From Potential to Reality: Gaps in Realizing Human–AI Complementarity in Educational Tasks

Abstract

Human–artificial intelligence collaboration (HAIC) is emerging as a transformative approach in education, yet empirical evidence regarding its impact on educational task performance remains limited. This study adopts a complementarity perspective and distinguishes three types of HAIC outcomes: inherent complementarity, collaborative complementarity, and collaboration counter-productivity. An experiment was conducted to examine pre-service teachers’ task performance across conditions. Specifically, we measured complementarity potential, complementarity rates, and the functional roles of artificial intelligence (AI) participation in HAIC. The results showed that HAIC improved pre-service teachers’ immediate task performance, but this performance advantage was not sustained in the post-test. Inherent complementarity was realized substantially more often than collaborative complementarity. By contrast, collaborative complementarity was rarely realized: when both humans and AI initially made incorrect judgments, HAIC seldom produced a correct final judgment. In addition, AI roles in two-way collaboration were highly concentrated in assistive functions, including elaboration, knowledge supplementation, evaluation, feedback provision, and questioning.