摘要
尽管人工智能(Artificial Intelligence,AI)在组织管理中的应用得到持续的关注和热切的期待,我们必须同时清醒地认识到与AI相关的各类缺陷及其潜在的负面影响。如此,我们才能更加全面系统而且准确客观地审视AI在组织管理中的应用潜能。有鉴于此,本文提出了一个总括的分析框架,简称ABCD,分别指向AI在四个关键方面的潜在缺陷:问责性(Accountability)、有限性(Boundedness)、欺骗性(Cheating)和愚蠢性(Dumbness)。在事实判断层面,有限性和愚蠢性聚焦考察AI在自身能力方面的系统性不足。在价值判断层面,问责性和欺骗性则专注于AI在法律以及伦理道德方面的若干缺陷。具体而言,第一,AI缺乏责任担当,无法被问责(Accountability)。当AI参与的重大决策失败时,AI的不可解释性使人们很难明确失败原因并难以在人机决策主体间准确地分配责任。另外,AI无法在法律、财务或心理等任一维度上接受处罚。因此,即便是AI被认定为负有责任,它最终也无法承担责任。第二,只要AI的智能仍然是基于人类的知识和专长而学习与发展的,那么其智能水平将始终是有限的(Boundedness)。这种有限性既源于阻碍AI完全捕获训练数据及其模式的那些自然或技术障碍,也源于人类主体为维护自身利益而故意制造的人为障碍。第三,正如人类主体在提供训练数据或参与AI设计和应用时可能会欺骗或操纵AI一样,AI也可能出于各种或合理或奇特的原因欺骗人类(Cheat-ing)。第四,在极端情况下,人工智能可能会完全演变成人工愚蠢(Dumbness),以错误的判断以及绝对专制的方式压抑人类智能和主动性,甚至造成致命性的错误与灾难。本文最后探讨ABCD框架对未来研究与管理实践的影响与启发。总结而言,即使我们对AI在组织管理应用中的光明前景持有总体乐观的态度,我们也必须慎重考虑其潜在的缺陷,从而能够保持必要的平衡拿捏与合理的冷静客观。
Despite the sustained attention and keen expectations surrounding the application of Artificial Intelligence(AI)in organizational management,it is imperative to recognize the dark side and potential negative implications associated with AI.Such a recognition will enable a more comprehensive,objective,and precise evaluation of AI’s full potential within the sphere of organi-zational management.Accordingly,this paper advances an overarching analytical framework,referred to as ABCD,which hinges specifically on four major aspects of the potential drawbacks and problems of AI:Accountability,Boundedness,Cheating,and Dumbness.At the factual judgment level,boundedness and dumbness examine the systematic deficiencies in AI’s capabilities.At the value judgment level,accountability and cheating scrutinize AI’s flaws in legal and ethical dimensions.Specifically,first,AI lacks accountability and could not be held accountable.Naturally,AI’s unexplanability makes it difficult for the assignment of re-sponsibilities to specific decision-makers,humans or machines,in case of major decision failures.Moreover,even when AI is found to be responsible,it will not be able to bear the ultimate responsibility as it cannot be punished legally,financially,or men-tally.Second,so long as AI develops its intelligence based on data drawn from human knowledge and expertise,it will suffer the problem of bounded intelligence,as determined by a host of natural and technological barriers that prevent AI from fully capturing the training data as well as artificial barriers created deliberately by the human actors to protect their own interests.Third,just as human actors could cheat and manipulate AI when providing training data or engaging in AI’s design and application,AI has been found to cheat humans,for whatever reasons.Finally,in the extreme case,AI could turn squarely into artificial dumbness and might potentially cause fatal errors and unfortunate disasters,as it categorically and tyrannically overrides human intelligence and suppresses human initiatives.The paper concludes with an examination of ABCD’s implications for future research and manage-ment practice.In sum,while we are sanguine about the supposedly bright future of applying AI in organizational management,we must also take into account its dark side for a more balanced and reasoned account.
作者
苏孟玥
马浩
Mengyue Su;Hao Ma(Xi’an Jiaotong-Liverpool University,HeXie Management Research Centre;Peking University,National School of Development)
出处
《管理学季刊》
2024年第4期15-29,125,共16页
Quarterly Journal of Management
关键词
问责性
有限智能
欺骗性
人工愚蠢
人工智能
accountability
bounded intelligence
cheating
artificial dumbness
artificial intelligence
