摘要
ChatGPT系列和Sora等大模型作为人工智能技术的发展前沿,在助力人类生产生活的数字化和智能化转型提速的同时,也引发一系列伦理失范现象,成为全球性治理难题。本文对大模型伦理失范进行理论解构发现:大模型伦理失范的缘起是其区别于传统生产力和生产工具的技术非中性、内嵌人类伦理和将人类伦理与机器伦理糅合其中等属性特征,技术起点是算法,关键中介是数据,行为主体是人类。大模型所引发的诸如模型黑箱、数据版权侵害且主体责任难确定、冲击人类主体资格等伦理失范现象对现行伦理治理模式提出挑战。因此,本文以大模型生命周期为时间维度,以关键要素为核心,构建了基于大模型关键要素的全生命周期伦理治理框架。同时,为推动新伦理治理框架的有效运行,本文构建了包含两层行为主体的自我治理、两级守门人和全球性合作治理网络等子体系的大模型伦理治理生态体系。
Large model is at the forefront of artificial intelligence(AI)technology,becoming a“universal simulator”connecting the physical world and the digital world.Large model has achieved breakthrough and disruptive technological innovation,bringing about changes in production methods,technological innovation paradigms,content generation methods,and human⁃machine relationships.At the same time,it has also triggered a series of ethical misconduct phenomena,becoming a global governance challenge.Existing literature on how to prevent and govern ethical misconduct in a large model still needs to be further studied and enriched,and there are shortcomings such as weak targeting and applicability.In addition,the existing ethical governance model is no longer able to effectively address the issue of ethical misconduct in a large model,we urgently need to propose a new ethical governance framework.This article provides a theoretical deconstruction of ethical misconduct in a large model and finds that the origin of the ethical misconduct of a large model lies in its distinguishing characteristics from traditional productive forces and production tools,such as technological non⁃neutrality,embedded human ethics,and the mixing of machine ethics and human ethics.The starting point of the ethical misconduct of a large model is the algorithm,the key intermediary is data,and the behavioral subject is human.The ethical misconduct phenomena caused by a large model,such as black⁃box models,data copyright infringement,and the impact on human subject qualifications,which poses a challenge to the current ethical governance system.To address the challenge,with the life⁃cycle of a large model as the time dimension and key elements as the core,this article constructs a full life⁃cycle ethical governance framework for key elements of a large model.At the same time,this article aims to promote the effective implementation of the new ethical governance framework,and build an ethical governance ecosystem that includes subsystems such as two levels of self⁃governance,two levels of digital gatekeepers,and a global collaborative governance network.This article expands previous literature from the following aspects.Firstly,it deconstructs the attributes and key elements of a large model in theory,which is a re⁃understanding of the essence of generative AI technologies such as a large model.Secondly,the theoretical deconstruction of the causes of ethical misconduct in a large model is conducted to further reveal the role of key elements such as algorithms,data,and humans in ethical misconduct.Thirdly,the phenomenon of ethical misconduct in a large model is summarized and extended based on the phenomenon or risk of ethical misconduct in AI,such as“model black boxes”.Fourthly,this article proposes a new governance framework for ethical misconduct in a large model.Beyond the current single factor,result⁃oriented governance model,this article constructs an ethical governance framework with key elements as the core,combined with the life⁃cycle of a large model,and constructing a corresponding ethical governance ecosystem to effectively prevent and govern ethical misconduct in a large model.
作者
肖红军
张丽丽
XIAO Hong-jun;ZHANG Li-i(Institute of Industrial Economics,Chinese Academy of Social Sciences,Beijing 100006,China;Institute of Digital Economy,Beijing Academy of Science and Technology,Beijing 100089,China)
出处
《财经问题研究》
CSSCI
北大核心
2024年第5期15-32,共18页
Research On Financial and Economic Issues
基金
国家社会科学基金重大项目“国企混合所有制改革的实现路径选择研究”(20&ZD073)
中国社会科学院国情调研重大项目“数字科技伦理监管制度调查研究”(GQZD2023009)
中国社会科学院登峰战略企业管理优势学科建设项目。
关键词
大模型
伦理失范
伦理治理
large model
theoretical deconstruction
ethical governance
