人工智能在应急管理中的应用及其风险分类——基于文献与案例的综述研究

Application of Artificial Intelligence in Emergency Management and Its Risk Categories:An Overview Study Based on Literature and Cases

随着新型工业化发展稳步推进,人工智能(AI)在应急管理的不同阶段和不同灾害情境下的应用日益增多,风险也呈现新的特征和态势,但对AI赋能应急管理的应用以及存在的各类风险仍缺乏系统性梳理。为了提升AI赋能应急管理的效率与效果,亟待厘清AI在应急管理中的具体应用及其风险类别。因此,研究结合相关文献与具体案例,对AI在应急管理中的应用与潜在风险进行文献回顾。综述发现,AI广泛应用于风险管理、应急准备、应急响应和应急恢复等四个时期,涵盖风险预警、模拟预测、心理辅导等十六类具体应用,适用于洪灾、地震等多种情境。同时,基于“人—AI—人与AI交互”等三个维度,整合AI应用于应急管理面临的个人隐私与数据安全、数据偏差与算法歧视、交互稳健程度等九类风险挑战,并由人与AI交互维度的系统与算法、价值与伦理、公众可接受度、组织可接受度等问题对AI赋能应急管理进行系统性风险分类。

With the steady development of new industrialization,the application of artificial intelligence(AI)technology in different stages of emergency management and different disaster situations is increasing,and its risks present new features and trends.However,there is still a lack of systematic study of the application of AI empowered emergency management and various types of risks.In order to improve the efficiency and effect of AI technology empowering emergency management,it is urgent to clarify the specific applications of AI in emergency management and its risk categories.Therefore,in combination with relevant literature and specific cases,the paper reviews the applications and potential risks of AI technology in emergency management.The review finds that AI technology is widely used in four periods:risk management,emergency preparedness,emergency response,and emergency recovery,cove ring 16 categories of specific applications such as risk early warning,simulation and prediction,psychological counseling,etc.,which are applicable to a variety of scenarios such as floods and earthquakes.At the same time,based on the three dimensions of“human-AI-human AI interaction”,the paper summarizes nine types of risks and challenges faced by AI technology applied to emergency management,such as personal privacy and data security,data bias and algorithmic discrimination,and the degree of robustness of interaction,etc.,and classifies the systemic risks of AI ap-plication in risk management in terms of the systems and algorithms,values and ethics,public acceptability,organizational acceptability and other factors from the dimension of human-AI interaction.