日省吾身，维新至善——任务开放程度与元认知对人与AI 协作绩效的影响

Abstract

As AI is adopted widely across domains, people expect human–AI collaboration to deliver sustained performance gains. However, many individuals and organizations fall short after introducing AI; performance often stagnates or even declines. In practice, some people achieve substantial improvements with AI, while others stagnate due to overreliance on it or a lack of proactive adaptation. Accordingly, this study asks: Why do some individuals fail to achieve sustained improvements in collaborative performance when working in a humanin-the-loop” mode, despite AI's powerful capabilities?
To address this question, we introduce two variables—task openness and metacognition—to build a theoretical framework. We hypothesize that the degree of task openness determines the potential space for performance improvement in collaboration. In contrast, the level of metacognition determines whether individuals can exploit that space to realize continuous improvement. In other words, only when tasks are sufficiently open and individuals possess strong metacognitive ability can human–AI collaboration fully leverage the strengths of both parties and achieve sustained performance gains.
Methodologically, we combine theoretical deduction with a multi-case study for corroboration. Drawing on dynamic capability theory, self-regulation, co-regulation, and shared regulation, we construct a conceptual model that explicates the mechanisms through which task openness and metacognition influence performance improvement. We then conduct a multi-case analysis of representative roles across several organizations, classifying cases by task openness and comparing human–AI collaborative performance at different metacognitive levels to test the proposed hypotheses.
Our findings show that in low-openness tasks, AI’s substitution effect confines performance improvements to predefined metrics such as efficiency; most individuals’ performance ultimately converges to AI’s level, and even those with high metacognition struggle to outperform others. By contrast, in high-openness tasks, AI primarily plays a complementary and augmentative role. Driven by metacognition, individuals iteratively adjust the division of labor and collaboration strategies with AI, exploring superior approaches that continuously elevate collaborative performance. This suggests that only when the task environment is sufficiently open and individuals’ metacognitive levels are high can human–AI collaboration breakthrough performance bottlenecks and enter a virtuous cycle.
Theoretically, this study integrates task-context openness with individual metacognitive regulation to deepen understanding of the mechanisms underlying sustained improvements in human–AI collaborative performance. We develop a metacognition-centered model of dynamic collaboration processes, enriching theoretical perspectives in cognitive psychology, artificial intelligence, and organizational behavior. In practical terms, the results offer guidance for talent development and organizational management: organizations and individuals should not rely solely on AI technologies. They must shape sufficiently open task environments and cultivate people’s metacognitive and self-regulatory capabilities. Only by acting on both the task and human dimensions can we break through the ceiling on human–AI collaborative performance and foster a virtuous cycle of collaboration.

Date of Award	15 Nov 2025
Original language	Chinese (Simplified)
Awarding Institution	China Europe International Business School
Supervisor	Yi Lu (Supervisor) & Yan Gong (Supervisor)