Can artificial intelligence help or hinder the compliance of vulnerable taxpayers?: case studies from the Australian National Tax Clinic Program

Artificial intelligence (AI) has received a lot of media attention recently due to the release of ChatGPT, a generative AI tool that is much more advanced than previous AI tools and can seemingly answer a wide variety of text"‘based questions. This led to some discussion regarding the potential for AI to replace employees in the workforce or to take over the tasks of professionals including accountants and lawyers. It also led to some discussion about the harm that AI may cause if it is used inappropriately or naively and what the role of AI ought to be. This article contributes to these discourses by considering how AI might assist tax professionals in providing tax services and advice as part of the Australian National Tax Clinic Program (NTCP) and the Australian Taxation Office in administering the Australian tax system. The NTCP provides free tax services to taxpayers who need it with many taxpayers in extremely vulnerable situations coming forward for assistance. Since AI is here to stay, this article highlights these taxpayers' situations and the types of decisions that the tax professionals participating in the NTCP must make to correctly address these taxpayers' issues. Since many taxpayers who seek help from the NTCP are in extraordinary circumstances, it is pertinent to consider whether AI would have the capability to assist them, especially since AI training data is unlikely to contain examples of such circumstances. Implications exist for other jurisdictions, including New Zealand, that may adopt a program like the NTCP in the future. Implications also exist for any jurisdiction that uses AI to any extent in tax administration and service provision. This research uses auto"‘ethnographic methods to present seven NTCP case studies that the researchers have reflected on regarding whether AI could have addressed each taxpayer's issue and, if so, how it might have done so. It demonstrates contrary to extant literature, that either AI or human involvement can decide on the correct course of action depending on the nature of the case rather than both working together. This is because a solution to a case either requires cognitive and analytical skills which AI is currently well suited to, or it requires an intuitive or common"‘sense solution that considers the social and emotional context to which humans are relatively more well suited. This means that with current AI capabilities, there is a potential role for appropriately trained AI to aid tax professionals in cases where cognitive and analytical skills are required thereby freeing up resources to carry out other tasks. Importantly, this research highlights the potential dangers of ceding too much responsibility to AI in helping vulnerable taxpayers.