摘要
智能体一直是人工智能的主要研究领域之一,任何独立的能够同环境交互并自主决策的实体都可以抽象为智能体.随着人工智能从计算智能到感知智能,再到认知智能的发展,智能体已逐步渗透到无人驾驶、服务机器人、智能家居、智慧医疗、战争武器等人类生活密切相关的领域.这些应用中,智能体与环境、尤其是与人类和社会的交互愈来愈突出,其中的伦理和道德问题日益凸显.人工智能应用的伦理风险和挑战引起了人们的普遍关注,伦理智能体是人工智能伦理的重要研究内容.本文从人工智能伦理的工程设计与实现角度,对伦理智能体及图灵测试、伦理智能体的设计范式、伦理智能体的逻辑程序设计、伦理智能体的形式化验证、伦理困境及分析等进行了介绍和讨论.同时,对伦理智能体及设计所面临的挑战和进一步研究方向进行了述评和展望.
Artificial agents have always been one of the main research fields of artificial intelligence.Any independent entity that can interact with the environment and make decisions autonomously can be abstracted as an agent.With the development of artificial intelligence from computational intelligence to perceptual intelligence,and then to cognitive intelligence,artificial agents have gradually penetrated into many fields closely related to our human life,such as unmanned driving cars,service robots,smart households,voice assistants,intelligent medical care and war weapons.In these applications,the interactions between agents and among agents and the environment,especially humans and society,are becoming more and more prominent,and the interactions of agents with humans and society inevitably resulting in ethical threats and moral attentions.Issues regarding artificial moral agents are important research contents of ethical artificial intelligence.The ethical risks and challenges in the application of artificial intelligence have aroused more and more public concerns.The ethical issues of artificial intelligence belong to the interdisciplinary research of social science,philosophy,psychology,cognitive science,computer,artificial intelligence,etc.On the one hand,philosophy and humanistic social scientists focus on the moral subjects of artificial intelligence,the effects of artificial intelligence on the human society,the ethical norms and moral codes of artificial intelligence and so on.On the other hand,researchers in the engineering fields such as computer,cognitive science and artificial intelligence are interested in the embedded ethical decision design,the implementation of moral artificial intelligence,the ethical supervision and control of artificial intelligence and others.In the framework of ethically aligned design,realization and deployment of moral artificial intelligence,this paper reviews the design methodology of artificial moral agents in the following aspects.Firstly,the concepts of artificial moral agents are introduced,and in terms of multiple behavioral characteristics,such as autonomy,adaptation,evolution and consciousness,the categories of artificial moral agents are illustrated.Although there is still controversial on the effectiveness of the Turing test,as a promising way to identify and evaluate moral artificial agents,the moral Turing test is addressed briefly.Ethical norms and moral codes are the premises for the design and implementation of artificial moral agents,and many government agencies,industry associations,companies and international organizations have put forward a series of moral codes and ethical norms to manage and supervise artificial intelligence.The open problems,such as how to properly choose and strictly abide by them,and how to translate them to machine or computer codes are discussed,etc.Secondly,the three design paradigms of artificial moral agents,including top-down,bottom-up and hybrid approach are analyzed,and the design of moral artificial agents via logical programming is elaborated,which integrates two different concepts of both logic and programming.Formal verification and validation of artificial moral agents are also outlined.Finally,the ethical dilemma is a common puzzle of philosophy,social science and engineering disciplines,and the potential solution using computational ethics and crowdsourcing technology to defend it is described.More importantly,a comprehensive outlook of the challenges and the further research directions of moral artificial agents is presented in the paper.
作者
古天龙
李龙
GU Tian-Long;LI Long(College of Information Science and Technology/College of Cyber Security,Jinan University,Guangzhou 510632;Guangxi Key Laboratory of Trusted Software,Guilin University of Electronic Technology,Guilin,Guangxi541004)
出处
《计算机学报》
EI
CSCD
北大核心
2021年第3期632-651,共20页
Chinese Journal of Computers
基金
国家自然科学基金(U1711263,U1811264,61966009)
广西自然科学基金(2019GXNSFBA245049)资助。
关键词
伦理智能体
人工智能伦理
伦理设计
逻辑程序设计
形式化验证
artificial moral agents
artificial intelligence ethics
ethically aligned design
logic programming
formal verification
