Your Lawyer May Soon Ask This AI-Powered App for Legal Help
When Jimoh Ovbiagele was ten years old, his parents decided to get a divorce. But as the couple got deeper into the process, the legal fees grew more and more expensive, until they ended up abandoning the whole plan. “It had a negative impact on my family,” Ovbiagele says.
In high school and beyond, when Ovbiagele was looking into various career options, he discovered that most of a lawyer’s time is actually spent researching cases. Remembering his parents’ difficulties, that idea nagged at him. Ovbiagele ended up studying computer science rather than law, but when he had the opportunity to pursue an artificial intelligence project at the University of Toronto, he had a pretty good idea of what he wanted to work on.
“I thought back to that big problem lawyers face in their day-to-day work, and how it impacts regular people,” Ovbiagele tells WIRED. “I thought we should apply the capabilities of machine learning to tackle this problem and make things better for lawyers and for clients.”
And so the idea for ROSS Intelligence was born. Ovbiagele became the company’s CTO, and along with co-founders whose backgrounds range from law to neuroscience to computers, the team came up with a voice recognition app powered by IBM Watson, the machine learning service based on the company’s Jeopardy-playing cognitive system, that doles out legal assistance.
The app is yet another example of the ways machine learning is infiltrating our everyday lives. These days, it’s not just AI algorithms themselves that have improved, but the ability to deliver them across the Internet that has made so many new applications possible. Just this week, a toy startup called Elemental Path started taking preorders for the CogniToy dinosaur bot, which also taps into IBM Watson for its brains. SRI International, the Silicon Valley incubator where Apple’s Siri digital assistant was born, recently announced a voice-recognition add-on for mobile banking apps that lets customers ask questions about their accounts. Ross is another incarnation of the trend.
Asking Natural Questions
Ross works much like Siri. Users can ask it any question the same way a client might—for instance, “If an employee has not been meeting sales targets and has not been able to complete the essentials of their employment, can they be terminated without notice?” The system sifts through its database of legal documents and spits out an answer paired with a confidence rating. Below the answer, a user can see the source documents from which Ross has pulled the information; if the response is accurate, you can hit a “thumbs up” button to save the source. Select “thumbs down” and Ross come up with another response.
According to Ovbiagele, it’s a huge improvement over current research databases that rely heavily on keyword search. Plus, he says, the system learns from the feedback its users give and gets smarter with more input.
Work on Ross began last September. Andrew Arruda, another Ross cofounder, explains that the team started with a “blank slate” version of IBM Watson. They fed it thousands of pages of legal documents, then trained it on the taxonomies and ontologies of law using one of Watson’s question-and-answer APIs. Then they built a machine-learning layer of their own dubbed LegalRank—a play on Google’s PageRank algorithm—to further refine the system.
“Without exactly giving away our secret sauce,” Arruda says, “LegalRank can figure out which results get preferential results, whether that’s prioritizing a case that has more citations, knowing that a Supreme Court case should rank higher than a local decision, and other nuances.”
For the moment, Ross focuses on bankruptcy and insolvency law, but Ovbiagele and Arruda remain optimistic about its ability to scale and move into other areas. They say there’s no shortage of demand for ways to make legal research easier: according to their data, attorneys devote nearly a fifth of their working hours to legal research and law firms spend $9.6 billion on research annually. The app could also help unburden much of the basic legal gruntwork that is often outsourced to places like India and the Philippines, where labor is often cheaper and workers have high English fluency.
Without having tried Ross myself, there’s no good way of knowing whether the app works as advertised. But Ovbiagele and Arruda say they’ve tested the system in small-scale pilot programs inside law firms since June, and they’re confident in the results they’re seeing so far. Another indication that the app shows promise: A subsidiary of global law firm Dentons, NextLaw Labs, has signed ROSS Intelligence as its first portfolio company.
Via: Google Alert for ML