摘要
物质、能量、结构和时间是人类认知和机器认知的最基本要素,是人类智能和人工智能共同的奠基石。信息不是物质,不是能量,是在结构、时间上形成的衍生物。位于四要素之上的是思维软构体和物质硬构体,其上分别对应基于想象与推理的数学和基于观察与实证的物理,而基于数学和物理之上的是认知物理学。认知物理学的使命是用物理学的理论和方法解释人类认知及其形成过程,人工智能浓缩重演人类智能,用机器来模仿、拓展人的认知。未来,要突破图灵机局限,发展难以计算的记忆智能和具身交互智能;用云模型、数据场、拓扑势、云变换等认知物理学理论和方法,形式化人类的不确定性认知,形成认知螺旋,实现认知的自成长;研发可交互、会学习、自成长的认知机器,开辟各智其智、智人之智、智智与共的人工智能新方向。
Matter,energy,structure and time are the most basic elements of human cognition and machine cognition,and are the common cornerstone of human intelligence and artificial intelligence.Information is neither matter nor energy but a derivative of structure and time.On top of the four elements are the soft structure of thinking and the hard structure of matter,corresponding respectively to mathematics based on imagination and reasoning,and physics based on observation and empirical evidence.The mission of cognitive physics is to use the theories and methods of physics to explain human cognition and its formation process.Art ificial intelligence condenses and reproduces human intelligence,and uses machines to imitate and expand human cognition.In the future,breakthroughs should be made in the limitations of Turing machines and memory intelligence,and embodied interactive intelligence which are difficult to compute should be developed;the cloud model,data field,topological potential,cloud transformation and other cognitive physics theories and methods shall be used to formalize human uncertainty cognition,form cognitive spirals,and realize the self-growth of cognition;and the research and development of interactive,self-learning,self-growing cognitive machines shall be made to open up a new direction of artificial intelligence that combines various intelligence,human intelligence,and shared intelligence.
出处
《学术前沿》
CSSCI
北大核心
2024年第14期4-10,共7页
Frontiers
关键词
认知物理学
人工智能
人类智能
物质
能量
结构
时间
cognitive physics
artificial intelligence
human intelligence
matter
energy
structure
time
作者简介
李德毅,清华大学特聘教授,中国工程院院士,国际欧亚科学院院士,中国人工智能学会名誉理事长,中国指挥与控制学会名誉理事长,“吴文俊人工智能最高成就奖”获得者,认知物理学和不确定性人工智能主要开拓者。研究方向为计算机工程、不确定性人工智能、认知的形式化、机器认知、云模型、云变换、数据场、拓扑势、数据挖掘、知识发现和智能驾驶。
