人工智能使用过程中自动化偏差文献述评与研究展望

Literature Review and Research Prospects on Automation Bias in the Use of Artificial Intelligence

摘要:

人工智能（AI）辅助决策给管理者带来决策效能的提高，但当AI出现错误，决策者如不加以验证AI建议准确性就采纳其建议，将会导致严重错误。为此，对人类使用AI过程中的自动化偏差的概念、行为规律、作用机制、影响要素等进行系统述评，发现自动化偏差具有跨领域发展的趋势，并归纳出自动化偏差的不同特点，认为可以从系统、环境、组织和个体等4个方面归纳影响自动化偏差的主要因素，分别涉及包括智能系统能否提供预测性信息以及系统能否即时反馈错误，决策支持系统设计、任务难度、工作量、任务复杂性、时间限制、对整体表现或决策准确性负责、信任和信心、知识水平的高低、能力等。未来应加强对企业管理者受AI自动化偏差影响的实证研究，同时要综合系统、环境、组织和个体要素揭示自动化偏差的形成机制，并制定出有效的去偏方案，帮助决策者正确利用AI辅助决策。

Abstract:

Artificial intelligence (AI) assisted decision-making enhances decision-making effectiveness for managers. However, when AI makes errors, if decision-makers adopt its suggestions without verifying their accuracy, it can lead to serious mistakes. This study employs a literature review method to systematically review the concept, behavioral patterns, mechanisms, and influencing factors of automation bias in human use of AI.It identifies a trend of cross-domain development in automation bias and summarizes its characteristics, which include excessive reliance, reduced vigilance, neglect of verification, and individual differences.The main factors influencing automation bias can be summarized from four aspects: system, environment, organization, and individual. These factors include whether intelligent systems can provide predictive information and whether the system can provide immediate feedback on errors, as well as the design of decision support systems, task difficulty, workload, task complexity, time constraints, accountability for overall performance or decision accuracy, trust and confidence, knowledge level, and capabilities. Future research should strengthen empirical studies on the impact of AI automation bias on corporate managers, while also integrating system, environment, organization, and individual factors to reveal the formation mechanisms of automation bias. Additionally, effective debiasing strategies should be developed to assist decision-makers in correctly utilizing AI for decision-making.