Tech and law experts are collaborating to develop artificial intelligence capable of detecting tax loopholes more effectively than a large team of high-caliber tax accountants to eradicate tax loopholes that cause the federal government to lose billions of dollars annually.
Leaders of the effort based at Johns Hopkins University are working on developing artificial intelligence capable of detecting tax loopholes, causing the federal government to lose billions of dollars annually. Although the task seems challenging, the team is determined to complete it before a corporate-funded effort beats them to it and uses their AI to find even more tax loopholes. Benjamin Van Durme, a computer scientist specializing in AI, is leading the team and believes that AI can predict ways the law will work in the real world before laws get locked down.
Tax law is a vast and complex array of laws enacted by Congress, Treasury regulations, Internal Revenue Service rulings, and court decisions. Even if all of it had been enacted with the best intentions, there are workarounds for taxpayers who are clever enough to find them. These clever pairings by taxpayers result in the annual tax gap estimated by the IRS at about $500 billion. The team hopes to narrow this gap with better technology and is creating an AI system called Shelter Check that can easily scan proposed tax legislation or rulings for loopholes they might unintentionally create.
Shelter Check is like a spell checker but for tax shelters. The team wants to build a system that could read proposed changes in the law and inform Congress and the IRS about the ramifications of the tax code or warn people writing new policies about unintended side effects. The team includes Andrew Blair-Stanek, a tax attorney turned law professor, and Nils Holzenberger, a student recruited to help create Shelter Check.
Developing an AI capable of finding tax loopholes like an accountant or tax lawyer will require a machine that can read, parse, and understand complex tax laws. The project is still in its early stages, with the team relying on the lab’s machine-learning expertise to experiment with language processing options. However, the complexity of the legal language in tax code and thousands of pages of documentation peppered with tables pose a significant challenge to language-based artificial intelligence, such as GPT-3.
Despite recent advances, the latest AI technology struggles with the tax code. Even GPT-3 failed to answer basic questions about the tax code with high accuracy, achieving only 70%. Nevertheless, the team believes that progress is being made and experiments with GPT-4 have shown promise. However, the team fears that corporations, which have the most to gain from tax shelters, may already be working on such AI solutions.
The team hopes that their AI, called Shelter Check, will be adapted for use in other fields, such as medicine and business law. Although producing the judgments of a human tax professional with high accuracy remains a challenge, the team is committed to making progress on this project. Blair-Stanek, one of the team members, plans to devote the rest of his career to this endeavor.
