IRS using artificial intelligence to sniff for tax dodgers

When U.S. authorities want to ferret out abusive tax shelters, they send an army of forensic accountants, auditors and lawyers to burrow into suspicious tax returns.

Analyzing mountains of filings and tracing money flows through far-flung subsidiaries are notoriously difficult; even if the Internal Revenue Service manages to unravel a major scheme, it typically does so only years after its emergence, by which point a fresh dodge has often already replaced it.

But what if that quest could be done quickly by a computer? Could the federal tax laws be accurately rendered in an algorithm?

New academic research seeks to use artificial intelligence to combat tax evasion by corporate entities, from publicly traded multinationals to private partnerships. The goal is to give the IRS a better way to investigate sophisticated tax shelters that strip billions of dollars from federal coffers each year.

“We see the tax code as a calculator,” said Jacob Rosen, a researcher at the Massachusetts Institute of Technology.

“There are lots of extraordinarily smart people who take individual parts of the tax code and recombine them in complex transactions to construct something not intended by the law.”

A recent paper by Rosen and four other computer scientists demonstrated how an algorithm could detect a known tax shelter used by partnerships.

First, the researchers translated tax regulations governing partnerships, a growing source of tax trickery, into source code.

Then they rendered the transactions underpinning a questionable shelter known as “ installment-sale bogus optional basis,” or IBOB, as a series of codes. The IBOB shelter artificially inflates the basis value of an asset on a tax return to wipe out taxable gains when that asset is sold.

Next, the researchers mapped out in code the tangle of entities that make up typical partnerships. The results flagged specific combinations of transactions and partnership structures that were likely to produce the IBOB dodge.

Large corporations attract most of the attention when it comes to tax avoidance and tax evasion, but partnerships, which have separate tax rules, are a growing source of worry for the authorities.

Commonly used by hedge funds, private-equity funds, real-estate outfits and oil and gas concerns, partnerships are far less likely to be audited than corporations.

A Government Accountability Office report from 2010 said that the IRS knew of 1 million “ networks” involving partnerships and similar entities, adding that “the IRS also knows that many questionable tax shelters and abusive transactions rely on the links among commonly owned entities in a network.”

Rooting out fraud in corporate tax returns takes place largely through data mining, in which the IRS collects data from filed tax returns and analyzes them for patterns. The data go into a database within the agency’s Office of Tax Shelter Analysis.

The data-analytical approach depends upon already having a smoking gun, such as a suspicious deduction on a return.

By contrast, the artificial intelligence approach does not require pre-existing evidence. Instead, it focuses on rule mining, in which individual tax-code regulations are lined up against one another to ascertain whether they can be used collectively to create a sophisticated tax dodge.

Rule mining takes advantage of a tax-shelter feature: While their inner workings are complex, their general aim is usually simple — to lower tax bills by improperly generating bogus losses, deductions, offsets and credits that minus the shelters would not exist.