Paper Type
Complete
Description
Artificial intelligence (AI) automates decisions, usually without any transparency or insight into its internal mechanisms. Often, missing transparency results in a lack of trust in these systems, reducing their usage and adoption. Explainable AI and the provision of explanations were designed to address these shortcomings. However, the exact effects of explanations remain ambiguous in the literature. We address this empirically by measuring trust at multiple points throughout the interaction with a system. We find the effect of global explanations to be highly dependent on the level of trust already exhibited before receiving the explanations. Furthermore, we find local explanations to only be of use in the case of an expectation violation. Ultimately, the trust recovered with the help of a local explanation cannot entirely make up for the loss of trust associated with an expectation violation.
Paper Number
1491
Recommended Citation
de Zoeten, Marc Christoph; Ernst, Claus-Peter H.; and Rothlauf, Franz, "A Matter of Trust: How Trust in AI-Based Systems Changes During Interaction" (2023). AMCIS 2023 Proceedings. 15.
https://aisel.aisnet.org/amcis2023/sig_odis/sig_odis/15
A Matter of Trust: How Trust in AI-Based Systems Changes During Interaction
Artificial intelligence (AI) automates decisions, usually without any transparency or insight into its internal mechanisms. Often, missing transparency results in a lack of trust in these systems, reducing their usage and adoption. Explainable AI and the provision of explanations were designed to address these shortcomings. However, the exact effects of explanations remain ambiguous in the literature. We address this empirically by measuring trust at multiple points throughout the interaction with a system. We find the effect of global explanations to be highly dependent on the level of trust already exhibited before receiving the explanations. Furthermore, we find local explanations to only be of use in the case of an expectation violation. Ultimately, the trust recovered with the help of a local explanation cannot entirely make up for the loss of trust associated with an expectation violation.
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.
Comments
SIG ODIS