Which Teachers Adopt AI in Teaching? An Interpretable Machine Learning Analysis of East Asian Data from TALIS 2024

Abstract

Against the backdrop of generative AI reshaping the global education ecosystem, identifying predictors of teachers’ artificial intelligence (AI) adoption has become an important issue. Most existing studies focus on single variables or traditional acceptance models, this study adopts a broader perspective, aiming to simultaneously examine multiple potential predictors. Guided by socio-ecological theory, this study employs machine learning methods (LASSO and AutoML) to predict and analyze factors associated with teachers' AI adoption across 25 factors at the individual, microsystem, and mesosystem levels in four East Asian economies (Singapore, Shanghai (China), Korea, and Japan). Results indicate that, in the overall East Asian sample, teaching self-efficacy, age, and perceived risk were the top three predictors; higher teaching self-efficacy was generally associated with a greater likelihood of teachers adopting AI, whereas higher age categories and perceived risk were associated with a relatively lower likelihood. Economy-specific results showed some variation: the top predictor in Singapore was workload stress, with higher workload stress generally associated with a greater likelihood of AI adoption; in Shanghai (China) and Korea, the top predictor was teacher professional collaboration, with higher collaboration levels generally associated with a greater likelihood of AI adoption, though their second-ranked predictors were teaching experience and age, respectively; in Japan, the top predictor was age, with the likelihood of AI adoption generally decreasing as age categories increased. These findings suggest that teachers’ AI adoption can be advanced by strengthening teaching self-efficacy at the individual level through practice-oriented training to strengthen teachers’ confidence; at the microsystem level, leveraging teacher professional collaboration to promote experience sharing and provide practical support through professional learning communities, peer modeling, and collaborative case development; and at the mesosystem level, further translating institutional support into accessible resources and technical support for teachers. Additionally, different economies can provide differentiated support tailored to teachers’ age and experience.