AI Use Quality and Sustainable Educational Equity: Evidence on the Socioeconomic Gap in Deep Learning Approach Among Chinese High School Students

Sustainable educational equity, the principle behind United Nations Sustainable Development Goal 4, calls for ensuring that disadvantaged students benefit from emerging educational technologies rather than being pushed further behind. As AI learning tools become routine in secondary schools, whether they reduce or widen socioeconomic gaps in learning has become a pressing question for sustainable educational policy. Building on digital divide theory and the resource substitution hypothesis, we tested whether family socioeconomic status (SES) moderates the link between students’ AI use quality and deep learning approach—specifically, whether high-quality AI use is more strongly associated with deep learning approach for students from lower-SES backgrounds. Data came from 548 students at three public high schools in Hangzhou, China. AI use quality was operationalized as a three-part construct (seeking, evaluating, applying). Deep learning approach was measured with the deep approach subscale of the R-SPQ-2F. We tested moderation with hierarchical regression and probed the interaction with simple slopes. Two results stood out. First, both SES and AI use quality positively predicted deep learning approach. Family SES moderated the association between AI use quality and deep learning approach: the link between AI use quality and deep learning approach was stronger for low-SES students than for their higher-SES peers, and when AI use quality was high, the deep learning gap across SES levels was correspondingly narrower. The data support the equalizer hypothesis: high-quality AI use can narrow the SES-related gap in deep learning approach and serve as a lever for sustainable educational equity. Schools that want AI to advance equity should treat AI literacy as an instructional priority across subjects, not as something students are expected to figure out on their own.