Does the size of the group in collaborative learning affect the productivity of working with AI?
In July of this year, the article titled ‘Collaboration with AI: How group composition affects solution quality in problem solving’ was published in an international academic journal. Its authors are Galina Shulgina, Aleksandra Getman – the Fellows of IOE, Ilya Gulenkov – a researcher from Faculty of Economic Sciences, and Jamie Costley – a researcher from UAE University, who decided to explore how the size and level of knowledge of the participants affect the results of the group work if AI is involved in the process.
As the researchers emphasise, although collaborative learning remains a cornerstone of higher education, the factors underlying effective group work have not been sufficiently studied. At the same time, the field has become particularly interesting since artificial intelligence (AI) began to be used as a co-author.
The study is based on data from group work involving 196 undergraduate students (55% men, and 45% women), who participated in group problem-solving assignments as part of the International Economics and Finance Program at a large Russian university. Participants were divided into groups of five to eight people, with varying levels of knowledge. During the first phase, the groups collaboratively worked on a series of analytical assignments without AI support. In the second phase, conducted over four seminar sessions, the same groups engaged with ChatGPT-3.5. The task was not simply to get an answer from AI, but to critically analyse it, apply the economic models from the course, and present a comprehensive solution.
The quality of the solutions was evaluated based on the accuracy and level of detail in the students’ responses. The highest scores were awarded to teams that both applied AI correctly and pointed out its limitations.
According to the data obtained, several patterns were identified. Firstly, the analysis revealed that groups with higher levels of prior knowledge generated more accurate and comprehensive solutions with the help of AI . Secondly, a negative correlation was found between group diversity and the quality of decisions generated by AI. In other words, teams with more variation in prior knowledge tended to demonstrate worse results. Finally, the researchers concluded that groups with more participants generated better solutions with the help of AI.
One of the most meaningful findings was the data indicating a positive correlation between larger group sizes and better AI-related results. On average, teams of seven to eight people performed better than groups of five to six participants. Each additional team member increased the final score. This contradicts the widespread belief in education that small groups are more effective than large ones. The researchers suggested that large teams have greater intellectual resources and a wider range of views and skills, allowing them to interact more productively with AI. It is assumed that there is a non-linear correlation, whereby large groups become less effective over time. However, this is a topic for future research.
Furthermore, the study has practical importance and its findings could be used regarding facilitating group work with the help of AI. The researchers suggest that teachers should consider forming larger groups for collaboration between students and AI, which can benefit from the diverse perspectives, knowledge and problem-solving resources that larger teams offer. Besides, teachers should consider grouping students with similar levels of prior knowledge. In the authors' opinion, this homogeneity can facilitate more effective use of AI to reinforce understanding and deepen learning.
For future research, the scientists suggest studying a wider range of group sizes to better understand their relationship with the effectiveness of collaborative work.
We congratulate our colleagues on this excellent publication, and wish them further success in their research!