AI in education : policy, practice, and empirical evidence from a field experiment in China

Abstract

This paper focuses on the issue of applying Artificial Intelligence (AI) technology in the education process. After reviewing and sorting out the relevant policies and industrial practices of "AI+ education" in China, this study empirically assesses the causal impact of AI-supported education technology on the academic achievement of middle school students in Math by conducting a large-scale Randomized Field Experiment across 38 schools in 18 Chinese provinces. The study finds that using iFLYTEK AI learning laptops has a significant positive effect on students' exam performance, with an increase of approximately 3.9% (1.1 points) in the score or 1.2% (10 places) in the ranking. The results are robust, having been tested for various checks and proven to be unlikely influenced by omitted variables, specification errors, or measurement errors, as evidenced by pre-trend and balance tests. The paper further reveals that the performance-enhancing effect of AI learning laptops exhibits heterogeneity favoring educational equity. Specifically, the effect is more pronounced in the west and northeast regions of China, where economic and educational development is weaker than in the eastern and middle regions. Additionally, schools located in prefecture-level cities with per capita education fiscal expenditure below the national median demonstrate a more significant effect than those above it. The effect is also more prominent for poor students ranking in the bottom third in the first exam compared to those with higher initial rankings. The study also found that students spent more time studying Math during self-study classes and more time on Math in total when using AI learning devices, compared to other subjects. Furthermore, the use of AI learning devices made students more engaged in studying Math, but it did not affect their interests in other subjects.