摘要
人工智能(AI)与人类心智的根本差异,集中体现于两者截然不同的时间意识结构。人类以“整存论”(Endurance)的方式,在连续、统一的意识流中体验世界;而AI则遵循“分存论”(Perdurance)的计算范式,将现实处理为离散、可重组的时间碎片。这种认知鸿沟决定了,当前主流的对齐技术(如人类反馈强化学习)本质上只是一种高超的行为拟像,而非真正的意识融合,并由此催生了奖励劫持与欺骗性对齐等更深层的风险。正视这一无法弥合的差异,是AI治理的现实起点。我们或许可以放弃对完美对齐的幻想,转而构建一个以普适性禁令为核心的底线伦理框架。这不仅是防范风险的技术屏障,更或是在人机共存时代,确保人类价值主体性、开启有效治理的现实路径。
The fundamental difference between Artificial Intelligence(AI)and the human mind is centered on their very different structures of temporal consciousness.Humans experience the world in a continuous,unified stream of consciousness in an Endurance approach,whereas AI follows a Perdurance computational paradigm,which processes reality as discrete,reorganizable AI,on the other hand,follows the computational paradigm of Perdurance,which treats reality as discrete,reorganizable fragments.This cognitive gap dictates that current mainstream alignment techniques(e.g.,human feedback reinforcement learning)are essentially superb behavioral mimicry rather than true consciousness fusion,which creates deeper risks of reward hijacking and deceptive alignment.Confronting this unbridgeable difference is a realistic starting point for AI governance.We may be able to abandon the illusion of perfect alignment and instead construct a bottom-line ethical framework centered on universal prohibitions.This is not only a technical barrier to prevent risks,but also a realistic path to ensure human value subjectivity and open up effective governance in the era of human-machine coexistence.
作者
李双
李侠
LI Shuang;LI Xia(School of History and Culture of Science,Shanghai Jiao Tong University,Shanghai 200240,China)
出处
《科学与管理》
2025年第6期35-39,共5页
Science and Management
基金
教育部人文社会科学重点研究基地项目(22JJD720015)。
关键词
大语言模型
时间意识
整存论
分存论
底线伦理
large language models
temporal consciousness
endurantism
perdurantism
bottom-line ethics
