Developing deep learning in science through an adaptive AI-based STEM instructional program

Evidence from sixth-grade classrooms

Bibliographic Data

ID22432363
AuthorsTahani Bin Bakheet (0009-0009-8383-3721, King Saud University), Hamdan Alamri (0000-0001-8350-4383, King Saud University, corresponding author), Fahad Alshaya (0009-0004-2626-1767, King Saud University)
Year2026
Volume13
Issue1
Publication date2026-06-28
Open AccessYes
TypeARTICLE
VenueInternational Journal of STEM Education (JOURNAL)
Journal identifiersISSN: 2196-7822 • E-ISSN: 2196-7822
PublisherSpringer Science and Business Media LLC (PUBLISHER)
DOI10.1186/s40594-026-00630-1
OpenAlexW7166468625
LanguageEN
References cited54

This study examines the effectiveness of an adaptive artificial intelligence–based instructional program grounded in the STEM approach in developing deep learning (defined here as students’ deep learning rather than machine learning algorithms) in science among sixth-grade students within a female-only primary school context. Using a concurrent mixed-methods design, the study integrated quantitative and qualitative evidence to evaluate how adaptive AI-driven learning pathways influenced students’ explanation, interpretation, application, and idea-generation skills. A cluster-randomized posttest-only control group design was implemented with 30 students assigned at the intact classroom level (one classroom per condition); therefore, findings are best interpreted as preliminary, within-sample evidence consistent with a program effect, and statistical estimates are treated as exploratory at the student level. The adaptive AI-based STEM program—designed to personalize content, adjust task difficulty, and provide real-time feedback—was implemented over eight weeks, with adaptive learning pathways governed by rule-based mastery mechanisms and machine-learning components used solely for performance monitoring rather than direct modification of learning trajectories. Quantitative results revealed statistically significant differences, based on median (IQR) summaries and Mann–Whitney U tests, favoring the experimental group across all deep learning dimensions, including explanation, interpretation, application, and idea generation. Large within-sample effect sizes indicated substantial rank separation within this sample, although classroom-level confounding and baseline differences cannot be fully ruled out in a two-cluster design. Qualitative findings from interviews with three science teachers corroborated these results, highlighting perceived improvements in analytical reasoning, conceptual integration, inquiry-based exploration, and creative scientific thinking. Teachers also reported that simulations, adaptive feedback, and hands-on STEM activities supported students’ conceptual understanding and engagement with scientific problem solving. Collectively, the findings provide context-specific pilot evidence suggesting that adaptive AI-based STEM instructional models may support deep learning in elementary science. The integration of adaptive AI mechanisms with inquiry-oriented STEM learning appears to promote deeper conceptual explanation, scientific interpretation, real-world application, and idea generation. However, because the study was conducted within a single female-only school using two intact classrooms, the findings should be interpreted cautiously and validated through larger, multi-site studies with stronger cluster-level experimental controls.

Adaptive Learning · Deep learning · Educational technology · Electronic learning · Instructional design · Science education · Science learning · Teaching method · Intelligent Tutoring Systems and Adaptive Learning · Neuroscience, Education and Cognitive Function · Online Learning and Analytics

Tools

Open DOIOpen Access