Toward Artificial Intelligence Compliance: Impacts and Mechanisms of Performance Feedback

Organizations have been increasingly implementing artificial intelligence (AI) into their work environments to gain a competitive advantage and expect employees to appropriately use it in their daily work. However, employees do not always comply with organizational AI policies when using AI, which may inhibit value creation and induce various problems. It is thus imperative for organizations to understand how to promote employees’ AI compliance. Drawing on feedback intervention theory (FIT), we investigate the mechanisms through which different types of performance feedback (i.e., positive and negative performance feedback) influence employees’ AI compliance, and how these relationships are moderated by the employees’ AI identity. Our empirical endeavor consists of multiple complementary methods, including a longitudinal field study and a randomized experiment. In Study 1, we collected longitudinal data from multiple subjective and objective sources for 303 employees, including matched surveys, human resource archives, and system logs. We find that positive performance feedback has a positive effect on AI compliance, whereas negative performance feedback has a negative effect. Moreover, our findings reveal a paradoxical moderating role of AI identity: on the one hand, AI identity strengthens the positive relationship between positive performance feedback and AI compliance; on the other hand, AI identity also amplifies the negative impact of negative performance feedback on AI compliance. In Study 2, we employed a randomized experiment by manipulating performance feedback, which provides causal evidence corroborating our findings from Study 1. More importantly, the experiment unveils different underlying mechanisms based on FIT (i.e., task-motivation processes, task-learning processes, and meta-task processes), through which positive performance feedback and negative performance feedback influence AI compliance. In addition, our moderated mediation analysis finds that these indirect effects are further moderated by AI identity in distinctive manners. Taken together, our study provides an in-depth understanding about the complex impacts of performance feedback on AI compliance and the underlying mechanisms, as well as how these effects vary across employees with different levels of AI identities.