arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2512.04334cs.HC
收藏

可受人类控制的AI:有意义的人类控制

Human-controllable AI: Meaningful Human Control

School of Foreign Languages, Anhui Agricultural University(安徽农业大学外国语学院);Department of Psychology and Behavioral Sciences, Zhejiang University(浙江大学心理学与行为科学系)

Chengke Liu, Wei Xu

AI总结:

本文提出有意义的人类控制(MHC)作为AI发展的重要原则,强调技术设计、治理和人类共同作用的重要性,旨在实现人类中心的AI发展。

摘要:

开发可受人类控制的人工智能(AI)并实现有意义的人类控制(MHC)已成为应对这些挑战的关键原则,确保AI的伦理一致性以及有效的治理。MHC也是人类中心AI(HCAI)研究和应用中的关键焦点。本章系统地探讨了AI中的MHC,阐述了其基础原理和未来发展方向。MHC不仅仅是操作权,而是人类的理解、干预以及AI决策中责任可追溯性的统一,这需要技术设计、AI治理和人类共同发挥作用。MHC确保AI的自主性服务于人类而不阻碍技术进步。人类控制的模式需要与技术的层级相匹配,人类监督应平衡对AI的信任与怀疑。对于未来的AI系统,MHC要求人类可控性作为前提,需要:(1)具有嵌入式人类控制机制的技术架构;(2)优化的人类与AI交互以更好地获得人类理解;(3)AI系统的发展和谐统一智能与人类可控性。治理必须优先考虑HCAI策略:平衡创新与风险缓解的政策,以超越技术精英主导的人类中心参与框架,以及全球推广MHC作为通用治理范式以保障HCAI发展。展望未来,需要加强跨学科研究以探讨AI系统的可控性,增强利益相关者的伦理和法律意识,超越简单的技术设计视角,关注围绕人类控制的知识构建、复杂性解释和影响因素。通过促进MHC,人类可控AI的发展可以进一步推进,交付HCAI系统。

英文摘要:

Developing human-controllable artificial intelligence (AI) and achieving meaningful human control (MHC) has become a vital principle to address these challenges, ensuring ethical alignment and effective governance in AI. MHC is also a critical focus in human-centered AI (HCAI) research and application. This chapter systematically examines MHC in AI, articulating its foundational principles and future trajectory. MHC is not simply the right to operate, but the unity of human understanding, intervention, and the traceablity of responsibility in AI decision-making, which requires technological design, AI governance, and humans to play a role together. MHC ensures AI autonomy serves humans without constraining technological progress. The mode of human control needs to match the levels of technology, and human supervision should balance the trust and doubt of AI. For future AI systems, MHC mandates human controllability as a prerequisite, requiring: (1) technical architectures with embedded mechanisms for human control; (2) human-AI interactions optimized for better access to human understanding; and (3) the evolution of AI systems harmonizing intelligence and human controllability. Governance must prioritize HCAI strategies: policies balancing innovation and risk mitigation, human-centered participatory frameworks transcending technical elite dominance, and global promotion of MHC as a universal governance paradigm to safeguard HCAI development. Looking ahead, there is a need to strengthen interdisciplinary research on the controllability of AI systems, enhance ethical and legal awareness among stakeholders, moving beyond simplistic technology design perspectives, focus on the knowledge construction, complexity interpretation, and influencing factors surrounding human control. By fostering MHC, the development of human-controllable AI can be further advanced, delivering HCAI systems.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑