黑格尔“逻辑学”对人工智能的启示

What can Artificial Intelligence Learn from Hegel's Logic?

将黑格尔哲学与人工智能研究联系在一起的纽带,乃是所谓的"框架问题":编程者该如何设计智能体,才能够使其在进行趋利避害的考量中,自动避免过多地涉及不相关獉獉獉的事项,并以此提高其整体行为效率呢?从语言哲学与逻辑哲学的角度看,该问题的真正的矛头指向乃是基于真值函项语义学的传统编程方式,因为这种编程方式的确无法完美地刻画出语义层面上的"相关性"与"不相关性"。而黑格尔的《逻辑学》在此被引入的必要性也正体现于如下考量:从人工智能或认知科学的角度看,《逻辑学》所讨论的诸种范畴,其实就是智能体用以在不同层面上进行"相关性"指派时所必须依赖的工具箱。若进一步以《逻辑学》的"‘是’论"为主要分析对象,那黑格尔带给人工智能研究的启发就体现在:将作为系词的"是"联接与综合各种表象的统一功能拆分为不同的谓词,否则我们就难以避免陷入"如何周延地列举基本谓词"这一技术困境;黑格尔对于"变"这一范畴基本地位的强调,既批判了公理化进路的符号人工智能对于语境性变化的迟钝,也批判了人工神经元网络技术或深度学习技术用单纯的"映射法"回避跨语境推理的做法;黑格尔在"是—否—变"之间展开的辩证推理过程,并不能落实为某种充满神秘主义色彩的"辩证逻辑",而必须被理解为对于认知主体探索真理的基本步骤的一种概括。因此,与之能够相提并论的人工智能对应物,也并不是一阶谓词逻辑、贝叶斯网络或人工神经元网络,而是可以被图示化为"流程图"的某种宏观问题解决策略,譬如司马贺与纽埃尔的"通用问题求解器"所涉及的"目标—手段"进路。最后,从总体上看,将黑格尔的《逻辑学》在现代哲学语境中加以祛魅化的基本思路,便是卸载《逻辑学》研究的形而上学负担,将其视为某种纯粹的认知逻辑研究。而反过来说,在变更这种基本解释策略的前提下对于人工智能要素的引入,也为黑格尔式逻辑自身的工程学化实现提供了契机。

Although the so-called ＂Heddeggerian Philosophy of Artificial Intelligence,＂thanks to the hard work of Hubert Dreyfus and John Haugeland,has been identified as a stable branch of philosophy of AI in the west,the resources of Hegelian philosophy does not play such a salient role in shaping any identifiable branch in Philosophy of AI,despite the affinity between two philosophers＇ writing styles. The core problem in AI which is requiring a Hegelian diagnosis is the so-called ＂frame problem,＂namely,the problem on how to design an artificial agent which can maximize its benefits by automatically avoiding ＂considering＂issues irrelevant to the task which is being handled.From the perspective of philosophy of language and philosophy of logic,this problem is tantamount to a problem on how to catch semantic relevance on a computable platform,and this problem looks so challenging to truth-functionsemantics-based AI,since semantics of this type is not apt enough to handle ＂relevance/irrelevance.＂As to Hegel＇s Logic,it is relevant to the frame problem in that the list of categories discussed by him can be viewed as a toolbox of different conceptual devices for handling semantic relevance on different levels. To be more specific,if the first part of his Logic,namely,＂the doctrine of being,＂is considered as the exemplar part of Logic for showing the possibility of developing a ＂Hegelian Philosophy of AI,＂then there are at least three morals that one can draw for AI from Hegel＇s relevant discussions：（ a） early analytical thinkers＇ proposal for splitting ＂being＂into different predicates does not deserve a serious consideration,in that this move would bring about a further problem on how to exhaustively list all of the primary predicates in programing;（ b） the fundamental status that Hegel attributes to the category＂becoming＂can be viewed as a criticism of the insensitivity to contextual changes in the axiomatic symbolic approach,as well as a criticism of the systematic reduction of cross-contextual inferences in terms of mapping procedures from data of a certain domain to outputs in the same domain in the approach of artificial neural networks or deep learning;（ c） the dialectical inference from ＂being＂to ＂nothing＂and to ＂becoming＂is actually not indicating a new logic which is different from formal logic. Rather,it is a flowchart on the same level of the ＂goal-means＂approach in Herbert Simon and Allen Newell＇s General Problem Solver Project,or in another way,it is a generalization of the procedures for probing the truth in the world. Hence,it is pointless to formalize the dialectical logic alone if it is detached from potential problemsolving contexts. Generally speaking,my research project for correlating Hegel with AI is not only beneficial to AI but also to the study of Hegel. Or put it differently,my strategy of reading Hegel＇s notion of ＂logic＂mainly in terms of a cognitive logic can make the disenchantment of the philosophical image of Hegel himself more easily,since my strategy can easily elude all of the metaphysical mysteries traditionally associated with his Logic. Moreover,the introduction of the AI-based perspective into the study of Hegel can also make it possible to demystify Hegelian speculations in terms of computer science.