Working as a legal research attorney can feel a bit like playing a never-ending game of “Jeopardy!” in which you are the only contestant. People constantly fire arcane inquiries your way, covering a diverse array of legal topics, and the monetary stakes for success or failure can get pretty high.
It seems reasonable then that someone who fares well as a legal researcher might expect to put up a good fight on “Jeopardy!”—and vice versa. And no one has ever dominated America’s favorite answer-and-question game show like Watson, the natural language question-answering machine created by IBM that in 2011 famously won a game against the finest human players in the world.
Of course, IBM created Watson with the expectation that its money-spinning potential would extend far beyond game show prizes.
Its unique talents, for instance, are already put to work improving the quality of medical care.
Now a group of young entrepreneurs from Toronto is angling to do the same thing for the practice of law.
Their new venture, ROSS Intelligence, seeks to use Watson’s computing power to answer legal research questions. ROSS’s founders were one of the winners of a contest IBM sponsored looking for the most promising commercial applications for Watson.
As a result, the group was able to gain continued access to Watson’s cloud computing platform to power ROSS. The idea behind it is that lawyers would be able to ask ROSS a question in a natural language, and get back documents that answer the question on point.
Andrew Arruda, the CEO and co-founder of ROSS, said that ROSS takes advantage of Watson’s cognitive computing, in which a computer’s “thinking” is designed to mimic as closely as possible the manner in which humans think. That allows computers to make sense of what computer scientists call “unstructured data,” of which things like case law and statutory text—basically, anything that doesn’t come neatly in a spreadsheet—would be perfect examples.
Cognitive computing is made possible chiefly by two breakthroughs in computer science: natural language processing and machine learning. The former aims to teach computers to do something very difficult for them, but trivially easy for humans: understand how the relationships between different words in a sentence affect their meaning. Arruda contrasted that against conventional legal research software, which relies on mining documents for various keywords prompted by the attorney.
“It allows lawyers to express what they’re researching the way that we communicate with one another, in natural plain English, and instead of bringing back thousands of documents that may or may not have your answer but just have the word you used in your research, ROSS is able to decipher and attempts to really understand the intent of your question,” Arruda said.
Machine learning, on the other hand, allows users of a software program to give the computer feedback on how well it performed a particular task. The computer then incorporates this feedback in order to order to get better at the task on future attempts.
Arruda analogized ROSS’s machine learning to IBM’s experiments in “cognitive cooking.” IBM used Watson to map out the various ingredients used in certain dishes and then tinker with the recipes based on what diners said they liked or didn’t like about certain combinations.
Conducting legal research is more intellectually daunting, of course, than making Peruvian potato poutine (an actual dish from IBM’s food truck). Arruda, however, claimed that ROSS will continue to get better as it interacts with lawyers.
“It’s much like a junior associate. When they come into the firm, you have to take them through the ropes and different things,” Arruda said. “ROSS is continuously learning and getting better, and our goal, our vision, is to have ROSS be that kind of senior partner that you can just talk to, ask questions of, and be able to really assist you in your case.”
That thought may be discomforting to existing senior partners, the sort that are human and well compensated for their expertise. Projects like ROSS and Watson conjure up visions of Kurt Vonnegut’s novel “Player Piano,” where in a dystopian future almost all laborers, even white collar ones, are made obsolete by computers.
Arruda contends, however, that technologies like ROSS will augment and enhance the services of lawyers, rather than replace them. He envisions a future in which lawyers, freed from much of the drudgery of repetitive tasks, have more time to engage in creative legal strategies. Arruda argues that cognitive computing will enable lawyers to reach a large untapped market of potential clients who need legal services but are unable to afford them.
Arruda said that ROSS still leaves it up to human lawyers to ultimately make all the legal decisions, and compared his company’s software to “centaur chess,” a form of chess introduced by grandmaster Garry Kasparov in which a computer program plays in tandem with a human, who is still fully in control of what moves are made. The most advanced computers are still unable to beat such human-and-hardware combinations. (At least for now, anyway.)
“By combining the forces of what you can do with cognitive computing with a lawyer, you’d be able to do much more than we thought was possible, and you address a market that needs to be addressed,” Arruda said. “It also becomes profitable for the lawyers, because using the current tools, there are a lot of inefficiencies, and, I like to say, sometimes effectively the lawyers can’t afford the clients, and the client can’t afford the lawyers.”
Arruda demurred when asked to forecast what cognitive computing might be capable of in five or 10 years, saying that such predictions were inherently difficult. But he noted that the raw computing power of Watson is much more powerful now than when it won “Jeopardy!” in 2011. Then, Watson was housed in large room. Now, it’s slimmed down to about the size of three large pizza boxes. But while Watson’s physical properties are getting smaller, the ambitions surrounding it are only getting bigger.
Follow David Donovan on Twitter @SCLWDonovan