余胜泉,熊莎莎.基于大模型增强的通用人工智能教师架构[J].开放教育研究,2024,30(01):33-43.
Yu Shengquan, Xiong Shasha. The Architecture of General Artificial Intelligence Teachers Enhanced by Large Models [J]. Open Education Research, 2024, 30(01): 33 - 43.
题名：基于大模型增强的通用人工智能教师架构
Title: The Architecture of General Artificial Intelligence Teachers Enhanced by Large Models
作者：余胜泉,熊莎莎
Authors: Yu Shengquan, Xiong Shasha.
关键词：人工智能教师;生成式人工智能;预训练大模型; 人工智能教育应用;
Keywords: Artificial intelligence teachers; Generative artificial intelligence; Pretrained large models; Applications of artificial intelligence in education.
摘要：以ChatGPT为代表的预训练大模型在教育界产生了巨大影响，也为发展通用人工智能教师带来了曙光。预训练大模型应用于教育存在幻觉、深度逻辑缺失、社会情感缺失等局限，如果这些关键问题不解决，大模型不可能在教育中得到真正有价值的应用。本文提出通过增强大模型构建通用人工智能教师架构，其核心思路是精调训练增强场景知识、检索增强认知、外部智能组件编排增强推理、多模态融合增强感知、情感计算增强社会情感，再通过教育知识图谱对大模型输出进行监督。通用人工智能教师主要有六种应用场景：需要渊博知识的场景、洞察创意增强场景、约束与管理场景、社会情感互动场景、个性化指导与反馈场景、多模态内容表现场景。文章最后分析了通用智能时代人类教师面临的机遇与挑战，提出教师需要主动拥抱、使用、驾驭并控制智能技术，解放、增强、进化自我，并跨越陷阱。
Abstract: Pretrained large models represented by ChatGPT have had a huge impact on the education field and also brought hope for the development of general artificial intelligence teachers. However, when applied in education, pretrained large models have limitations such as hallucination, lack of in-depth logic, and lack of social emotions. If these key issues are not resolved, it will be impossible for large models to have truly valuable applications in education.
This paper proposes to construct the architecture of general artificial intelligence teachers by enhancing large models. Its core ideas are as follows: fine-tuning training to enhance scene knowledge, retrieval to enhance cognition, orchestration of external intelligent components to enhance reasoning, multimodal fusion to enhance perception, and emotional computing to enhance social emotions. Then, the output of large models is supervised through the educational knowledge graph.
General artificial intelligence teachers mainly have six application scenarios: scenarios requiring profound knowledge, scenarios for enhancing insight and creativity, constraint and management scenarios, social-emotional interaction scenarios, personalized guidance and feedback scenarios, and multimodal content presentation scenarios.
Finally, the article analyzes the opportunities and challenges faced by human teachers in the era of general intelligence, and proposes that teachers need to actively embrace, use, master, and control intelligent technologies, liberate, enhance, and evolve themselves, and avoid falling into traps.