Using Artificial intelligence for qualitative analysis of adolescents’ interviews about moral growth.

Introduction. In May 2024, funded by the Recovery and Resilience Facility project No. 5.2.1.1.i.0/24/I/CFLA/007, grant No. LU-BA-PA-2024/1-0011, 171 Latvian pupils in Years 7 and 9 (14-16 years old) participated voluntarily in a written structured interview about their experiences, current situation, and plans regarding their moral growth. This paper presents the lessons learnt from the qualitative analysis of those interviews, combining AI and computer-assisted potentialities. 

Goals and Methods. The research question addressed in this paper is: How useful and reliable was the AI tool for qualitative analysis of interviews? First, the interviews of each class were summarised separately using the AI tool ChatGPT-4o, and two researchers reviewed and corrected the summaries for reliability. Secondly, a frequency analysis of particular codes was implemented, first using AQUAD 8 software with a human-created code catalogue and then using the AI tool for checking AI potential and reliability in this numeric task. Finally, relevant quotations were retrieved in the data of Year 7 pupils’ interviews, first using AQUAD 8 software and then the AI for checking AI potential and reliability in this task, which required a higher degree of sophistication.

Results. The AI tool was useful and reliable for summarising pupils’ reported opportunities (e.g., family kindness), challenges (e.g., lack of motivation), and strategies (e.g., helping others) guiding their moral development (step 1). The AI tool had a low reliability for counting the most frequent codes in the data (step 2), often missing the textual context in Latvian. The AI tool was useful and moderately reliable for selecting significant quotations according to predetermined topics (step 3), but human-driven verification was still necessary.

Conclusions. The findings highlight the intervention's impact on adolescents' moral growth. This study shows that the AI tool ChatGPT-4o can notably accelerate qualitative data analysis, but human oversight is still necessary for ensuring reliability.