胡塞尔的意向性理论与人工智能关系刍议

Some Remarks on the Relationship between Husserl’s Notions of “Intentionality” and Artificial Intelligence

中文关键词:  意向性；悬搁；信念盒；推论主义；人工智能；意向相关项

英文关键词:intentionality, epoch, belief box, inferentialism, artificial intelligence, noema

中文摘要:

      胡塞尔的意识哲学对于人工智能的意义，一向被学界严重低估。实际上，胡塞尔的“现象学悬搁”方法虽貌似对一切自然主义的认知建模工作构成了威胁，但依然能在某种变通的意义上对人工智能构成启发。具体而言，“现象学悬搁”对于意识表征中意向模式的普遍存在性的提示，完全可以在人工智能语境中被转述为下面这个工作思路：建造一台具有自动“修偏”机能的智能机器，只有在为其配属了丰富的心理模式的情况下才是可能的，而心理模式的配置本身又会倒逼系统将对于世界的素朴表征转化为那些悬置了外部真值判断的“内存在表征”。此外，对于胡塞尔的“Noema”概念的推论主义解释模式，也可以告诉我们为何福多所提出的关于命题态度的“盒喻”式处理方案乃是不对的。不过，目前主流AI在处理意向性问题方面的表现，依然是十分笨拙的：这些主流进路要么根本就无法在符号表征的层面上触及意向性问题（如深度学习），要么不得不采用福多的“盒喻”而导致模型的不灵活性（如某些符号AI技术），要么就在过分强调环境与主体的相互关系的同时预设了胡塞尔所反对的“自然主义态度”（如某些能动主义技术路线）。由此看来，我们如果要设计一种在最低限度上符合胡塞尔精神的AI系统，我们就必须与目前的主流AI技术路线分道扬镳。

英文摘要:

      The relevance of Edmund Husserl’s notion of “intentionality” to artificial intelligence (AI) has long been underestimated. As a matter of fact, although the phenomenological methodology of “epoché” appears to be undermining any attempt to do cognitive modelling, it can still offer inspirations for AI in a deeper sense. To be more specific, the “omnipresence” of psychological modes in pure consciousness as the residue of “epoché”, if appropriately reinterpreted, can provide principles an AI oriented systematic reconstruction of psychological modes, which could facilitate an AI system to revise its beliefs in a flexible manner. Moreover, Husserl’s notion of “Noema”, if reinterpreted in accordance with Brandom’s inferentialism, can be regarded as a competing theory with Jerry Fodor’s “Language of Thought Hypothesis”, which embraces a theory of psychological modes, according to which there are different “boxes”, each of which specifies a certain psychological mode. Husserl’s notion of “Noema”, by contrast, can explain belief revisions without presupposing the existence of these “boxes”. However, the mainstream AI approaches simply ignore Husserl’s insights of intentionality. For example, the connectionist/deep learning approach cannot handle intentionality on the symbolic level, and the symbolic AI approach cannot resist the temptation of presupposing Fodorrian “belief boxes” in an inflexible manner, whereas the enactivist approach overemphasizes the importance of the interplays between artificial agents and their environments by modelling both in a way not from a certain phenomenological perspective. Therefore, it is not unfair to conclude that the algorithmic realization of Husserlian intentionality has to be “suspending” most mainstream AI approaches.