If Edmonton researcher Randy Goebel has his way, artificially intelligent judges and attorneys will become players in the courtrooms of the future.
A professor in computing science at the University of Alberta, Goebel has partnered with scientists in Japan to develop artificial intelligence programs designed for the legal world.
His team has already designed an algorithm capable of passing the Japanese bar exam. Now the computer scientists are taking their research one step further.
The latest project is new artificial intelligence software that could weigh contradicting legal evidence, rule on cases and predict the outcomes of future trials.
“It’s just extending the research required to more deeply understand language, so you can support or take over some of the things that humans do with legal reasoning,” Goebel said in an interview with CBC Edmonton’s Radio Active.
“The next stage is to become more aggressive and (develop) not just yes-no questions, but do free-form questions.”
Outside the lab, lawyers put the new algorithm to practical use, using it to study legal precedents and navigate the best path to a legal victory.
The program is designed to understand language and provide answers, just like a real flesh-and-blood legal expert — and in some cases may be more accurate due to its ability to reference obscure cases.
Search engines like Google are already commonplace in the courts, and artificial intelligence is the next logical step, said Goebel.
Reducing the research burden for lawyers and clerks will be among the most practical applications.
“A lot of the work that lawyers do in law offices is so-called discovery, trying to find out what data, what documents are related to a case that you’re managing,” said Goebel, who also serves as principal investigator with the Alberta Machine Intelligence Institute.
“Speeding that up by having computer programs that do a more in-depth analysis of the language can save time and money.”
Work on the bar exam software began in 2012. It is now considered a world leader in the field. The software finished first at the 2016 Competition on Legal Information Extraction/Entailment, in international competition that pitted the best AI programs against each other.
The successful program was inspired by failure, said Goebel.
“My colleague started studying law [in Japan], he took a law degree and started writing bar exams and he failed the bar exams,” said Goebel.
“So the question was, why does a human do so poorly? Could a machine do better? That’s an interesting motivation. Building machines that understand language well enough to reason about law is a pretty interesting challenge.”
Building a computer network that mimics the human brain is complex. Designing one capable of interpreting overlapping, often contradicting facts of law, is particularly challenging.
Though computer programs may be the end result of his work, the research is also a way to better understand language and the human mind, said Goebel.
“It’s a very common misconception that AI researchers are trying to solve a particular problem. For the most part, they aren’t. What they’re trying to do is to do is solve classes of problems.
“We don’t know very well how humans use language, and trying to understand it well enough to program it, gives you insight.”