【摘要】 当前，随着以ChatGPT为代表的生成式人工智能技术的兴起，教学活动正从传统的“师-生”二元主体结构转变为“师-生-机”三元结构。在此背景下，作为新主体的AI代理应该秉持怎样的设计原则，才能够实现不输于传统“师-生”互动的有效教学？“人-人”教学中的有效教学策略又能否迁移到“人-机”教学的场景中？为回答上述问题，文章回顾了教学代理的相关研究，围绕两个方面展开研究：一是保证情感互动。在媒介层面，提升机器的社会临场感，使其能够通过拟人化的声音、表情和行为与学习者进行情感化互动；在社交层面，重视社交规范的作用，重新构建机器和学习者之间的互动规则，促进有效的教学交互。二是保证信息有效性。针对大模型的“幻觉”问题，提出结合领域知识库和技术护栏机制来提升信息的准确性；关注人机信任的平衡，避免过度信任和算法厌恶，培养学习者的批判性思维和数字素养。文章通过这一系列讨论，旨在提出构建适用于“人-机”教学场景的设计原则，为人工智能教育应用的开发设计和教学实践创新提供参考。更多还原

【Abstract】 With the emergence of generative artificial intelligence technologies exemplified by ChatGPT, the fundamental structure of educational activities is transitioning from the traditional binary “teacher-student” framework to a ternary “teacherstudent-machine” structure. What design principles should AI agents, as new actors in this structure, follow to establish effective teaching interactions that are comparable to traditional teacher-student dynamics? What differences between humans and machines pose unprecedented challenges when translating existing human-to-human teaching theories and models to humanmachine teaching scenarios? This paper addresses two critical issues: First, ensuring emotional interaction-at the media level,enhancing the machine’s social presence to enable emotional engagement with learners through humanized voice, expressions,and behaviors; at the social level, emphasizing the role of social norms and reconstructing interaction protocols between machines and learners to promote effective teaching interactions. Second, ensuring information validity-addressing the “hallucination” issue of large language models by proposing the integration of domain knowledge bases and technical guardrail mechanisms to improve information accuracy; focusing on the balance of human-machine trust to avoid both over-reliance and algorithm aversion while fostering learners’ critical thinking and digital literacy. This paper aims to propose design principles for future human-machine teaching scenarios through these discussions.更多还原