Abstract:
|
There is considerable excitement about new artificial intelligence (AI) and machine learning (ML) approaches to auditing. In fact, some feel that these could eventually replace many statistical approaches to auditing such as audit sampling. However, it is important to understand not only the strengths of AI and ML, but also their limitations. In this presentation, We will articulate how statistical auditing, and audit sampling in particular, will still play an essential role in audit work. The power of AI and ML approaches is that they can very efficiently detect items and issues related to things that have been discovered before. Furthermore, we may even be unable to understand the manner of the relationship but we can still find the items at fault. The challenge arises when totally new classes of anomalies and fraud arise over time that have no relationship to those that were included in a model's training set. In such cases the model would be "untrained" on those features and aspects that might have served as targets for the model. However, should these new targets become sufficiently prevalent, audit sampling or other approaches will pick them up.
|