When chatbots make errors: Cognitive and affective pathways to understanding forgiveness of chatbot errors

Abstract

This study aims to investigate whether individuals can forgive chatbots for their errors as they do for humans. Drawing on the contrasting theoretical frameworks of Computer are Social Actors (CASA) and machine heuristic in the Human-AI interaction (HAII), the study examines individuals’ forgiveness towards errors made by chatbots with different levels of anthropomorphism. Specifically, this study focuses on the affective and cognitive pathways in shaping individuals’ forgiveness towards chatbots. An online experiment (N = 580) with a two (anthropomorphism levels: low vs. high) × two (chatbot types: task-oriented vs. relationship-oriented) between-subjects design was conducted. Results indicated that compared to chatbots with low anthropomorphism, those with high anthropomorphism tend to elicit greater forgiveness for their errors. The effects of anthropomorphism on forgiveness to chatbot errors were mediated both through the affective route, by mitigating perceived severity and emotional aversion, and through the cognitive route, by attributing errors more to the users themselves. Our study also reveals the complex nature of forgiveness responses to chatbot errors, which are influenced by the specific context in which the chatbot is used. The theoretical and practical implications were discussed.