Role of Human-AI Interaction in Selective Prediction

被引:0
作者
Bondi, Elizabeth [1 ]
Koster, Raphael [2 ]
Sheahan, Hannah [2 ]
Chadwick, Martin [2 ]
Bachrach, Yoram [2 ]
Cemgil, Taylan [2 ]
Paquet, Ulrich [2 ]
Dvijotham, Krishnamurthy [3 ]
机构
[1] Harvard Univ, Cambridge, MA 02138 USA
[2] DeepMind, London, England
[3] Google Brain, Mountain View, CA USA
来源
THIRTY-SIXTH AAAI CONFERENCE ON ARTIFICIAL INTELLIGENCE / THIRTY-FOURTH CONFERENCE ON INNOVATIVE APPLICATIONS OF ARTIFICIAL INTELLIGENCE / THE TWELVETH SYMPOSIUM ON EDUCATIONAL ADVANCES IN ARTIFICIAL INTELLIGENCE | 2022年
关键词
D O I
暂无
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
Recent work has shown the potential benefit of selective prediction systems that can learn to defer to a human when the predictions of the AI are unreliable, particularly to improve the reliability of AI systems in high-stakes applications like healthcare or conservation. However, most prior work assumes that human behavior remains unchanged when they solve a prediction task as part of a human-AI team as opposed to by themselves. We show that this is not the case by performing experiments to quantify human-AI interaction in the context of selective prediction. In particular, we study the impact of communicating different types of information to humans about the AI system's decision to defer. Using realworld conservation data and a selective prediction system that improves expected accuracy over that of the human or AI system working individually, we show that this messaging has a significant impact on the accuracy of human judgements. Our results study two components of the messaging strategy: 1) Whether humans are informed about the prediction of the AI system and 2) Whether they are informed about the decision of the selective prediction system to defer. By manipulating these messaging components, we show that it is possible to significantly boost human performance by informing the human of the decision to defer, but not revealing the prediction of the AI. We therefore show that it is vital to consider how the decision to defer is communicated to a human when designing selective prediction systems, and that the composite accuracy of a human-AI team must be carefully evaluated using a human-in-the-loop framework.
引用
收藏
页码:5286 / 5294
页数:9
相关论文
共 32 条
[1]   Guidelines for Human-AI Interaction [J].
Amershi, Saleema ;
Weld, Dan ;
Vorvoreanu, Mihaela ;
Fourney, Adam ;
Nushi, Besmira ;
Collisson, Penny ;
Suh, Jina ;
Iqbal, Shamsi ;
Bennett, Paul N. ;
Inkpen, Kori ;
Teevan, Jaime ;
Kikin-Gil, Ruth ;
Horvitz, Eric .
CHI 2019: PROCEEDINGS OF THE 2019 CHI CONFERENCE ON HUMAN FACTORS IN COMPUTING SYSTEMS, 2019,
[2]  
[Anonymous], 2021, ECOLOGY EVOLUTION, DOI DOI 10.1111/2041-210X.13504
[3]   Optimally Interacting Minds [J].
Bahrami, Bahador ;
Olsen, Karsten ;
Latham, Peter E. ;
Roepstorff, Andreas ;
Rees, Geraint ;
Frith, Chris D. .
SCIENCE, 2010, 329 (5995) :1081-1085
[4]  
Bansal GG, 2021, AAAI CONF ARTIF INTE, V35, P11405
[5]   Context R-CNN: Long Term Temporal Context for Per-Camera Object Detection [J].
Beery, Sara ;
Wu, Guanhang ;
Rathod, Vivek ;
Votel, Ronny ;
Huang, Jonathan .
2020 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR 2020), 2020, :13072-13082
[6]   Uncertainty as a Form of Transparency: Measuring, Communicating, and Using Uncertainty [J].
Bhatt, Umang ;
Antoran, Javier ;
Zhang, Yunfeng ;
Liao, Q. Vera ;
Sattigeri, Prasanna ;
Fogliato, Riccardo ;
Melancon, Gabrielle ;
Krishnan, Ranganath ;
Stanley, Jason ;
Tickoo, Omesh ;
Nachman, Lama ;
Chunara, Rumi ;
Srikumar, Madhulika ;
Weller, Adrian ;
Xiang, Alice .
AIES '21: PROCEEDINGS OF THE 2021 AAAI/ACM CONFERENCE ON AI, ETHICS, AND SOCIETY, 2021, :401-413
[7]   The Behavioral and Neural Mechanisms Underlying the Tracking of Expertise [J].
Boorman, Erie D. ;
O'Doherty, John P. ;
Adolphs, Ralph ;
Rangel, Antonio .
NEURON, 2013, 80 (06) :1558-1571
[8]   To Trust or to Think: Cognitive Forcing Functions Can Reduce Overreliance on AI in AI-assisted Decision-making [J].
Buçinca Z. ;
Malaya M.B. ;
Gajos K.Z. .
Proceedings of the ACM on Human-Computer Interaction, 2021, 5 (CSCW1)
[9]  
CHOW CK, 1970, IEEE T INFORM THEORY, V16, P41, DOI 10.1109/TIT.1970.1054406
[10]   A Case for Humans-in-the-Loop: Decisions in the Presence of Erroneous Algorithmic Scores [J].
De-Arteaga, Maria ;
Fogliato, Riccardo ;
Chouldechova, Alexandra .
PROCEEDINGS OF THE 2020 CHI CONFERENCE ON HUMAN FACTORS IN COMPUTING SYSTEMS (CHI'20), 2020,