The effects of artificial intelligence– and web-supported argumentation-based science learning approaches on middle school students' scientific argumentation skills, digital literacy levels and views on the nature of science

In today's world, access to information and learning processes have changed radically. The integration of artificial intelligence into our lives has further accelerated digital transformation. With this rapid change, the ways in which students access information and use knowledge have been restructured. In particular, artificial intelligence–based applications have begun to be used as supportive "digital guides" that assist students in inquiry, problem-solving, and knowledge construction processes. How this digital guide functions within the Argumentation-Based Science Learning (ABSL) approach, which supports students in producing evidence-based scientific knowledge, constitutes the starting point of this study. Accordingly, the study aims to examine in depth how both artificial intelligence–supported ABSL applications and web-supported ABSL applications affect students' scientific argumentation skills, digital literacy levels, and views on the nature of science. For this purpose, a mixed-methods research design, in which quantitative and qualitative data collection techniques are used together, was adopted and supported by a strong experimental design, namely the Solomon four-group design. The study was conducted with four classes, two of which received instruction through ChatGPT-supported ABSL applications, while the other two were taught using web-supported ABSL applications. The research was carried out with 6th-grade students attending a public middle school in the Gazipaşa district of Antalya province. Quantitative data were collected using the Views of Nature of Science Questionnaire (VNOS-E), the Digital Literacy Scale, and the Scientific Argumentation Test. Qualitative data were obtained through semi-structured interviews, ChatGPT interaction records, and student portfolios. In the analysis of quantitative data, assumptions of normality and homogeneity of variance were first tested. Parametric tests were applied when assumptions were met, whereas non-parametric tests were used otherwise. In line with the Solomon four-group design, one-way analysis of variance (ANOVA) and, when necessary, post-hoc tests were employed for between-group comparisons. Appropriate tests for independent and dependent samples were used in pre-test–post-test comparisons, and two-factor and robust ANOVA analyses were conducted to control for pre-test effects. In the analysis of qualitative data, a content analysis approach was adopted. The qualitative results obtained were interpreted in relation to the quantitative findings within the integrity of the mixed-methods design. According to the results of the study, students' argumentation skills improved in both learning environments. The quantitative results indicate that both AI-supported and web-supported IBSE (Argumentation-Based Science Education) practices raised students' argumentation levels to similarly high levels. Important insights regarding the nature of this improvement were provided by the qualitative data. The qualitative findings revealed that both learning environments supported argumentation from different dimensions. In the ChatGPT-supported groups, processes such as engaging in cognitive inquiry, establishing cause–effect relationships, and restructuring ideas came to the forefront, whereas in the web-supported groups, the use of evidence, development of counter-arguments, and evidence-based discussion processes were more prominent. With regard to digital literacy, the qualitative findings indicate that both learning environments contributed to the development of students' digital literacy skills. However, the results also showed that the web-supported groups performed more strongly particularly in source verification, critical evaluation, comparison of information from multiple sources, and adherence to digital ethical principles. Consistent with these results, the quantitative analyses revealed that the digital literacy scores of the web-supported groups were significantly higher than those of the AI-supported groups (Mweb_{web}web=4.14; MAI_{AI}AI=3.86), with a medium effect size (d = –0.68). On the other hand, the qualitative results indicated that in the AI-supported groups, skills such as rapid access to information, simplification, restructuring, and transformation of information into presentation formats were more dominant. The results regarding the nature of science (NOS) show that both learning environments improved students' understanding of the nature of science. However, the findings also reveal that the AI-supported IBSE approach significantly enhanced students' awareness of the experimental, changing, subjective, and creative aspects of science compared to the web-supported learning environment. Robust ANOVA analyses indicated that the type of resource used had a significant effect on all NOS sub-dimensions with medium-level effect sizes. This demonstrates that AI-supported learning contributes to students' perception of science as a process that is open to discussion, interpretable, and continuously reshaped over time. Overall, the findings indicate that the artificial intelligence–supported ABSL approach enhances cognitive depth in students' scientific thinking and argumentation processes. Nevertheless, web-supported learning environments appear to foster stronger development in students' digital literacy skills, particularly in critical awareness, information verification, and source selection.