IRS has primarily targeted poor families with audits because it’s easier and cheaper than chasing down the complicated tax affairs of wealthy claimants. Technology could help level the playing field
In his latest budget proposal, President Joe Biden will unveil a series of new tax hikes for wealthy Americans. Instead of raising taxes, maybe he should just focus on collecting what some of them already owe.
Every year the IRS takes in $600 billion less than it should. By one estimate, half of this is due to underreporting by those among the super-rich who hide their income by forming elaborate partnerships or other ventures. If you’re worried about the size of the US debt, these numbers should get your attention.
The IRS just doesn’t have the resources to hunt them down. After years of budget cuts and understaffing, the IRS has mostly targeted poor families with audits because it’s easier and cheaper than pursuing the complicated tax affairs of wealthy claimants. But artificial intelligence could change that balance of power and help the archaic, ailing agency better chase the real money.
The Inflation Reduction Act allocates nearly $80 billion to the IRS over the next decade. Once nominated IRS commissioner Daniel Werfel is confirmed, one of his first tasks should be to allocate some of the funds to unlocking AI to overhaul the entire audit process.
Take corporations structured as partnerships, where audit rates have dropped to 0.05 percent and the average tax rate is just 16 percent. (The highest state income tax rate is 37 percent.) According to a recent study led by Stanford University economists, about 15 percent of partnerships are complicated — meaning they can build LLC on LLC on LLC and so on and have overlapping partners.
Some efforts are underway, but it’s still very difficult for the IRS to determine whether these complicated partnerships are reporting the correct amount of income. And many of the agency professionals specializing in this area have retired or are about to retire.
But looking at more than 7 million partnership companies from 2013 to 2015, the researchers found that machine learning successfully helped predict which companies were noncompliant — in other words, they didn’t pay all of their tax debts. This research shows that AI has the potential to peel off the layers more easily and efficiently, alerting human agents to non-compliant partnerships that they could pursue.
The IRS is fairly tight-lipped about the AI, or machine learning, it currently uses on the enforcement front, but during a 2018 webcast, the agency revealed that the technology helped it root out certain violations in minutes. It used to take people weeks or months to do this.
Honest taxpayers should rejoice at the prospect. Today, too many compliant payers are saddled with unnecessary audits. It would benefit both taxpayers and the agency to stop wasting time on this painful process when it is not necessary. AI could recognize patterns and lead auditors to audits that pay off.
Still, there are some caveats. We are not headed for a future where the IRS is run by green-visored robots. AI can only expand, not replace, the capacity of IRS examiners. As Janet Holtzblatt, a senior fellow at the Urban-Brookings Tax Policy Center, put it, “People still have to be the teachers and evaluators.”
The Netherlands is a good example of how relying on AI can bring new problems. In 2013, the Dutch tax authorities started using a self-learning machine algorithm to check that childcare grants go to the right recipients. The algorithm suffered from an ingrained racial bias and innocent families were forced to return their loans without appeal. (The prime minister and his entire cabinet resigned in 2021 following the scandal.)
In the US, new research shows that the IRS’ current algorithms can also be discriminatory. A new working paper shows black taxpayers are more likely to be scrutinized than other taxpayers. In the most egregious example, a single Black man with dependents applying for the Earned Income Tax Credit is nearly 20 times more likely to be screened than a non-Black applicant who is married and filing jointly. whoops
But that doesn’t mean we should stop using technology to improve tax compliance. Daniel E Ho, an economist at Stanford who worked on both this paper and the one on complex partnerships, told me, “There’s this concern about machine learning, but it can also lead to the discovery of differences in established legacy systems. Basically, machine learning helped uncover the injustice, and now it’s up to humans to fix it.
If applied correctly and given proper oversight, AI could go a long way toward making IRS auditing fairer, more focused, and more profitable for the US government. There’s nothing Orwellian about that. It’s progress.
Alexis Leondis is a Bloomberg Opinion columnist covering personal finance. Views are personal and do not represent the status of this publication.
Photo credit: Bloomberg
Alexis Leondis is a Bloomberg Opinion columnist covering personal finance. Views are personal and do not represent the status of this publication.
