摘要
当前以深度学习代表的人工智能技术所能解决的问题往往需要具有大量可以借鉴学习的训练集,同时要求所获取的信息是完备的.但在复杂时变的真实应用环境中,采集到的数据往往包含大量噪声与不确定性,且仅包含环境的局部信息,导致了基于深度学习的人工智能应用的局限性.而在同样环境下,人类却往往可以依靠直觉做出较为迅速且适宜的判断,这为我们发展新的人工智能理论解决上述问题提供了灵感.本文系统地对人类直觉的概念、机理、类别等内容进行论述,分析了不同学科已有相关研究进展及存在的不足.在此基础上,提出了机器直觉这一新的交叉学科研究方向,并给出了实现机器直觉的基本准则,以及通过让机器实现类似甚至超过人类的直觉能力、赋予机器洞察力与创造性的研究目标.本文尝试设计了机器直觉的通用整体架构,确定了全息感知、直觉认知、直觉决策和博弈行动几个主要功能模块的基本原理和内涵.最后,从脑科学、认知科学、人工智能等多学科交叉研究角度,对机器直觉的应用与未来研究方向进行了展望,为后续展开机器直觉的研究提供方向性指引.
Currently,the problems that can be solved using deep learning-based artificial intelligence technology often require a large training data set for learning,and simultaneously,the information contained in the data set should be complete.However,in a real time-varying complex application environment,the collected data often contain significant noise,uncertainty,and only partial information of the environment,which limits the prospects of artificial intelligence applications based on deep learning.However,in a similar environment,humans can often make rapid and appropriate decisions based on intuition,providing inspiration to develop new artificial intelligence theories to solve the above problems.This article systematically discusses the concepts,mechanisms,categories,and other aspects of human intuition and analyzes the progress and shortcomings of existing research from different disciplines.Based on this analysis,machine intuition,a new cross-disciplinary research direction,is proposed,along with its basic criteria.The objective of machine intuition research is to facilitate machines with insight and creativity abilities to ultimately achieve intuitive intelligence similar or even superior to human instincts.Moreover,this paper attempts to design the general overall architecture of machine intuition and determines the basic principles and connotations of several main functional modules,such as holographic perception,intuitive cognition,intuitive decision-making,and game action.Lastly,from the viewpoint of cross-disciplinary research in brain science,cognitive science,and artificial intelligence,among others,the potential applications of machine intuition and future research directions are prospected,thus providing directional guidance for subsequent research on machine intuition.
作者
翟鹏
张立华
董志岩
万森
郭雨晨
甘中学
戴琼海
Peng ZHAI;Lihua ZHANG;Zhiyan DONG;Sen WAN;Yuchen GUO;Zhongxue GAN;Qionghai DAI(The Institute of AI and Robotics,Fudan University,Shanghai 200433,China;Institute for Brain and Cognitive Sciences,Tsinghua University,Beijing 100084,China;Ji Hua Laboratory,Foshan 528200,China;Engineering Research Center of AI and Robotics,Ministry of Education,Shanghai 200433,China;Shanghai Engineering Research Center of AI and Robotics,Shanghai 200433,China;Engineering Research Center of AI and Unmanned Vehicle Systems of Jilin Province,Shanghai 200433,China)
出处
《中国科学：信息科学》
CSCD
北大核心
2020年第10期1475-1500,共26页
Scientia Sinica(Informationis)
基金
上海市科委项目(批准号:19511132000)资助。
关键词
人工智能
直觉机理
机器直觉
直觉架构
全息感知
直觉认知
artificial intelligence
mechanism of intuition
machine intuition
structure of intuition
holographic perception
intuitive cognition
