482. https://doi.org/10.1177/0735633115622213 Janssen, J., & Kirschner, P. A. (2020). Applying collaborative cognitive load theory to computer-supported collaborative learning: Towards a research agenda. Educational Technology Research and Development, 68(2), 783–805. https://doi.org/10.1007/s11423-019-09729-5 Kim, J., & Lee, S.-S. (2023). Are Two Heads Better Than One?: The Effect of Student-AI Collaboration on Students’ Learning Task Performance. TechTrends, 67(2), 365–375. https://doi.org/10.1007/s11528-022-00788-9 Kirschner, P. A., Sweller, J., Kirschner, F., & Zambrano R., J. (2018). From Cognitive Load Theory to Collaborative Cognitive Load Theory. International Journal of Computer-Supported Collaborative Learning, 13(2), 213–233. https://doi.org/10.1007/s11412-018-9277-y Liverpool-Tasie, L. S. O., Adjognon, G. S., & McKim, A. J. (2019). Collaborative learning in economics: Do group characteristics matter? International Review of Economics Education, 31, 100159. https://doi.org/10.1016/j.iree.2019.100159 Retnowati, E., Ayres, P., & Sweller, J. (2018). Collaborative learning effects when students have complete or incomplete knowledge. Applied Cognitive Psychology, 32(6), 681–692. https://doi.org/10.1002/acp.3444 Zambrano R., J., Kirschner, F., Sweller, J., & Kirschner, P. A. (2019). Effects of prior knowledge on collaborative and individual learning. Learning and Instruction, 63, 101214. https://doi.org/10.1016/j.learninstruc.2019.05.011