重新找回人工智能的可解释性

详细信息    查看全文 | 推荐本文 |

英文篇名：Refining the interpretability of artificial intelligence 作者：何华灿 英文作者：HE Huacan;School of Computer Science, Northwestern Polytechnical University; 关键词：人工智能 ; 可解释性 ; 演化 ; 不确定性 ; 泛逻辑学 ; 柔性命题逻辑 ; 柔性神经元 ; 数理辩证逻辑 英文关键词：artificial intelligence;;interpretability;;evolution;;uncertainty;;universal logic;;flexible propositional logic;;flexible neurons;;mathematical dialectic logic 中文刊名：ZNXT 英文刊名：CAAI Transactions on Intelligent Systems 机构：西北工业大学计算机学院; 出版日期：2019-04-25 16:38 出版单位：智能系统学报 年：2019 期：v.14;No.77 基金：国家自然科学基金面上项目(60273087);; 西北工业大学基础研究基金重点项目(W18101) 语种：中文; 页：ZNXT201903001 页数：20 CN：03 ISSN：23-1538/TP 分类号：5-24

摘要

针对深度神经网络AI研究的可解释性瓶颈,指出刚性逻辑(数理形式逻辑)和二值神经元等价,二值神经网络可转换成逻辑表达式,有强可解释性。深度神经网络一味增加中间层数来拟合大数据,没有适时通过抽象把最小粒度的数据(原子)变成粒度较大的知识(分子),再把较小粒度的知识变成较大粒度的知识,把原有的强可解释性淹没在中间层次的汪洋大海中。要支持多粒度的知识处理,需把刚性逻辑扩张为柔性命题逻辑(命题级数理辩证逻辑),把二值神经元扩张为柔性神经元,才能保持强可解释性。本文详细介绍了从刚性逻辑到柔性逻辑的扩张过程和成果,最后介绍了它们在AI研究中的应用,这是重新找回AI研究强可解释性的最佳途径。
        In view of the restrictions on the interpretability of artificial intelligence(AI) research on deep neural networks, it is indicated that rigid logic(mathematical formal logic) and binary neurons are equivalent. Moreover, a binary neural network can be converted into a logical expression, which is highly interpretable. The deep neural network blindly increases the number of intermediate layers to fit big data without the timely abstraction of data with the smallest granularity(atom) into knowledge with larger granularity(molecule), changes knowledge with smaller granularity into knowledge with larger granularity, and submerges the original strong explanatory power in the ocean of intermediate layers. To support knowledge processing of multiple granularities, rigid logic should be expanded into flexible propositional logic(proposition-level mathematical dialectic logic) and binary neurons should be expanded into flexible neurons to maintain the strong explanatory power. This paper introduces in detail the achievement of the expansion process from rigid logic to flexible logic and its application in AI research, which is the best method to recover the interpretability of AI.

引文

[1]何华灿.人工智能导论[M].西安:西北工业大学出版社,1988.11-15
    [2]何华灿.泛逻辑学理论:机制主义人工智能理论的逻辑基础[J].智能系统学报,2018,13(1):19-36.HE Huacan.Universal logic theory:logical foundation of mechanism-based artificial intelligence theory[J].CAAItransactions on intelligent systems,2018,13(1):19-36.
    [3]何华灿,何智涛.从逻辑学的观点看人工智能学科的发展[M]//.涂序彦.人工智能:回顾与展望.北京:科学出版社,2006:77-111.
    [4]谭铁牛.人工智能:天使还是魔鬼?[EB/OL].(2018-06-13)[2019-04-10].http://www.sohu.com/a/235446077_453160.
    [5]何华灿,王华,刘永怀,等.泛逻辑学原理[M].北京:科学出版社,2001.
    [6]HE Huacan,WANG Hua,LIU Yonghuai,et al.Principle of universal logics[M].Beijing:Science Press,2006.
    [7]HE Huacan.The outline on continuous-valued logic algebra[J].International journal of advanced intelligence,2012,4(1):1-30.
    [8]张文修.不确定性推理原理[M].西安:西安交通大学出版社,1996.
    [9]伊利亚·普里高津.确定性的终结:时间、混沌与新自然法则[M].湛敏,译.上海:上海科技教育出版社,1998:1-12
    [10]张金成.逻辑及数学演算中的不动项与不可判定命题(I)[J].智能系统学报,2014,9(4):499-510.ZHANG Jincheng.Fixed terms and undecidable propositions in logical and mathematical calculus(I)[J].CAAI transactions on intelligent systems,2014,9(4):499-510.
    [11]何华灿.S型超协调逻辑中的一项重大研究突破:评张金成《逻辑及数学演算中的不动项与不可判定命题》[J].智能系统学报,2014,9(4):511-514.
    [12]张金成.悖论、逻辑与非Cantor集合论[M].哈尔滨工业大学出版社,2018.
    [13]钟义信.机制主义人工智能理论:一种通用的人工智能理论[J].智能系统学报,2018,13(1):2-18.ZHONG Yixin.Mechanism-based artificial intelligence theory[J].CAAI transactions on intelligent systems,2018,13(1):2-18.
    [14]汪培庄.因素空间理论:机制主义人工智能理论的数学基础[J].智能系统学报,2018,13(1):37-54.WANG Peizhuang.Factor space-mathematical basis of mechanism based artificial intelligence theory[J].CAAI transactions on intelligent systems,2018,13(1):37-54.