On the Effect of Explainable AI on Programming Learning: A Case Study of using Gradient Integration Technology


Feng Hsu Wang, Ming Chuan University, Taiwan


AI-based learning technologies, especially deep learning, hold significant promise for enhancing students' learning experiences in educational systems. However, providing accurate predictions or answers to students' learning problems through high-performance deep learning models is not sufficient for students to achieve effective learning. This study explores Explainable Artificial Intelligence (XAI) in reducing students' cognitive load and improving learning outcomes within the realm of object-oriented programming education. Specifically, this study examines the application of Gradient Integration to generate coloured code segments associated with code errors predicted by a Performer-based deep learning classification model for debugging tasks. Thirty-six participants took part in a controlled experiment assessing students' cognitive load and learning performance through the XAI system. They were randomly assigned to a control group (N=18) and an experiment group (N=18). The independent-samples Wilcoxon-Mann-Whitney test results revealed that the coloured code segments reduce students' cognitive load (p=0.006) and improve their exam scores (p=0.006) significantly. This study contributes to an appropriate application of the XAI technique that can reduce students' cognitive load and improve learning outcomes in educational settings.


Explainable Artificial Intelligence, Deep Learning Technology, Human-Computer Collaborative Learning, Programming Education

Full Text  Volume 14, Number 12