Enhancing Early Childhood Education Assessment Through Human-AI Collaboration: A Mixed-Methods Study of DeepSeek-Supported Assessment

Early childhood assessment is critical to developmental monitoring, yet traditional methods suffer from subjectivity, inefficiency, and limited actionable insights. We evaluated a Human-AI Collaboration (HAIC) model which integrates the AI (DeepSeek) with teachers’ contextual expertise in early childhood education assessment. In a mixed-methods design, 120 paired assessment reports of 5–6-year-olds’ regional learning activities were analyzed—60 using teacher-independent methods (TIM) and 60 via HAIC. Wilcoxon signed-rank tests indicated HAIC increased the use of high-leverage strategies, like evidence-based judgment (r = 0.875). Furthermore, HAIC improved assessment quality overall, and multiple regression confirmed HAIC-generated strategies positively predicted quality ratings. However, the explanatory power of HAIC was lower than that of TIM for scientific objectivity (R2: 0.810 vs. 0.932) and developmental focus (R2: 0.550 vs. 0.861), yet higher for comprehensiveness (R2: 0.795 vs. 0.309). Semi-structured interviews and interaction logs confirmed HAIC reduced assessment time and enhanced standardization, but generated nonspecific positive reinforcement and theory-practice discrepancies requiring expert teacher correction. We concluded that optimal task allocation delegates data-intensive functions to AI while reserving contextual validation for teachers. Sustainable implementation requires dual-competency professional development that combines pedagogical expertise with AI literacy, thereby enhancing the synergistic effect to strengthen developmental sensitivity and pedagogical authenticity.