AI Experts Aim to Combat Tax Loopholes Costing Billions
A collaborative team of AI and law experts from Johns Hopkins University is working to develop an artificial intelligence system capable of identifying tax loopholes more effectively than tax accountants. The aim is to reduce the annual tax gap, estimated to be around $500 billion, that results from the manipulation of tax laws.
The research team is developing an AI system named Shelter Check. This software will enable Congress, the IRS, or courts to scrutinize proposed tax legislation or rulings for unintentional loopholes.
“That's why we call it Shelter Check—it's like a spell checker, but for tax shelters,” said lead researcher Benjamin Van Durme, a Johns Hopkins computer scientist specializing in AI, in a Johns Hopkins Hub article. “We want to build a system that could read proposed changes in the law and inform Congress and the IRS about ramifications for the tax code or warn people writing new policies about unintended side effects.”
Van Durme is joined by Andrew Blair-Stanek, a University of Maryland law professor and tax attorney who is also a Ph.D. student in Van Durme’s computer science program, along with student Nils Holzenberger.
The tangled web of U.S. tax laws can sometimes enable taxpayers to circumvent tax liability by combining various rulings from Congress, the Treasury, the IRS, and related court decisions.
Creating an AI that can understand and apply this complex tax law as expertly as a human tax professional could be a challenging prospect. The legal language is intricate, and the tax code includes thousands of pages with tables essential for interpreting tax outcomes. In this early stage of the project, the team has experimented with ChatGPT and GPT-3, but both AI models were stumped by the tax code.
“GPT-3 was completely baffled by the tax code,” said Blair-Stanek in a Hub article. “Literally flipping a coin on these would get 50% of the questions we were asking right, and GPT-3 only got about 70%. And these were just basic questions about the tax code like 'so-and-so is a dependent, makes $100,000 a year, does this tax section apply?' It couldn't handle it.”
Despite these initial difficulties, the researchers remain optimistic, and preliminary experiments have been promising when using GPT-4, OpenAI’s latest large language model that boasts expanded reasoning capabilities.
The heat is on for making progress with this project, as the team is concerned that corporate-funded efforts may outpace them in developing similar AI systems that could potentially find even more tax loopholes.
However, the researchers also believe their AI could be adapted for broader applications in fields like medicine and business, highlighting the far-reaching implications of their work: “I’m planning to spend the rest of my career trying to make it work,” Blair-Stanek said.
This article originally appeared on sister site Datanami.