Effects of AI Explanations on Human-AI Collaboration: An Experimental Study on Decision Performance and Reliance
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
This study examines how AI explanations affect human-AI collaborative decision-making. We test whether explainable AI (XAI) improves accuracy, speed, confidence, and reliance when distinguishing correct from incorrect AI suggestions. Using call center agents in three conditions—human only, human with AI, and human with XAI—we evaluate decisions made with classifiers and LIME-based explanations. Results show that explanations significantly increase decision accuracy, reduce overreliance on AI, and promote appropriate non-reliance. These findings emphasize the critical role and applicability of AI explanations in human-AI collaboration and contribute practical insights into designing AI assistants.
Abstract
This study examines how AI explanations affect human-AI collaborative decision-making. We test whether explainable AI (XAI) improves accuracy, speed, confidence, and reliance when distinguishing correct from incorrect AI suggestions. Using call center agents in three conditions—human only, human with AI, and human with XAI—we evaluate decisions made with classifiers and LIME-based explanations. Results show that explanations significantly increase decision accuracy, reduce overreliance on AI, and promote appropriate non-reliance. These findings emphasize the critical role and applicability of AI explanations in human-AI collaboration and contribute practical insights into designing AI assistants.