Challenges of Explaining the Behavior of Black-Box AI Systems

被引:60
作者
Asatiani, Aleksandre [1 ]
Malo, Pekka [2 ]
Nagb, Per Radberg [3 ]
Penttinen, Esko [4 ]
Rinta-Kahila, Tapani [5 ]
Salovaara, Antti [6 ]
机构
[1] Univ Gothenburg, Informat Syst, Dept Appl Informat Technol, Gothenburg, Sweden
[2] Aalto Univ, Sch Business, Stat, Espoo, Finland
[3] It Univ Copenhagen, Copenhagen, Denmark
[4] Aalto Univ, Sch Business, Informat Syst, Espoo, Finland
[5] Univ Queensland, Australian Inst Business & Econ, Brisbane, Qld, Australia
[6] Aalto Univ, Sch Arts Design & Architecture, Espoo, Finland
关键词
D O I
10.17705/2msqe.00037
中图分类号
G25 [图书馆学、图书馆事业]; G35 [情报学、情报工作];
学科分类号
1205 ; 120501 ;
摘要
There are many examples of problems resulting from inscrutable AI systems, so there is a growing need to be able to explain how such systems produce their outputs. Drawing on a case study at the Danish Business Authority, we provide a framework and recommendations for addressing the many challenges of explaining the behavior of black-box AI systems. Our findings will enable organizations to successfully develop and deploy AI systems without causing legal or ethical problems.(1,2)
引用
收藏
页码:259 / 278
页数:20
相关论文
共 10 条
[1]  
[Anonymous], 2017, HARVARD J LAW TECHNO
[2]  
[Anonymous], 2010, Intelligence artificielle: Avec plus de 500 exercices
[3]  
[Anonymous], 2020, The Guardian
[4]  
Bajkowski J., 2019, ITNEWS
[5]  
Buranyi S., 2017, GUARDIAN
[6]   Designing Ethical Algorithms [J].
Martin, Kirsten .
MIS QUARTERLY EXECUTIVE, 2019, 18 (02) :129-142
[7]   AI and the path to envelopment: knowledge as a first step towards the responsible regulation and use of AI-powered machines [J].
Robbins, Scott .
AI & SOCIETY, 2020, 35 (02) :391-400
[8]   Explainability in human-agent systems [J].
Rosenfeld, Avi ;
Richardson, Ariella .
AUTONOMOUS AGENTS AND MULTI-AGENT SYSTEMS, 2019, 33 (06) :673-705
[9]  
Vincent J, 2019, VERGE 0125
[10]  
Warren T., 2018, VERGE 0524