可解释教育人工智能研究:系统框架、应用价值与案例分析  

Research on Explainable Artificial Intelligence for Education:System Framework,Application Value and Case Analysis

作　　者：王萍 田小勇 孙侨羽 Wang Ping;Tian Xiaoyong;Sun Qiaoyu(School of Education,Shanghai International Studies University,Shanghai 201620)

机构地区：[1]上海外国语大学国际教育学院,上海201620

出　　处：《远程教育杂志》2021年第6期20-29,共10页Journal of Distance Education

基　　金：上海市教育科学研究项目“人工智能环境下翻译教学的影响因素与提升策略研究”(项目编号:C2021311)研究成果之一。

摘　　要：人工智能正在加速教育创新,重塑未来教育新生态。尽管当前教育人工智能在大数据背景下体现出了积极价值,但以深度学习为代表的智能技术存在的黑箱问题,使得智能教育系统的决策结果缺乏透明度和可解释性。因此,智能教育的有效性因缺乏解释能力和因果机制而受到限制和质疑。可解释教育人工智能(XAI)通过有意义的交互,为使用者提供智能系统做出决策的缘由与解释,使教育系统可理解、可信任、可管理。在系统框架上,构建包括教育需求分析、教育数据与特征解释、教育模型解释、可解释界面等模块的可解释教育系统模型,旨在应用价值层面上探索可解释人工智能对因果推断教育研究的新路径,分析其对深层次知识发现的推进机制,构建助力教育人工智能治理的框架。并且在教育实践中的案例表明,可解释教育人工智能在智能导师系统、学习推荐系统、学习分析系统中的应用,验证了其能够有效提升教与学的效果。未来对可解释人工智能的研究,可从赋能决策支持、融合知识图谱、评估解释方法、加强跨学科合作、编制指南规范等维度,加以实施和完善。Artificial intelligence is accelerating educational innovation and reshaping the new ecology of future education.It has embodied positive value based on big data.But the black box problem of intelligent technology represented by deep learning makes the decision results of intelligent education system lack transparency and interpretability.So,the effectiveness of intelligent education is limited and questioned due to the lack of interpretability and causal mechanism.The explainable artificial intelligence provides users with the reason and interpretability of intelligent system decision-making through meaningful interaction,so that the educational system can be understood,trusted and managed.In the system framework,explainable artificial intelligence system is built by educational data and feature explanation,educational model explanation,and explainable interface.In terms of application value,explainable artificial intelligence will explore new path of the educational research on causal inference,promote deep knowledge discovery and construct the governance framework of educational artificial intelligence.In educational practice,explainable artificial intelligence has proved to effectively promote the effect of teaching and learning in intelligent tutor system,learning recommendation system and learning analysis system.The future research can be carried out from the aspects of enabling decision support,integration of knowledge graph,evaluation of explanation methods,strengthening interdisciplinary cooperation,compiling of guidelines.

关 键 词：教育人工智能 可解释人工智能 可解释性 机器学习 深度学习 因果推断 可信人工智能 

分 类 号：G420[文化科学—课程与教学论]