acm-header
Sign In

Communications of the ACM

ACM News

Predicting Lawsuits


View as: Print Mobile App Share:
Deep learning yields the ability to forecast the potential for litigation.

A new type of software uses deep learning to help keep companies out of court.

Credit: rippleshot.com

A software developer has come up with a deep learning program that can sift through employee emails to sniff out potential lawsuits before they happen.

Slated for release early this year, the application—created by an attorney, with help from a software programmer—promises to help shield corporations from the millions of dollars in litigation risk they face each year.

"We're currently pricing our system based on the number of attorneys in the corporate legal department of our clients," says Nick Brestoff, attorney and founder of Intraspexion.  "Where a department has one to nine attorneys, we charge $7,500 per month.; for 10 to 39 attorneys, $15,000 per month, and for departments with more than 40 attorneys, $25,000 per month."

Brestoff's software, now readying for a pilot test at a New York Stock Exchange-level corporation, is one of a number of new applications attempting to leverage deep learning to predict the future.

For example, researchers led by Carlos Leon, research and development manager in the Financial Oversight Department of Banco de la Republica (the central bank of Columbia), are experimenting with deep learning software to identify banks that appear destined to fail, based on their balance sheets.

Meanwhile, a research team led by Australian researcher Stewart Jones is working with deep learning to identify corporations that are most likely headed for bankruptcy.

And, Croatian researchers led by Nina Pavlin-Bernardic are experimenting with deep learning to predict which schoolchildren will most likely turn out to be gifted.

One of the reasons deep learning is burgeoning is that there are a number of open source deep learning software packages that programmers can use for free to create custom applications.

Brestoff's Intraspexion, for example, is built on Google TensorFlow, a free, open source, deep learning software developed by researchers and engineers on the Google Brain Team. TensorFlow is flexible enough to be run on a single smartphone, or across thousands of computers in datacenters spanning the world.

"TensorFlow is quickly becoming a viable option for companies interested in deploying deep learning," says Rajat Monga, engineering leader, TensorFlow at Google.

Brestoff's customization of the TensorFlow software (now patented) is programmed to sniff out employee emails for potential employee discrimination suits, simply because those suits are among the most common. However,  Brestoff says he can easily rework his code to do the same kind of monitoring for breach-of-contact suits, fraud suits, and more than 150 other categories of lawsuits that businesses must avoid each day.

Brestoff says he trains the deep learning algorithm of his software by feeding it reference text already identified as discriminatory; then, he asks the software to search for text in company emails that matches, or very closely matches, that reference text.

Interestingly, Intraspexion does not find those matches by seeking out keywords (Brestoff is not a huge fan of keyword matching).  "I know that keyword search had been found by Blair and Maron to be a leaky boat," Brestoff says, referring to a paper published in the March 1985 issue of Communications. "Blair and Maron found that they were finding only about 20%" of the matches they should have been discovering when using keywords as an identifier, Brestoff says.

Instead, Intraspexion identifies problem emails by viewing each letter in an email the same way a computer truly sees that letter: as a unique number string. The software compares all those unique number strings, along with the unique way in which those number strings relate to one another in that email, against reference text the algorithm already knows is discriminatory.

Of course, there's also some hardware involved:  "CPUs with GPU cards," Brestoff says.  "We're currently using P2s in Amazon Web Services.  But we've also used an NVIDIA Tesla K40 GPUs with about 2,500 cores and close to 12 GB in GPU memory."

The result?  In tests so far, Instraspexion's accuracy has clocked in at 99%, Brestoff says.

Moreover, once a problem email is found, mitigating the risk is simply a matter of getting the company's legal department to take a look at it and respond with appropriate precautions.

All told, Intraspexion is a rather impressive go at helping corporations protect themselves in a world fraught with litigation, according to James P. Groton, retired partner of Eversheds Sutherland, a global law firm with offices in 29 countries.

"The Intraspexion software penetrates so deeply into a company's big data that it can safeguard/protect businesses from virtually any kind of potential risk that can be described in words," Groton says.

Adds Thomas D. Barton, a professor at the California Western School of Law in San Diego,  "Intraspexion software has breakthrough potential for attorneys who want to practice more preventively.  I see a strong future for this risk-alerting software."

Joe Dysart is an Internet speaker and business consultant based in Manhattan. 


 

No entries found

Sign In for Full Access
» Forgot Password? » Create an ACM Web Account