误区与正道:法律人工智能算法问题的困境、成因与改进  

Misunderstandings and Correct Ways:the Dilemma,Causes and Improvements of Algorithm Problems in Legal Artificial Intelligence

作　　者：洪凌啸[1] HONG Ling-xiao(Law School,Sichuan University,Chengdu,Sichuan 610207,China)

机构地区：[1]四川大学法学院

出　　处：《四川师范大学学报（社会科学版）》2020年第1期58-70,共13页Journal of Sichuan Normal University(Social Sciences Edition)

摘　　要：随着AlphaGo在围棋领域的成功,人们开始思考是否能够将以"深度学习"算法为代表的计算统计概率型算法移植至法律领域。目前,法律科技公司往往打着"深度学习""强化学习""神经网络"等旗号宣传自身的法律人工智能产品,但其实际效果却往往不佳。在司法实践中真正得到使用的仍然是以"知识图谱"为代表的传统符号型算法。而效果较好的、使用了"深度学习"算法的语音文字转换系统也是一种通用型算法,并非为法律领域量身定制。同时,算法还存在着不透明、不公正、不中立等问题。在这一现象背后有商业、技术、人才三方面原因,法律科技公司囿于经济生存压力,不得不选择目前看来最稳妥的传统符号性算法;在技术方面,法律自身的特点以及法律标签数据缺失、法律数据质量不高、代表性不足等缺陷也使统计计算型算法在短期内尚无用武之地;而法律人工智能领域人才的匮乏更是制约其发展的重要掣肘。未来,需要开发一种符号型与统计概率型算法相结合的、专门针对法律领域的新型算法,同时,需要在对算法进行可视化操作的同时,进行算法警告、算法开源与算法审计。With the success of AlphaGo in the field of Go,people have begun to think about whether it is possible to transplant the statistical probabilistic calculation algorithms represented by "deep learning" algorithms into the legal field.At present,legal technology companies often promote their legal artificial intelligence products in the name of "deep learning", "reinforcement learning",and "neural network",which do not work well in practice.What is really used in judicial practice is still the traditional symbolic algorithm represented by the "knowledge map".The voice-to-text conversion system,which uses the "deep learning" algorithm,is also a general-purpose algorithm,not tailor-made for the legal field.It has problems such as opacity,injustice,and lack of neutrality.The three reasons behind this phenomenon are business,technology and talents.Legal technology companies have to choose the traditional symbolic algorithms that seem to be the most secure at present due to economic reason.Technically,the characteristics of the law itself also make statistical probabilistic algorithm useless in the short term.The lack of talents in legal artificial intelligence is a major constraint to its development.In the future,it is necessary to develop a new type of algorithm specifically for the legal field that combines symbolic and statistical probabilistic algorithms.Meanwhile,it is necessary to perform algorithm warnings,algorithm open source,and algorithm audits while visualizing the algorithms.

关 键 词：法律人工智能 算法 深度学习 强化学习 知识图谱 

分 类 号：DF0-059[政治法律—法学理论]