Abstract
Abstract
Artificial Intelligence offers innovative potential for personalised learning, yet its integration remains hindered by instructional voids and restrictive mobile-device policies globally. This study addresses these challenges by investigating AI-powered chatbots as a scalable scaffolding tool for science learning in resource-constrained settings where a scarcity of personalised instructional resources presents a significant barrier to science outcomes. Employing a pre-experimental single-group pre-test/post-test design, the research assessed changes in cognitive outcomes and holistic competencies among 31 pre-Senior High School students in Ghana using researcher-developed, validated instruments. The intervention integrated the chatbot as a self-directed learning and inquiry support system over four weeks. Qualitative data were analysed thematically and subsequently quantified. Quantitative data were analysed using paired t-tests and the robust Wilcoxon signed-rank test, supplemented by Spearman’s rank-order correlation and Least Absolute Deviations (LAD) regression. Findings revealed statistically significant improvements across the measured domains, with large effect sizes observed for cognitive outcomes (Cohen’s d > 2.0) and moderate-to-large effect sizes for holistic outcomes (Wilcoxon r = 0.41–0.68). The findings suggest that the chatbot functioned as a More Knowledgeable Other (MKO) within the structured learning intervention and may have supported cognitive and self-directed learning processes, consistent with Cognitive Load Theory (CLT) and Self-Determination Theory (SDT) interpretations. Critically, the study found a strong positive correlation (ρ > 0.8) between students’ learning outcomes and their self-reported Perception of Utility. This research provides preliminary evidence on the feasibility and potential value of AI-supported learning for expanding personalised learning opportunities in resource-constrained educational contexts.