To close tax loopholes that cost the federal government billions of dollars each year, tech and legal experts are working together to develop artificial intelligence that can find loopholes better than an army of top tax accountants.
While the task is daunting — a recent test with ChatGPT where the tax code was “totally baffled” — the leaders of the Johns Hopkins University-based effort believe it is doable. Not only do they believe there is, but they worry that corporate-funded efforts will overwhelm them and harness their power. Find even more tax loopholes with our unique AI.
Benjamin Van Dahme, a computer scientist specializing in AI at Johns Hopkins University who is leading the effort, said: “The rules of the game are that if you are operating within the law, It’s okay to bend the law to save taxes.” “It’s very difficult to predict how laws will all work in the real world. The point of using AI is to be able to infer these laws before they are locked down.”
What is understood as “tax law” is a vast and unwieldy set of laws enacted by Congress, Treasury regulations, Internal Revenue Service rulings, and court decisions. Even if everything was enacted with the best of intentions, workarounds exist for taxpayers smart enough to spot it.
“The problem is that smart taxpayers combine one ruling with another to come up with ideas that save a lot of money,” says Andrew Blair Stanek, a tax accountant turned law professor who is part of the AI team. said. “This is a combination of all these tax laws from smart lawyers.”
The IRS estimates a tax gap of about $500 billion annually, partly due to these clever combinations. Wanting to fill the gap with better technology, Blair-Stanek enrolled at Johns Hopkins University where he earned his doctorate in computer science in Van Dürme’s lab. They recruited another student, Nils Holzenberger, to help create an AI system called Shelter Check.
The idea is to create software that Congress, the IRS, or the courts can use to easily scan proposed tax laws or rulings for loopholes they may inadvertently create.
“That’s why we call it Shelter Check. It’s like a spell checker, but for tax shelters,” Van Durme said. “We want to create a system that can read proposed law changes, notify Congress and his IRS about the impact of tax laws, and warn people writing new policies of unintended side effects. I have.”
Such advanced analytics require machines that can not only read, parse and understand complex tax laws, but blend it all together like an accountant or tax attorney.
The project is in its early stages and the team is exploring language processing options and testing them based on the lab’s extensive work in machine learning. Yet they have not found a way to conquer legal terminology.
Despite recent advances in language-based artificial intelligence, the latest technology is no match for tax law. It’s complicated. Then, thousands of pages of documents are littered with tables essential to interpreting tax results.
The team is currently experimenting with language prediction techniques such as ChatGPT and GPT-3. The tax law baffled both.
“GPT-3 was completely embarrassed by the tax code,” said Blair-Stanek, a law professor at the University of Maryland. “Literally a coin toss, we got 50% of the questions we were asking right, and GPT-3 only got about 70%. It was just a basic question about tax law, like earning dollars, etc. Year, does this tax section apply?
Still, the team feels a solution is possible. For example, his recently published first experiment with GPT-4 suggests that progress is being made, but there is still work to be done.
“We haven’t yet found an easy way to make judgments for human tax professionals with high accuracy, but we are making progress,” Van Durme said. “But if we’re moving forward, it’s certainly possible that the efforts of larger, more well-funded companies could accelerate it.”
The team is concerned that the companies profiting the most from tax shelters are also working on this kind of artificial intelligence and may already have it. Team members believe their AI can also be applied to broader legal applications in medicine and business.
Blair-Stanek said, “I’m going to spend the rest of my career making it work.”
