人工智能生存性风险:演变、分层与应对  

Existential Risks of Artificial Intelligence:Evolution,Stratification,and Responses

作　　者：姚旭 张傲 YAO Xu;ZHANG Ao(Fudan Development Institute,Fudan University,Shanghai 200433,China;School of International Relations and Public Affairs,Fudan University Yangpu District,Shanghai 200433,China)

机构地区：[1]复旦大学发展研究院,上海200433 [2]复旦大学国际关系与公共事务学院,上海200433

出　　处：《智能物联技术》2024年第5期7-11,共5页Technology of Io T& AI

基　　金：复旦大学-金光集团思想库2024年度基金项目研究课题(项目编号:JGSXK2408)。

摘　　要：人工智能生存性风险的讨论由来已久,人工智能生存性风险的基础是通用人工智能乃至超级人工智能的实现,但这一过程仍有阻碍。人工智能生存性风险的现状大多源于目标偏差,已引发算法和就业等一系列社会问题。人工智能生存性风险的终局被认为可能导致人类灭绝,其中包括推动战争形态和环境气候的变化。为应对人工智能生存性风险,需要统筹好人工智能的发展和安全,强化人工智能发展中的治理和技术软硬兼施的安全手段,推动国际人工智能安全治理合作,进一步打破发达国家和发展中国家的治理鸿沟和治理隔阂。The discussion surrounding the existential risks of artificial intelligence(AI)has been longstanding,fundamentally rooted in the realization of artificial general intelligence and even artificial super intelligence.However,significant obstacles still lie ahead in this process.The current state of existential risks largely stem from goal misalignment,leading to a series of social issues such as algorithmic bias and employment disruption.The ultimate consequence of these existential risks is perceived as potentially leading to human extinction,encompassing the escalation of warfare and environmental changes.To address these risks,it is essential to coordinate the development and safety of AI,enhance governance in AI advancement,implement a blend of technical and regulatory safety measures,and promote international cooperation in AI safety governance,thereby bridging the governance gap between developed and developing nations.

关 键 词：人工智能 生存性风险 发展安全 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]