Socially (un)acceptable errors of AI: Consumer perceptions of different AI-induced errors

Artificial intelligence (AI) commonly errs in practice. This study investigates consumer responses to two distinct types of errors: technical errors stemming from technological disruptions in algorithmic processes and social errors, which involve violations of social norms. These distinctions are critical, as our research reveals different consumer response patterns based on error type and error severity. Grounded in the theory of mind perception and expectation disconfirmation theory, we present findings from multiple experiments demonstrating that severe errors, regardless of type, evoke negative consumer responses. In contrast, minor social errors seem anticipated and mostly elicit responses more akin to those for error-free AI performance. However, in the realm of self-learning AI, these minor social errors are problematic. They can perpetuate the stigmatization of minorities and ethnic groups, highlighting the urgent need to prevent AI from violating social norms.