Navigating AI feedback: Changing perceptions and argumentation skills in science teacher education

This study aimed to examine pre-service teachers’ (PST) perceptions of AI-supported argument feedback (AIAF), changes in their perceptions, argument level and component development, process experiences, and to derive principles for AIAF design. The research was conducted in two cycles using the Design-Based Research approach. In the first cycle, prompts based on the Toulmin argumentation model (2003) and Erduran et al. (2004)’s argument level framework was developed for ChatGPT (GPT-4). In the second cycle, these prompts were tested through transcripts of scenario-based discussions with PSTs to provide personalized feedback. Data were collected from semi-structured interviews, discussion transcripts, and AI-generated feedback outputs. The study group consisted of nine science PSTs selected through purposive sampling from a state university in Turkey. The findings showed that the PSTs’ perceptions of feedback evolved from a descriptive level to a pedagogical and affective understanding; they began to see AI as a complementary tool. While the quantity of arguments decreased, their quality increased; progress was made in refutation and multi-component argument production. PSTs found AI feedback positive in terms of speed and motivation but limited in terms of personalization. The study recommends the principles of holistic argument analysis, speed, interaction, and individualization for AI-supported feedback design. This study presents original data to the relevant literature, and its findings form the basis for a comprehensive study currently being conducted with a larger sample using a pre-test–post-test design.