分析哲学的逻辑知识真对人工智能有用吗?  

Is the Knowledge of Logic in Analytic Philosophy Genuinely Applicable for Artificial Intelligence?

作　　者：徐英瑾[1] Xu Yingjin

机构地区：[1]复旦大学哲学学院,上海200433

出　　处：《南开学报（哲学社会科学版）》2020年第3期87-98,共12页Nankai Journal:Philosophy,Literature and Social Science Edition

基　　金：国家社会科学基金重大项目(15ZDB020);教育部哲学社会科学研究重大课题攻关项目(19JZD010)。

摘　　要：英美分析哲学的主要学术训练方式,是建立在对于肇始于弗雷格的现代逻辑(一阶谓词逻辑与命题逻辑)的尊重之上的。因此,分析哲学所说的“分析”,大约就可以等于“运用现代逻辑的工具进行分析”。但这样的学术训练,显然是没有办法应对于当下基于深度学习的人工智能发展现状的。另外,即使对于基于符号表征与明述规则的传统人工智能而言,主流分析哲学的训练所能够提供的思想支持也未必始终是积极的。毋宁说,基于现代逻辑的主流分析哲学的“理想化”讨论方式往往会忽略智能系统的运作效率问题,而这一点又会倒逼着天然具有工程学面相的人工智能研究通过某些“特设化”的修正(如珀拉克的“奥斯卡规划”所给出的那种修正)来部分弥补相关的缺憾。然而,这种零敲碎打的修补方式显然是很难达到真正的普遍性的,遑论为具有多重用途的通用人工智能设备的设计铺展可行的路线图。The major tutelage in mainstream Anglophone analytic philosophy is tethered on the modern logic originating from the respective era of Gottlob Frege.The“philosophical analysis”in analytic philosophy fundamentally refers to the analysis executed in the light of modern logic.However,form the perspective of the philosophy of artificial intelligence(AI),logical training of this sort may not be entirely beneficial to understanding the development of some branches of AI,especially the branch of“deep learning”.Furthermore,with respect to traditional AI based on symbolic expressions and explicit rules,the intellectual support provided by the tutelage of analytic philosophy does not continuously satisfy AI significantly.The idealized approaches to the discussion in the modern logic of mainstream analytic philosophy usually neglect the operational efficiency of AI systems.As such,this will make the study of AI partly compensate for related contritions with various ad hoc remedies(those offered by John Pollock’s“OSCAR Project”).However,this kind of impromptu compensations cannot achieve genuine universal applicability,let alone provide a feasible roadmap for the design and development of universal AI equipment with multiple functions.

关 键 词：分析哲学 人工智能 现代逻辑 心理学 词项逻辑 非单调推理 

分 类 号：B089[哲学宗教—哲学理论] TP18[自动化与计算机技术—控制理论与控制工程]