Abstract:
The advancement of Artificial Intelligence (AI) technologies increased significantly in the last few years. Moreover, the application of AI models expanded to a broader range. Hence, auditors are progressively encountering AI like systems, models and algorithms during audit and assurance projects. The growing scientific domains of eXplainable AI (XAI) and Responsible AI raise concerns around the transparency, explainability, and other ethicalities. These concerns, in combination with upcoming legislation, demand audit statements on reliability, integrity, and other aspects of AI models. Where auditing is a formal well-established practice, AI auditing is a novel practice. This research includes literature research, exploration of AI audit cases, and interviews with AI experts in order to discover relevant methods and specificalities of AI audits. Through the methodology of design science, a first formalised AI Audit Process is developed and proposed in order to provide AI auditors with a flexible reference frame to conduct customised AI audits. This research is a step towards the advancement of an AI auditing method and offers valuable insights for science and practice.