Future of AI in the Tax World Prompts Both Excitement and Fear

Artificial intelligence and robotics have a long history in the world of taxation. In the early 1970s, Harvard University assistant professor Thorne McCarty—sometimes referred to as the father of AI and law—developed a program called Taxman, which dealt with group reorganizations under the US Internal Revenue Code. McCarty saw corporate tax as ripe for “rule based reasoning” as it was “drained of much of the content of the ordinary world.”

In more recent times, AI and robotics have featured in the plans of revenue administrations for understanding large pools of data for risk assessment purposes and for creating chatbots to answer questions on taxation law. However, with the recent advent of large language models such as ChatGPT, the capabilities of new AI are outrunning the legal and ethical structures for dealing with risks presented by its misuse.

This is true of the world of taxation. Clearly there are potential positive uses of AI. It has been useful for revenue authorities for detecting fraud patterns for value-added tax refunds, identifying phoenix activities where businesses are liquidated and then rise from the ashes in a new form, mapping money trails from millions of documents obtained through data leaks, and assessing aggressive or fraudulent tax practitioners.

AI also can benefit both businesses and revenue authorities in finding likely data errors and in real time analytics to understand the business through a taxation lens. It can assist in understanding documents to determine likely taxation impacts, including withholding taxes, and in taxation law insights, including how similar expenses are treated under different tax codes throughout the world.

It also has been argued that AI could assist in applying the principal purpose test in tax treaties arising from the OECD/G20 BEPS 1.0 initiative.

Ethical Framework

While the potential benefits of the use of AI in a taxation environment are substantial, these benefits can only be realized if AI is placed in an ethical framework.

In the Netherlands, a childcare benefits scandal known as “toeslagenaffaire” uncovered racial bias in a Dutch tax administration system designed to detect childcare fraud based on machine learning. This led to the resignation of several government officials.

In a revenue context, AI should not be left unsupervised. The Australian Taxation Office has framed its use of AI in strong ethical practices, as explained by Rebecca Saint, the deputy commissioner of public groups, in an address delivered to the KPMG Asia Pacific Tax and Legal Summit on May 24, 2023. The ATO believes decisions, and the logic behind them, should be known and explainable. The use of AI across a wide spectrum of taxpayers, be they businesses or individuals, should be piloted and then progressively applied across a wider population.

Incorrect decisions must be rectified rapidly, and the potential benefits should greatly outweigh negative outcomes. There needs to be human supervision and processes to test for bias and help ensure that it doesn’t take place. Rita de la Feria and Maria Amparo Grau Ruiz have recently argued that one of the inherent potential sources of machine learning bias is the confusion of correlation and causation.

Furthermore, privacy concerns need to be respected. Many AI tools collect data from a large number of sources. Such data can reveal the private lives of individuals and needs to be secure from hacking. While this applies to AI generally, it’s particularly acute for taxation information.

AI can indeed create biases, but as Harvard Professor Cass Sunstein argues in an article titled Algorithms, Correcting Biases, AI can help reveal and thus overcome the effects of cognitive biases in systems. One can view our AI future generally with both excitement and fear, and this is certainly true of AI and taxation. We must be cleverly ethical and ethically clever. Otherwise, here be dragons.

This article does not necessarily reflect the opinion of Bloomberg Industry Group, Inc., the publisher of Bloomberg Law and Bloomberg Tax, or its owners.

Author Information

Grant Wardell-Johnson is KPMG Global Tax Policy Leader.

We’d love to hear your smart, original take: Write for us.