Paper Number

1349

Paper Type

Complete Research Paper

Abstract

Explainable Artificial Intelligence (AI) aims to provide insight into the inner workings of black-box AI systems and thereby increase trust through the provision of local and global explanations. Nonetheless, the precise effects of explanations on AI trust remain ambiguous. We investigate (1) the effect of known trust antecedents on trust over the course of an interaction with an AI-based system, and how (2) a global explanation influences these antecedents, as well as (3) how usage of a system with/without experiencing an expectation violation influences these antecedents, and lastly (4) how the provision of a local explanation influences these antecedents, differentiated by whether an expectation violation had previously been experienced. We found all but one investigated antecedents to be significant predictors of trust. Additionally, we demonstrate the precise effects of global explanations, system usage with and without experiencing an expectation violation, and local explanations on trust antecedents.

Share

COinS
 
Jun 14th, 12:00 AM

The Effect of Explainable AI on AI-Trust and Its Antecedents over the Course of an Interaction

Explainable Artificial Intelligence (AI) aims to provide insight into the inner workings of black-box AI systems and thereby increase trust through the provision of local and global explanations. Nonetheless, the precise effects of explanations on AI trust remain ambiguous. We investigate (1) the effect of known trust antecedents on trust over the course of an interaction with an AI-based system, and how (2) a global explanation influences these antecedents, as well as (3) how usage of a system with/without experiencing an expectation violation influences these antecedents, and lastly (4) how the provision of a local explanation influences these antecedents, differentiated by whether an expectation violation had previously been experienced. We found all but one investigated antecedents to be significant predictors of trust. Additionally, we demonstrate the precise effects of global explanations, system usage with and without experiencing an expectation violation, and local explanations on trust antecedents.

When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.