A Generative AI-Driven Framework for Human–AI Collaborative Teaching: Design, Implementation, and Empirical Evaluation

Abstract

The rapid development of Generative Artificial Intelligence (GAI) technologies is driving profound changes in education. This study investigates the construction of human–Artificial Intelligence (AI) collaborative teaching models supported by GAI, with the goals of improving teaching efficiency, enabling personalized learning, and optimizing the allocation of educational resources. The proposed framework integrated intelligent content generation, real-time tutoring, adaptive learning pathways, and a teacher–student–AI collaboration mechanism. The generative model employed was LLaMA-2-13B, which was domain-adapted through supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF). The experiment was conducted in a university course titled Data Structures and Algorithms, involving 60 students in the experimental group and 60 students in the control group. Multi-dimensional data were collected and analyzed, including academic performance, student engagement, interaction depth, technology acceptance, and long-term retention. The quality of AI-generated content was evaluated using Bilingual Evaluation Understudy (BLEU, 0.74) and Recall-Oriented Understudy for Gisting Evaluation (ROUGE, 0.79), with a Cohen’s κ value of 0.86 indicating high inter-rater consistency. The results showed that the GAI-driven human–AI collaborative model significantly improved final exam scores (85.6 vs. 78.3, p < 0.001), average assignment grades (91.2 vs. 84.7, p < 0.001), and learning satisfaction (p < 0.05), while reducing cognitive load and enhancing personalized and interactive learning. This study provides both a theoretical framework and practical guidance for innovating educational models in the era of intelligent technology, offering valuable insights for advancing the digital transformation of education.