Self-regulated learning is essential to academic success, especially in online environments, where learners are expected to self-manage large amounts of information and take control of their learning process. Learners who struggle with self-regulated ...
Self-regulated learning is essential to academic success, especially in online environments, where learners are expected to self-manage large amounts of information and take control of their learning process. Learners who struggle with self-regulated learning experience cognitive load during this process, which can lead to decreased academic self-efficacy and academic performance. One strategy to support them, metacognitive prompts, has been shown to be effective in helping them reflect and regulate their learning process. However, there are still many challenges in integrating them into the classroom. Recent advances in generative artificial intelligence offer the possibility of easily integrating metacognitive prompts into online learning, but empirical studies validating their effectiveness are scarce. Therefore, this study aims to explore the effects of metacognitive prompts using generative AI on learners' cognitive load, academic self-efficacy, and academic performance.
The present study used an experimental design to investigate how metacognitive prompts affect learners’ cognitive load, academic performance, and academic self-efficacy. For this study, ChatGPT was customized using the GPTs feature that allows to create a customized ChatGPT tailored to personal needs. A total of 40 undergraduate and graduate students in South Korea participated, and they were divided into two groups. Students in the experimental group (n=20) received metacognitive prompts by ChatGPT, while the comparison group (n=20) only used ChatGPT without prompts.
Prior to the experiment, all students conducted a pre-survey on academic self-efficacy. The experiment consisted of two sessions, which followed identical procedures. During each session, all students watched a 30-minute video lecture on Python programming. After, students received a programming problem-solving task relevant to the video lecture, where each group used ChatGPT with or without metacognitive prompts. Participants completed a cognitive load survey and took an immediate test upon finishing the given task. Students took a retention test a week after each experiment session. Two sessions varied in task difficulty, with the first having a low-level task and the second having a high-level task. After the end of the second session, all students answered a post-survey on academic self-efficacy, marking the end of the experiment.
The present study collected self-reported surveys on cognitive load and academic self-efficacy and used immediate and retention test scores to measure academic performance. Conversation logs with ChatGPT from all participants were also collected to investigate the problem-solving process and to explore further how these processes affected the academic performance of the participants. The collected data were analyzed using the Mann-Whitney U test, Aligned Rank Transform Test, and Chi-Square test. An Aligned Rank Transform (ART) test was performed to determine whether there were differences in cognitive load scores between groups based on task difficulty and the presence of metacognitive prompts. Also, an ART test was conducted to examine whether there was a difference in pre-post academic self-efficacy scores within and between groups. A Mann-Whitney U test was conducted to see if there are differences between groups in academic performance. Finally, conversation logs with ChatGPT were coded into the four stages of the problem-solving framework of Carlson and Bloom (2005), and a Chi-square test was run to see if there was a difference in the problem-solving process between groups.
The results show that the experimental group scored lower on cognitive load than their counterparts, but the difference between groups and task difficulty in cognitive load across sessions was not statistically significant. For academic performance, participants who received metacognitive prompts scored higher than those who did not, but the difference was not statistically significant. Both groups reported increased academic self-efficacy after the experiment, with the group with metacognitive prompts leading to higher post-test scores. However, the interaction effect was not statistically significant. On the other hand, there was a meaningful difference in problem-solving processes. The experimental group engaged evenly across all phases of ‘Orientation’, ‘Planning’, ‘Execution’, and ‘Checking’. In contrast, the comparison group mostly engaged in ‘Orientation’, where they acquired conceptual knowledge for problem-solving, and was overly dependent on ChatGPT's responses, showing less balanced engagement. These findings suggest that metacognitive prompts can support learners' self-regulatory processes by helping them stay productive during the problem-solving process, which can alleviate cognitive load, improve academic performance, and build confidence in their abilities in this process. Despite these findings, the lack of statistically significant between-group differences highlights the need for further research with larger sample sizes and in different contexts to better understand the potential for integrating metacognitive prompts and generative AI.