Challenges for Leveraging Explainable Artificial Intelligence in Audit Procedures

Abstract

This paper discusses the challenges encountered by the audit industry in light of the dearth of well-labeled data and the increasing adoption of machine learning technologies. Although existing machine learning techniques have their merits, they have limitations when it comes to transactional data. Explainable artificial intelligence (XAI) can be a potential solution for applying machine learning models in audit procedures. Primarily, this study discusses challenges related to dependence on preprocessing, verification of explanation, variation in XAI techniques, limitations for feature importance explanation, auditors’ attitude to XAI, and computation time. The paper provides some potential solutions for these challenges.