人工智能的"理解困境"及其化解途径——从东西方文化的视角看
The"Understanding Dilemma"of Artificial Intelligence and its Solution:From the Perspective of Eastern and Western Cultures
任晓明1
作者信息
- 1. 四川大学哲学系(四川成都 10065)
- 折叠
摘要
近年来,机器学习建模领域得到了飞速发展,但是它在带来了巨大利益的同时伴随着巨大的风险,人工智能的技术实现是主要的风险来源.论文从可解释人工智能的视角,探讨如何破解其"理解困境"的途径和方法并指出,首先需要澄清诸如智能、理解之类的基本概念,认为智能的目标不仅是对于未来的预测,而且是对世界的理解.理解世界需要一个普适的架构,与统计相关模型相比,因果结构模型更具可解释性,是一种"深层理解".文章讨论了因果推理的认知框架,分析和比较它们在可解释性方面的优势和劣势,认为倾向理论比反事实依赖理论更具可解释性,这就为破解人工智能的"理解困境"开辟新的思路.最后指出,目前的深度学习系统虽然在预测精度方面取得进展,但在可解释性、可理解性、公平性和可信性等方面有所欠缺,在应对因果涌现等复杂性问题时还需要借鉴东方智慧,特别是借鉴印度正理哲学的因果思想.
Abstract
In recent years rapid advances have been made in machine learning model-ing.Nevertheless,big benefits have come along with big risks,of which the technological realization of artificial intelligence(AI)is a major source of risks.From the perspective of explaining AI,this paper discusses the solu-tion to the dilemma,pointing out that we should first of all clarify such con-cepts as intelligence and understanding,and assuming that the goal of intel-ligence is not only a prediction of the future but also an understanding of the world.To understand the world we should have a universally-applicable framework,and compared with statistics-related models the cause-effect model is more interpretative and hence a"deep understanding".This paper discusses the cognitive frameworks for the cause-effect inference,analyzes and compares their pros and cons in interpretation,and holds that the dispo-sition framework is more interpretative,which opens a new path of thinking for solving the"understanding dilemma"of AI.Finally,this research indi-cates that although much progress has been made in the current deep-learning system in terms of prediction accuracy,it is deficient in interpreta-tion,understanding,fairness and reliability.While addressing the complexity of cause-effect issues we should resort to oriental wisdom,particularly In-dian orthodox philosophy regarding the causal phenomena.
关键词
人工智能/理解困境/不确定性推理/正理哲学Key words
artificial intelligence/understanding dilemma/uncertainty inference/or-thodox philosophy引用本文复制引用
基金项目
2022年国家社会科学基金重点项目(22ZBX019)
出版年
2024
NETL
NSTL
万方数据