Ideas regarding the use of Generative AI for auditing

Generative artificial intelligence (AI) offers a wide range of potential uses for various areas, business models and processes. It will also have an impact on auditing work. Auditors will therefore need to familiarise themselves with the basic features of the underlying technology as well as the opportunities and challenges arising from this innovative application.

The sophistication and possibilities of AI became visible to the general public with the release of Chat GPT. The media are talking about AI’s ‘iPhone moment’ because of the fast-growing number of users, and the software’s performance has even surprised experts. GPT models are artificial neural networks with a special architecture. Generative AI refers to a category of AI algorithms in the field of deep learning. This results in a broad applicability of this technology – generating creative content such as text, images, music, and videos. Previous AI systems were usually designed only for one specific use case. 

At the moment the use of generative AI is focused on assistance functions. These include the creation of offers, reports and documentation papers, as well as the collection and utilisation of available information. A second possibility of use is to create specific expert systems via Foundation Models with comparatively little effort. These systems can be customised with specific and individual domain knowledge and design alternatives – for example, for use by tax departments. Auditors can then concentrate on the review of such statements or decision templates. 

Generative AI can also be used to support various phases of an audit, for example, in the preparation of market analyses as part of risk analysis, or to support the generation of audit documentation. Auditors can also examine audit evidence created in part or in full using generative AI. Generative AI itself can then be used to identify and verify this evidence. 

Challenges that may arise using AI include:

• Hallucinating: Producing incorrect results, presented very convincingly;
• Bias: Questionable quality of the training data of the models on which it was programmed;
• Traceability: AI models are complex; the better the model is understood, the better the results can be interpreted;
• Resource consumption: Fine-tuning of AI systems requires large amounts of data and computer resources;
• Regulation: The data may be subject to regulatory requirements, liability and ethical considerations; and
• Training: Many people are as yet unfamiliar with AI, and there is the need to train employees regarding functions and risks of AI.

With an emerging shortage of skilled staff, generative AI should be seen as an opportunity to support auditors in their tasks and also to guarantee the high quality of their work with fewer staff in the future. By utilising generative AI, the activities of auditors can be automated to a greater extent and carried out more efficiently.