论Sora的伦理风险与我国治理因应——也谈我国参与人工智能算法伦理全球治理的基本路径
作者:
作者单位:

(同济大学法学院,上海200092)

作者简介:

刘祖兵(1988—),男,博士研究生,主要从事数据法学和知识产权法学研究。E-mail:ncliuzb@126.com

通讯作者:

中图分类号:

D92.11

基金项目:

教育部人文社会科学规划基金项目(21YJA820032);上海市2022年度软科学研究项目(22692104400)


On Sora’s Ethical Risks and China’s Governance Response: Also on the Basic Path of China’s Participation in the Global Governance of Artificial Intelligence Algorithm Ethics
Author:
Affiliation:

(Law School, Tongji University, Shanghai 200092, China)

Fund Project:

  • 摘要
  • |
  • 图/表
  • |
  • 访问统计
  • |
  • 参考文献
  • |
  • 相似文献
  • |
  • 引证文献
  • |
  • 文章评论
    摘要:

    以Sora为代表的文生视频模型触发多重伦理风险,亟须探索复归理路。Sora替代应用诱发弱化人类主体性风险,“绝对”失业潮恶化贫富差距,可能引发“新卢德运动”,影响人类社会稳定;Sora接受偏见视频数据“浸润”,刻画对特定群体或观点的刻板印象,其生成内容欠缺价值多样性与包容性,诱发高度隐蔽性的社会歧视风险,影响人类社会公平;Sora生成的虚假视频具有高度违法性风险,致使诈骗视频信息大行其道,诈骗犯罪高发,影响社会和谐。我国生成式人工智能伦理监管规范体系阙如,存在具体应用场景中的算法监管细则语焉不详、使用行为监管规则欠缺和算法伦理安全规则缺位等问题;视频数据使用者行为监管失衡,既有过度重视数据保护而阻碍数据要素丰富的宏观治理问题,也存在数据生产者行为规范缺位、部分法条可适用性不强的微观治理问题。应建立基于场景化的分类分级审查制度,纵向细化平台数据使用行为规则,进行算法听证,确保多元主体参与算法设计;从制度和技术维度强化视频数据使用者的行为监管,推动训练Sora的数据实现透明化和价值多样性。此外,我国应把握全球视野,参与生成式人工智能算法底层模型技术标准的制定与全球共享,致力于推动科技创新人才培养和域外专业人才引进;积极推动我国参与人工智能算法伦理治理的南北对话和南南合作,为人工智能全球伦理治理贡献中国智慧。

    Abstract:

    The cultural video model represented by Sora triggers multiple ethical risks and urgently needs to explore the path of rationalization. The substitution of Sora application triggers the risk of weakening human subjectivity, worsening the wealth gap due to the “absolute” unemployment wave, triggering a possible “New Ludendorff Movement”, and affecting the stability of human society. Sora accepts biased video data infiltration, depicting stereotypes about specific groups or viewpoints. The generated content lacks value diversity and inclusiveness, triggering highly covert social discrimination risks and affecting human social equity. The fake videos generated by Sora have a high risk of illegality, leading to the proliferation of fraudulent video information, high incidence of fraud crimes, and affecting social harmony. There is a lack of ethical regulatory framework for generative artificial intelligence in China, with unclear rules for algorithm regulation in specific application scenarios, insufficient rules for regulating usage behavior, and a lack of ethical safety rules for algorithms. There is an imbalance in the regulation of video data users’ behavior, which includes macro governance issues such as excessive emphasis on data protection and hindering the enrichment of data elements, as well as micro governance issues such as the lack of behavioral norms for data producers and weak applicability of some legal provisions. We should establish a scenario based classification and grading review system, vertically refine platform data usage behavior rules, conduct algorithm hearings, and ensure the participation of multiple stakeholders in algorithm design. We should strengthen the supervision of video data messenger behavior from institutional and technological dimensions, and promote the transparency and value diversity of Sora training data. We should grasp a global perspective, participate in the formulation and global sharing of technical standards for generative artificial intelligence algorithm underlying models, and strive to promote the cultivation of scientific and technological innovation talents and the introduction of professional talents from outside the field. And we should actively promote China’s participation in the North-South dialogue and South-South cooperation on ethical governance of artificial intelligence algorithms, and provide Chinese wisdom for global ethical governance of artificial intelligence.

    参考文献
    相似文献
    引证文献
引用本文

刘祖兵.论Sora的伦理风险与我国治理因应——也谈我国参与人工智能算法伦理全球治理的基本路径[J].河海大学学报(哲学社会科学版),2024,26(5):99-112.(LIU Zubing. On Sora’s Ethical Risks and China’s Governance Response: Also on the Basic Path of China’s Participation in the Global Governance of Artificial Intelligence Algorithm Ethics[J]. Journal of Hohai University (Philosophy and Socail Sciences),2024,26(5):99-112.(in Chinese))

复制
分享
文章指标
  • 点击次数:
  • 下载次数:
  • HTML阅读次数:
  • 引用次数:
历史
  • 收稿日期:
  • 最后修改日期:
  • 录用日期:
  • 在线发布日期: 2024-10-31
  • 出版日期: