AAA Spark Meeting of the Regions

“ALEXA, AUDIT LOAN GRADES!”: Does Humanizing Artificial Intelligence Enhance Auditor Reliance?

• In Event: DSFI/SE - Session 2 (Experimental)

Abstract

Emerging research finds that, in some circumstances, auditors exhibit “algorithm aversion” and will discount their own firm’s evidence more heavily when it comes from an AI system instead of a human specialist. However, little is known about the persistence of this phenomenon or factors that can exacerbate and mitigate it. In an experiment with experienced auditors, we predict and find that auditors more heavily discount audit evidence from AI systems than human specialists after specialist errors become more salient via their colleagues’ observations. Thus, we identify second-hand observations of errors as a factor that amplifies auditors’ algorithm aversion. Importantly though, we find that humanizing the system (e.g., using a face, name, personal pronouns) mitigates this effect, causing auditors to react less severely to system errors. Firms, therefore, can facilitate continued reliance on AI systems and mitigate auditors’ algorithm aversion with a low cost and readily implemented design feature – humanization.

Authors

• Suzanne Mullinnix, AAMVA

• Benjamin P. Commerford, University of Kentucky

• Sean Dennis, University of Central Florida

• Jennifer R. Joe, University of Delaware

• Jenny Ulla, University of Kentucky