The IRS bought a $99 million AI tool that will catch all tax evaders through social media

There’s a level of public fear that arises when we hear that the government is using AI in some way. In the past I’ve covered the collapse of Digital Privacy thanks in part to Palantir Technologies and their government contract to end terrorism. This week I talked about the racial biases encoded in AI and how dangerous it can be when applied to criminal justice algorithms.

But the most recent government agency to bring AI into their arsenal, is the one agency we all secretly fear the most: the IRS.

Although not nearly as physically terrifying as, say, the Department of Defense, the IRS is one government agency you don’t want to mess with. If you do, the list of people that beat the IRS is not a very long list – just ask Pete Rose, Willie Nelson, Wesley Snipes, and Martha Stewart.

Essentially, the IRS has invested $99 million in a contract with Palantir Technologies to:

provide hardware, software, and training to “capture, curate, store, search, share, transfer, perform deconfliction, analyze and visualize large amounts of disparate structured and unstructured data.”

Michael Tennant, The New American

In other words, they’re doing everything in their power to make sure no one evades taxes ever again. And they believe this can be done by analyzing disparate sources of data that human brains can’t handle.

This includes, and is not limited to, analyzing social media information, phone records, bank records, credit card transactions, etc. So, if you want to continue keeping all of your Vegas winnings under wraps, you’re going to need to be a lot more careful with the data you’re generating.

Not To Be Taken Lightly

This seems all theoretical, but you must realize that Palantir is a very serious company with unbelievable technology. Not to mention, their track record with government agencies is pretty legit:

I would not take this application of artificial intelligence lightly. And I know for a fact that tax attorneys and accountants are going to be walking on eggshells, knowing that this “all-knowing” system is keeping a watchful eye.

This would allow the IRS to meaningfully link tens of millions of tax returns, billions of information returns, and trillions of bank and credit card transactions, phone records and even social media posts. For example, if a U.S. citizen moves money from a Swiss bank to some other offshore bank, then uses credit or debit cards to spend the money in the U.S., Palantir’s software can link those transactions. It could also flag a person whose tax return shows relatively low annual income but whose social-media posts indicate something entirely different.

Daniel Pilla, WND

“Be careful what you post on social media.” We’ve all heard this before. However, now the phrase is taking on an entirely different meaning.

Out of Necessity

I think the reason the IRS is getting this help from AI is two-fold. One, their mission is to not let a single owed tax dollar slip through the cracks. Two, they are hurting badly (like many government agencies) and need help.

The IRS’s volume of data in 2017 was 100 times larger than the amount in 2007, according to a strategic plan from the agency. In 2018 alone, the IRS Criminal Investigation division collected 1.67 petabytes—or 1 million gigabytes. Approximately 900,000 plain text files can fit into a single gigabyte.

Siri Bulusu, Bloomberg BNA

Add on top of that the sizable dent they took to their budget:

Congress has cut the IRS budget by $715 million since 2010. The IRS Criminal Investigation division cut 522 special agents between 2011 and 2016, according to its fiscal year 2016 annual report. In FY 2018, according to the annual report for that year, another 140 special agents and 51 professional staffers were cut.

Siri Bulusu, Bloomberg BNA

The IRS’s plan to adopt intelligent data analysis algorithms comes out of pure necessity. The contract is set to run from 2018-2022, although I’d expect them to continue it in some capacity for the foreseeable future.

The simple fact is that the window for error is going to get smaller and smaller. Although this tool will be particularly beneficial to finding unaccounted for dollars, nobody is safe from this tool.

Today, the IRS is filing fewer tax-cheating charges than it once did:

However, using Palantir will most likely increase their capacity for criminal pursuit and help them identify the cases it has the best chances of winning and that will yield the highest return. It’s not all doom and gloom from here on out, though.

Already a future thinker?
Then become a friend.

A Better Tax System

In particular, there’s an outlook where I believe the IRS can use this tool to actually make the entire tax collection process simpler. And even put a system in place that makes the criminal pursuit of tax evaders better for all parties.

Undeniably, the IRS is going to be collecting a lot data over the next five years. And one of the most interesting sides of data collection and analysis is finding ways to predict an outcome before it happens. We’re in a time where the outcomes we can predict are just flatout unbelievable.

Perhaps the most well-known research of this kind is John Gottman’s 1992 study of 52 newlywed couples, in which Gottman’s team interviewed them and observed their interactions, then asked them to fill out a questionnaire three years later. Gottman was able to develop a model predicting the likelihood of whether a couple would get divorced with more than 94 percent accuracy.

In 2013, Facebook engineer Lars Backstrom and Jon Kleinberg of Cornell University co-authored a paper identifying a number of factors that contribute to long-term relationship success, such as whether a couple had lots of friends in common or whether they posted a lot of photos together. The researchers found they were able to determine with 60 percent accuracy whether a couple would break up.

It’s a well-established fact in the marketing world that people are more likely to spend lots of money after a bad breakup. “When a relationship ends, people go shopping,” Sasidharan says. It’s also a well-established fact that companies like Facebook track this information — per their own data, users spend 25 percent more on travel-related purchases after changing their relationship statuses.

E.J. Dickson, Vox

Six years later, I would venture to guess that Facebook has gotten really good at predicting the outcome of relationships, months or even years in advance. There’s even a growing population of people that want what’s called “expiration dating” to become a feature of dating sites.

Now, let’s think about this level of prediction in the context of the IRS, who is going to be tapping into a lot of this personal data.

What if the IRS could predict that you were going to evade taxes before it happened? What if they knew you couldn’t afford to pay the looming taxes you were going to owe, some six or more months before your return was due?

They could use this knowledge and foresight to reach out to the people in this predicament and figure out a better solution – a solution that isn’t going to ruin that person’s life and also alleviate the time and resources the IRS uses chasing people down and taking them to court.

Ultimately, people make mistakes. Thousands of athletes, businessmen, and average people overspend what they can afford and then get burned on their taxes every year. In the end, the IRS always gets their money. How they get it, though, could be changed.

Their partnership with Palantir previews a world where tax season isn’t the darkest time of the year. The government works with everyone to get in what they owe and work with people that cannot afford their taxes. This is a way for them to make it simpler and less stressful on all parties involved. And for that, I’m very hopeful.