Unlocking the Value of Real-Time AI Assistance: Who Benefits, and Why?
46 Pages Posted: 4 Jul 2023 Last revised: 9 Jan 2024
Date Written: December 16, 2023
Abstract
As companies increasingly integrate artificial intelligence (AI) into their services, it is becoming crucial to understand AI's impact on service users. One novel application of AI is to provide real-time assistance throughout the service process, featuring a more dynamic and interactive human-AI relationship. In this paper, we investigate how human users derive value from such real-time AI assistance and how this value varies according to users' experience levels. We address these questions through a collaboration with an international car-sharing platform that recently underwent a technology upgrade involving an AI-powered driver monitoring and feedback system. Combining theoretical insights with empirical analyses, we uncover a heterogeneous effect characterized by a U-shaped relationship: users with low or high experience levels benefit from AI assistance, while users with medium experience levels do not. Our findings further reveal that different user groups derive value from AI in distinct ways: low-experience users derive value passively due to their reliance on AI assistance, whereas high-experience users actively enhance AI's value by adapting their behaviors to mitigate the costs associated with AI during the real-time interaction. On the other hand, medium-experience users struggle to overcome the cost-benefit trade-off of AI, as they derive less benefit from AI's assistance compared to their less experienced peers and lack the adaptability of their more experienced peers to mitigate its costs. These insights are useful for guiding AI adoption and algorithm customization in service applications, and provide broader implications for other human-AI interaction processes, such as AI-assisted medical diagnosis and ChatGPT.
Keywords: human-AI interaction, real-time feedback, service platform, consumer behavior, causal inference
JEL Classification: C23, D03, L80, M15
Suggested Citation: Suggested Citation