Robo-justice

TOMORROW, WHEN John Roberts appears before the Senate Judiciary Committee for the start of his Supreme Court confirmation hearings, the senators will apply themselves to the futile task of looking into the future. How might a Justice Roberts rule on an abortion case? A school prayer case? A DNA profiling or human cloning case? For Roberts, who has written few opinions during his brief tenure as a federal appellate judge, the crystal ball is especially dim, but even the most explicit paper trail would be no guarantee of his future opinions.

That even Supreme Court justices, the high priests of the constitutional order, can recant and reconsider is as vivid a reminder as there is of the fundamentally human nature of the law. From traffic court on up to the high court, our laws, in all their power and complexity, are interpreted and applied by sometimes fickle, always fallible people.

But what if ''John Roberts" were not a human being but a piece of artificial intelligence software--a robot with the whole of law mapped out neatly in his circuit boards? Granted, the resulting Roberts-bot would not offer the personal charms of the wry, unflappably affable human Roberts. But consistency, predictability--surely those count for something.

Computer judges, of course, aren't going to be ascending to the bench in the foreseeable future. ''Nobody thinks that's a good idea," says Carole D. Hafner, a Northeastern University computer scientist and pioneer in using artificial intelligence to study the law. Judging, and most especially Supreme Court judging, is a complex and subtle mix of imagination, acuity, and political calculation. Still, at a time when doctors are starting to use software to aid in their diagnoses and when hedge funds are using computer models to make multibillion-dollar investment decisions, there is growing interest--even in an American legal establishment usually resistant to change--in finding ways to incorporate artificial intelligence into the law.

In the last few years, as a number of studies, and even a few commercially available products, have set out to demonstrate how our increasingly powerful computers can assist in the practice of law, the computer scientists and legal scholars who work in this small, emerging field believe they are doing something revolutionary: making the legal system more transparent, more efficient, and more fair.

. . .

Put simply, artificial intelligence is the branch of computer science that deals with getting machines to think like human beings. Its application to the law dates, in its earliest form, to the 1950s, when mathematicians first tried to use formal logic to model legal reasoning. Enthusiasm about its possibilities reached a peak in the 1980s, with suggestions that computers might soon be arguing cases, or even deciding them. But when those predictions proved hugely premature, the field's reputation suffered.

Today, true AI of the Hal 9000 variety remains the stuff of space operas, but programmers have started incorporating some forms of machine intelligence into software, giving it the ability to reason and learn in a rudimentary way. Once again, the law--with its web of rules and its gradations of interpretation--has provided a tempting target for computer scientists looking for ways to explore human logic. According to Northeastern's Hafner, ''the law fascinates computer scientists interested in understanding how people argue." Many see in it, she says, ''a model that applies to human decision-making more broadly, that might be applied to understanding how people argue with themselves when they have to make a decision."

Some of the fruits of this fascination, however, have been decidedly practical, from intelligent document retrieval systems that use fuzzy logic to search not just by keyword but by concept (the only AI application widely used in American law firms) to programs that predict the outcomes of court cases or evaluate potential clients.

The computer scientists John Zeleznikow of the University of Melbourne and Andrew Stranieri of the University of Ballarat, for example, have developed two pieces of legal software currently in use in their native Australia. One, SplitUp, calculates with impressive accuracy the likely results of divorce proceedings--its effect has been to encourage settlements, thus preventing unnecessary litigation. Another, GetAid, is used by an Australian government agency to appraise applicants for legal aid--a complicated calculation based on employment history, household income, the likelihood that the case will be won, and myriad other factors.

For Zeleznikow, such programs promise both to make lawyers more efficient and untangle a court system that's tied up in endless administrative procedures. Much of the work done by lawyers is the application of relatively straightforward statutes or the drafting or relatively standard documents, tasks that Zeleznikow and other similarly minded programmers believe can easily be handled by today's AI. The aid applications that GetAid now reviews, Zeleznikow points out, used to take up the vast majority of the aid agency's time and money. ''So now, rather than have a lawyer dealing with 100 applications a week, they're only having to deal with about 20 applications a week. Eighty percent of somebody's time that used to be processing applications now can be turned to trying cases."

But some supporters present AI-based legal applications not only as a lawyer's tool but a lawyer's surrogate. Some, like Zeleznikow, point out that since AI won't approach the abilities of a human lawyer anytime soon, this isn't a threat to the legal profession. (Zeleznikow, for one, is also careful to protect human agency in the design of his programs: GetAid, for example, can only approve applicants, passing along the rest to a lawyer to deal with.) Even crude AI legal advice is better than nothing, the argument goes, and could help level the playing field in the many cases where one party can't afford legal advice at all. ''Really the worst thing about the legal domain, and it's more serious in the US, is the actual cost," says Zeleznikow. ''Not everybody has access to a lawyer, especially access to an expensive lawyer."

But Marc Lauritsen, president of the Massachusetts-based legal technology consultancy Capstone Practice Systems, believes that AI's foray into the law will usher in a period of healthy tumult. ''We're just at the beginning of something that's going to disrupt the comfortable legal profession," he says. He points to online companies like Legal Zoom that offer basic interactive contracts for everything from patent filings to prenuptial agreements. They're the precursor, he says, for a whole new sort of lawyerless law firms, ''organizations out there saying, 'Hey, you don't need a lawyer, come to us and we'll take care of it."' As legal software gets more sophisticated, he believes, it will eat more and more into the work of lawyers.

One current area of interest for programmers, according to Northeastern's Hafner, is designing programs that could not only draft contracts but enforce them: ''If the contract is represented in computer-understandable terms and the transactions take place online," she says, ''there is the idea that the computer could monitor compliance."

In such an imagined future, legal advice would become a commodity. As Richard Susskind, the information technology advisor to the Lord Chief Justice of England, wrote in his 2000 book ''Transforming the Law," ''I envisage in the world to come that much of the lawyer's work will shift from being advisory in nature to becoming, in large part, a form of information service.... Much of today's conventional legal work will be systematized, routinized, and proceduralized." Customers could shop around for legal advice, that is, the way they now hunt for a good mortgage.

And, while computers aren't likely to replace judges, they might end up restraining them. Uri J. Schild, a computer scientist at Bar-Ilan University in Israel, has developed a software system that weights the importance of previous crimes to aid judges in criminal sentencing. It's only a tool, Schild acknowledges, and it promises to take some of the caprice out of the process, but it's easy to see how its adoption in the United States might strengthen those politicians currently pushing for stricter federal sentencing guidelines.

For Susskind, though, the liberty of judges is not the point. ''I belong to the school that thinks the law should be as reliable as possible," he said in an interview. If AI makes the law more bloodlessly uniform, he argues, so much the better: ''You should find out with almost unerring accuracy whether what you're saying or doing is within the scope of the law."

It's a high standard, and may not be humanly possible. Whether it's nonhumanly possible only time will tell.

Drake Bennett is the staff writer for Ideas. E-mail drbennett@globe.com.