AI used in Canlii to automatically classify and summarize cases
Lexum AI launched a new AI tool on Canlii, a function where AI can generate case summaries and summarize legislation. The tool was piloted using Saskatchewan data, and Lexum will roll out to Alberta and PEI in 2024.
The “CatLii” pilot project began earlier this year. The Lexum blog describes CatLii as “case analysis merges the power of artificial intelligence with the grace and curiosity of a cat.”
The AI tool began to summarize Saskatchewan case law in April. The Law Foundation of Saskatchewan partly funds the project, and any future Saskatchewan content uploaded to CanLii will be summarized using AI, including administrative tribunal decisions. AI analysis of legislation and regulations will be coming. The Law Foundations in Alberta, Manitoba and PEI will provide funding for its provincial case law summaries.
Most Read
“We initially wanted to test the waters,” says Pierre-Paul Lemyre, VP of business development at Lexum. “We now hope other provinces will jump on board so we can have full coverage across Canada.”
Building AI into legal research
Lexum has been deeply invested in AI. Lexum launched its automatic classification system earlier this year using court case data from Saskatchewan and Ontario. Lexum’s generative AI program, LexKey, used Canlii’s millions of cases, commentaries and legislation to train the program to choose the right keywords. LexKey automatically adds subject tags to each case, including areas of law, legal topics, jurisdiction and level of court. Eventually, the tags will be clickable so users can view all cases based on the topic.
The next step is case summaries. Canlii’s AI program reads cases, creating summaries including facts, a list of legal issues discussed, an analysis of each party's arguments, and the outcome.
Lexum took this approach as part of its mandate to make legal information more accessible to lawyers and the public.
“It’s a way to quickly grasp what a case is about, and it speeds up the process of legal research,” says Lemyre. “The work we’ve been doing is to use AI to automate legal services that are too expensive to complete manually at large volume. Now, it becomes possible to open any case and have quick and easy access to a list of the legal issues involved or the complete procedural history.”
Challenges in making AI accurate
Hallucinations, when AI creates false data to fill gaps, remain an issue in generative AI. Lemyre says hallucinations won’t be an issue because only data from a case file can be used to create the summary. The case summaries also have links to citations within the document showing the source of the analysis. “We’re not relying on any pre-existing knowledge in our use of the large language models (LLMs),” says Lemyre.
One of the challenges is how AI analyzes complex cases. Lemyre says cases with simple legal issues are easier to summarize than those with multiple, complex issues. “When there are several legal issues in a case, sometimes it can mix them up or ignore some of them,” says Lemyre. “For now, it does better with simpler cases that address up to maybe 2 or 3 legal issues, but it is getting better and better.”
The program has a 90 percent accuracy rate. Lexum asks users to contact them about any errors in the summaries.
“It’s not perfect, but it’s more than adequate to go forward,” says Lemyre. “CanLII helps over 2 million visitors a month search across 3 million documents. It enables us to provide them with added-value information in an efficient manner. Users are able to verify the summaries in real-time, so it’s totally transparent. In this context, we’re confident to keep moving with this technology.”
