可解释人工智能问题  

Problem with Explainable Artificial Intelligence

作　　者：顿新国[1] DUN Xin-guo(Institute for Contemporary Philosophy of Intelligence and the Future of Humanity,Nanjing University,Nanjing 210023,China)

机构地区：[1]南京大学当代智能哲学与人类未来研究中心,南京210023

出　　处：《江苏行政学院学报》2023年第3期44-51,共8页The Journal of Jiangsu Administration Institute

基　　金：国家社科基金重大项目“新一代人工智能驱动的逻辑学研究”(20&ZD047);南京大学文科卓越研究计划中长期专项“形式知识论研究”的阶段性成果。

摘　　要：当代人工智能技术在给人类带来多方面福利的同时,面临黑箱、鲁棒性和可解释性等问题,发展值得信赖的可解释人工智能已成为人工智能领域的核心关切。论文在论证可解释人工智能为什么重要的基础上,对可解释人工智能的内涵进行界定,进而提出发展可解释人工智能需遵循的一些原则,它们也可看作可解释人工智能需达致的目标和要求。论文最后提出了两点关于实现可解释人工智能途径的构想.The development of trustworthy and explainable Artificial Intelligence(AI)has become a central concern in the field of AI,as contemporary AI technologies bring multiple benefits to humans while facing problems,such as black box,robustness and explainability.This paper defines the meaning of explainable AI based on the argument of why explain-able AI is important,and then proposes some principles to be followed in developing explainable Al,which can also be re-garded as the goals and requirements to be achieved by explainable AI.The paper concludes with two ideas on pathways for achieving explainable AI.

关 键 词：人工智能 可解释性 可信赖性 透明性 可达性 

分 类 号：N02[自然科学总论—科学技术哲学]