ARTIFICIAL INTELLIGENCE AS A PEDAGOGICAL PARTNER: EMPIRICAL EVIDENCE FOR ADAPTIVE TEACHING, STUDENT LEARNING, AND ASSESSMENT

Abstrak

Artificial intelligence (AI) is now being viewed as a pedagogical collaborator, as opposed to a teaching instrument, and has the potential to transform adaptive teaching, learner learning, and assessment. Although there is an increased interest in this field, there is little empirical data on how AI-based engagement affects these spheres in a large scale. The research paper fills that gap by studying the EdNet KT3 dataset which comprises over 18 million interaction records of 50,000 stratified learners. The system approach to the methodology was adopted, comprising pre-processing of the data, feature extraction, and multi-level analysis. Descriptive statistics and ANOVA tests were used to compare the differences between groups, and regression modelling and predictive comparisons were used to determine the contribution of the engagement features identified by AI. Findings indicate that learners who were highly engaged were much more accurate (M = 81.6%), faster (M = 46 seconds), and more widely mastered (average of 145 skills per learner) than learners who were lowly engaged. The addition of engagement features to predictive models significantly enhanced the performance, with the improvement in the AUROC of 0.72 to 0.87 and the enhancement of the assessment validity. On the whole, the results demonstrate the prospect of AI-assisted interaction to improve the flexibility of instruction, learning effectiveness, and the validity of learning evaluation.

Identifier

• OpenAlex: https://openalex.org/W7160669048
• DOI: https://doi.org/10.5281/zenodo.20078523