AI experts will crack down on hundreds of billions of dollars in tax loopholes

A collaborative team of artificial intelligence and legal experts at Johns Hopkins University is developing an artificial intelligence system that can identify tax loopholes more effectively than tax accountants. The aim is to reduce the annual tax gap caused by manipulation of tax laws, estimated at around $500 billion.

The research team is developing an artificial intelligence system called Shelter Check. The software will enable Congress, the IRS or the 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 havens,” said lead researcher Johns Hopkins University, who specializes in artificial intelligence Computer scientist Benjamin Van Durme said in a Johns Hopkins Center article. “We want to build a system that can read proposed changes to the law and notify Congress and the IRS about the tax law’s impact, or alert those developing new policies about unintended side effects.”

Professor of Law at the University of Maryland and Van Durme was joined by tax attorney Andrew Blair Stanek and student Nils Holzenberger, also a doctoral student in Van Durme's computer science program.

The intricate web of U.S. tax laws sometimes allows taxpayers to avoid tax obligations through a combination of various rulings from Congress, the Treasury Department, the IRS, and related court rulings.

Creating an AI that can understand and apply this complex tax code as proficiently as a human tax professional can be a challenging prospect. The legal language is convoluted, and tax laws include thousands of pages of tables that are critical to interpreting tax results. In the early stages of the project, the team had experimented with ChatGPT and GPT-3, but both AI models were stumped by tax laws.

"GPT-3 is completely confused by the tax code," Blair Stanek said in a Hub post. "Literally, flipping a coin on these issues would get 50% of the problems, and GPT -3 only gets about 70%. These are just basic questions about tax law, like 'So-and-so is a dependent and makes $100,000 a year, does this tax portion apply?' It can't be handled."

Despite these initial difficulties, the researchers remain optimistic that preliminary experiments using GPT-4 are promising. GPT-4 is OpenAI’s latest large-scale language model with expanded inference capabilities.

The team is concerned that developing similar artificial intelligence systems may uncover more tax loopholes, so the progress of the project has attracted much attention.

However, the researchers also believe that their AI could be adapted for wider applications in areas such as medicine and business, underscoring the far-reaching impact of their work: "I plan to spend the rest of my career trying to make It works," said Blair Stanek.

The above is the detailed content of AI experts will crack down on hundreds of billions of dollars in tax loopholes. For more information, please follow other related articles on the PHP Chinese website!