Promise or Peril? When Human Efficacy Meets AI Capability Augmentation

Posted: 11 Dec 2022 Last revised: 13 Dec 2022

See all articles by Tian Lu

Tian Lu

Department of Information Systems, Arizona State University

Xianghua Lu

Fudan University

Yiyu Huang

Fudan University

Hai Wang

Singapore Management University - School of Computing and Information Systems; Carnegie Mellon University - Heinz College of Information Systems and Public Policy

Date Written: December 9, 2022

Abstract

Recent literature has demonstrated that humans high in self-efficacy tend to resist AI advice despite its superior capabilities. In this study, we question the validity of this finding for a prevalent practice, in which AI capability is dynamically augmented and humans accumulate knowledge and interaction experience with AI. As such, humans high in self-efficacy, compared with those low in self-efficacy, are more likely to recognize AI capability augmentation due to their strong learning ability. We thus posit that this phenomenon, together with the AI's superior capability compared with that of humans, would invoke individuals' active cognitive reasoning by comparing themselves with respect to the AI system. Consequently, they tend to change their attitude to follow AI. We leverage a unique natural experiment of AI capability augmentation in the context of an on-demand food delivery service to verify our conjecture. We obtain several interesting findings: (1) Human riders high in self-efficacy indeed become more compliant with AI advice in the presence of an AI capability augmentation, in both the short and long run. However, (2) riders' short-term food delivery performance significantly decreases. Further mechanism analysis reveals that (3) such decline mainly occurs during an adaptation (learning) window when riders shift their decision-making objective for the delivery sequence from their own to the one recommended by the AI system; in the long run, their delivery performance eventually recovers to a high level. By contrast, (4) riders low in self-efficacy are trivially affected by AI capability augmentation. From a dynamic perspective, our findings shed light on the critical role of cognitive reasoning triggered by high human self-efficacy in human-AI collaboration, which prompts humans to better leverage the wisdom of AI and their own to achieve better collaborative decision-making outcomes.

Keywords: AI Capability Augmentation, Human Efficacy, Human-AI Collaboration, Cognitive Reasoning, On-demand Food Delivery

Suggested Citation

Lu, Tian and Lu, Xianghua and Huang, Yiyu and Wang, Hai, Promise or Peril? When Human Efficacy Meets AI Capability Augmentation (December 9, 2022). Available at SSRN: https://ssrn.com/abstract=4298793 or http://dx.doi.org/10.2139/ssrn.4298793

Tian Lu (Contact Author)

Department of Information Systems, Arizona State University ( email )

Tempe, AZ 85287
United States

HOME PAGE: http://isearch.asu.edu/profile/tianlu1

Xianghua Lu

Fudan University

Yiyu Huang

Fudan University

Hai Wang

Singapore Management University - School of Computing and Information Systems ( email )

80 Stamford Road
Singapore 178902, 178899
Singapore

Carnegie Mellon University - Heinz College of Information Systems and Public Policy ( email )

5000 Forbes Avenue
Pittsburgh, PA 15213-3890
United States

Do you have a job opening that you would like to promote on SSRN?

Paper statistics

Abstract Views
165
PlumX Metrics