AI-driven intelligent tutoring systems (ITSs) generally show positive effects on K-12 students' learning and performance, but these effects are often mitigated when compared to non-intelligent tutoring systems.

Objective: This systematic review aimed to identify the effects of intelligent tutoring systems (ITSs) on K-12 students' learning and performance, and to examine which experimental designs are currently used to evaluate these systems.

Methods: The researchers conducted a systematic review following PRISMA guidelines, searching two databases (ERIC and Scopus) for peer-reviewed empirical studies published between 2009 and January 2025 that focused on K-12 students using ITSs in formal school contexts. From an initial 948 records, 26 articles containing 28 studies were included in the final analysis, involving a total of 4,597 students. The studies were categorized based on their experimental design: ITS vs. teacher, ITS vs. non-intelligent tutoring system, ITS vs. modified ITS, and ITS with no control group. The review examined various elements including authors' affiliations, publication dates, intervention durations, school levels, subjects, and experimental designs.

Key Findings:

• Most studies (54%) originated from the United States, with Asia (27%) being the second most common region.
• Studies primarily involved middle and high school students, with very few focusing on elementary students.
• The majority of studies (82%) examined ITSs in STEM subjects, with mathematics being the most common.
• Half of the interventions lasted less than a week, with some as brief as a single class period, raising questions about potential novelty effects.
• When comparing ITSs to traditional teacher-led instruction, seven out of eight studies reported significant positive effects favoring ITSs.
• When comparing ITSs to non-intelligent tutoring systems, results were more mixed: only one of five studies showed clear advantages for the ITS.
• Studies comparing different versions of ITSs or examining ITSs without control groups identified that personalization, adaptivity, immediate feedback, and self-regulation support were key factors in ITS effectiveness.
• Overall, the findings suggest that the effects of ITSs on learning and performance in K-12 education are generally positive but are mitigated when compared to non-intelligent tutoring systems.
• None of the studies addressed ethical implications of using AI in education.

Implications: The review highlights that ITSs can be effective educational tools when they include core components like personalization, adaptivity, and immediate feedback. The research suggests that ITSs should be leveraged to supplement rather than replace classroom instruction, providing additional learning support in parallel with teacher guidance. The study also reinforces that AI and teachers can collaborate effectively to optimize and facilitate student learning. The findings indicate that well-designed ITSs may help address global educational challenges, including teacher shortages and the need for personalized learning experiences.

Limitations: The review identified several limitations in the current research on ITSs:

• The short duration of many interventions (half lasting less than a week) makes it difficult to determine whether positive effects are due to the ITS or simply to novelty.
• Sample sizes varied significantly across studies, and there was limited research involving elementary school students.
• There was an overrepresentation of STEM subjects and a lack of research in other areas of the curriculum.
• Most studies used quasi-experimental designs with varying control groups, making direct comparisons difficult.
• Publication and reporting bias may have influenced the results, as studies with positive outcomes are more likely to be published.
• The review was limited to articles in English from only two databases.

Future Directions: The authors suggest several avenues for future research:

• Additional research with longer interventions and increased sample sizes with greater diversity is warranted.
• More studies should investigate the effects of ITSs in elementary school settings.
• Research should expand beyond STEM subjects to examine ITS effectiveness in other curriculum areas.
• The ethical implications of using AI for teaching should be investigated, as none of the reviewed studies addressed this aspect.
• Given the rapid advancement of generative AI, future research should explore how newer AI technologies can be integrated into ITSs.
• Researchers should consider the novelty effect when designing studies and implement longer interventions to better assess the true impact of ITSs.

Title and Authors: "A systematic review of AI-driven intelligent tutoring systems (ITS) in K-12 education" by Angélique Létourneau, Marion Deslandes Martineau, Patrick Charland, John Alexander Karran, Jared Boasen, and Pierre Majorique Léger.

Published On: May 14, 2025

Published By: npj Science of Learning, in partnership with The University of Queensland