An empirical framework for assessing AI in early childhood education: a global versus local perspective

Abstract The rapid integration of generative artificial intelligence into education necessitates a rigorous evaluation of its effectiveness, particularly in culturally and developmentally sensitive contexts. This study addresses this need by developing and applying a novel evaluation framework for early childhood education (ECE). We conducted a comparative analysis of four leading AI models—two global (ChatGPT and Gemini) and two prominent local Chinese (Doubao and DeepSeek) models—using seven theory-grounded prompts within a Chinese ECE context. Through blind expert ratings and statistical analyses (ANOVA, MANOVA), we found that local Chinese models significantly outperformed global models on tasks requiring cultural nuance and pedagogical alignment, with effect sizes (Cohen’s d) reaching as high as 14.63. These findings highlight the critical limitations of a one-size-fits-all approach and argue that deep localization is essential for developing effective and equitable AI-driven educational tools.