Alfred Bester’s classic science fiction novel, The Demolished Man, presents a future society where the Mosaic Multiplex Prosecution Computer — called “Mose” for short — has to approve every criminal charge before it can go to trial. He keeps the cops honest. Powell, a telepathic detective, tells the police commissioner, “You know Old Man Mose. He’s going to insist on hard fact evidence.”
Decades later, artificial intelligence really is getting into the legal field. The most impressive entry is a cloud service called ROSS. It isn’t actually an AI lawyer, in spite of attention-grabbing headlines. It hasn’t been admitted to practice in any state. But it’s a first-class paralegal, addressing many real-life issues and making a difference in negotiations and trials. Toronto-based ROSS Intelligence claims it lets lawyers “do more than humanly possible.”
An AI system is free of the preconceptions that a lawyer might hold.
Several law firms have started using ROSS. Baker & Hostetler was the first, using it to handle bankruptcy cases. It accepts questions in plain English, and ROSS gives an answer that a competent lawyer can easily understand. It finds all the relevant citations and boils them down to a small number of answers. Bob Craig, the firm’s chief information officer, explains, “ROSS is not a way to replace our attorneys – it is a supplemental tool to help them move faster, learn faster, and continually improve.”
Lawyers can talk back to ROSS, challenging its hypotheses. It accepts the feedback and adjusts its subsequent responses accordingly. ROSS could be right when the lawyer is wrong, though. In The Demolished Man, the detective was looking for a profit motive while Mose said there was a passion motive. The detective dismissed Mose’s hypothesis as a glitch, but it proved to be right. An AI system is free of the preconceptions that a lawyer might hold.
IBM’s Watson platform, best known for beating top players at Jeopardy, forms the basis of the ROSS service. Law firms access ROSS as a cloud subscription.
ROSS v. Traditional Software
Software to aid lawyers’ research has existed for many years. The traditional approach has been a static one that doesn’t adapt well to changing laws and to lawyers’ needs. Text searches can find laws that appear to be relevant. Links in databases can help to navigate the complex cross-references which every body of law has. The limitation of this approach is that it’s basically pattern matching, without any understanding of what underlies the law. Systems such as Bloomberg BNA and LexisNexis operate this way.
The more it’s used, the more ROSS learns. It incorporates feedback not just in giving advice on a particular case, but in improving its model of the law.
With no greater level of insight, software can’t tie in related topics. A precedent might be very relevant to a case, yet they might not have any keywords in common. A text-based search wouldn’t find the precedent. A good lawyer can get the relevant information by making the right queries, but the process is tedious. It takes a lot of time to do thorough research with conventional tools. That increases the cost to clients, makes lawyers less efficient, and can miss important information. Just learning how to use those tools well takes a significant chunk of time.
The more it’s used, the more ROSS learns. It incorporates feedback not just in giving advice on a particular case, but in improving its model of the law. Each day, it’s able to give better responses than the day before.
What it can’t do is weigh the reliability of factual information. Is a witness lying? How trustworthy is someone’s memory after five years? Are the measurements of a lot’s boundaries accurate? These are still questions for people to decide. Telling reliable accounts from guesses and lies might someday fall under Watson’s abilities, but not yet.
The growing robolegal field
ROSS isn’t the only robolegal in town. The website DoNotPay opens AI’s legal power up to the American and British public in the form of a chatbot. It’s dealt with a quarter of a million parking tickets and claims to have saved people $4 million in fines. The site has also dealt with landlord-tenant disputes and banking charges. DoNotPay claims to be “the world’s first robot lawyer,” which might get it into trouble with the legal establishment.
It lets attorneys find arguments which not only are objectively good, but are persuasive to the judge or counter the opponent’s arguments.
Lex Machina offers its Legal Analytics software, which pulls information out of vast quantities of litigation information. It extracts patterns to find information that’s relevant to arguing a case. It doesn’t just examine the law, but provides insight into how judges and opposing counsels think. It lets attorneys find arguments which not only are objectively good, but are persuasive to the judge or counter the opponent’s arguments.
Equal justice under AI
The total volume of federal laws and regulations is so big that the Library of Congress despairs of measuring its size. States and localities have their own laws to add to the bulk. Understanding it all is far beyond human comprehension. The ability to give legal advice and aid requires some way of dealing with so much information.
Some people see a “legal apocalypse” in the coming of AI for law research. They wonder where young lawyers will get their practice in research if computers can do it better. But those lawyers won’t miss the drudgery if, instead, they have the time to work with their clients and prepare documents more quickly.
A major complaint about the law is that the people with the most expensive legal teams win. They find the most obscure precedents and draw up the most impressive briefs, and victory too often goes to the party with the biggest bank account. Legal AI could level the field, letting even offices with modest budgets dive deep into the laws, get information on the latest cases, and present arguments that will give their clients a fair hearing.