�ʹ�����

From medical imaging to education, artificial intelligence (AI) is at the core of a technological revolution across sectors. But in some areas, AI potential is still limited by the lack of trustable models, capable of discerning technical and sensitive issues that should inform decision making. Law is one of these areas.

Systems like ChatGPT, known as large language models (LLMs), are trained to understand and generate language, like a human would. They learn from available data, such as online content, and from their interactions with users. But, by counting on data containing biased or false information, LLMs can generate flawed content and present it as true.

Current legal AI models can experience hallucination rates (when AI generates false information and presents it as real) as high as 66%, and a tendency toward “sycophancy,” parroting back what models think users want to hear rather than offering critical analysis. They also lack the customizable workflows, domain-specific language, and legal reasoning practitioners need. 

Solving these challenges is the goal of , a no-code AI platform tailored for legal applications created by �ʹ�����’s Conflict Analytics Lab (CAL) in 2023. The project is receiving over $1.3 million in support, including $462,160 from the Natural Sciences and Engineering Research Council (NSERC) Alliance Advantage and Mitacs Accelerate grants – which aims to foster university collaborations with private, public-sector, and not-for-profit organizations – and $400,000 from the Social Sciences and Humanities Research Council (SSHRC) Insight Grant program, plus contributions of project partners. Led by Samuel Dahan (Law), the team will use the funding to advance legal AI models, making them more reliable, customizable and accessible. 

OpenJustice was specifically trained on different legal systems from Canada, the U.S., and France, which improves reliability. To create a community of users, CAL has established partnerships with law schools, courthouse libraries, and self-help centers such as Pro Bono Canada. Each organization brings new users to OpenJustice. While users benefit from the open access to legal information, they also provide high quality feedback on how to improve the platform.

But what can bring OpenJustice to the next level, according to Dr. Dahan, is to get lawyers to instill their practice, knowledge, and experience into the framework, helping to build the AI model from scratch. In February, during a Legal AI Hackathon hosted by CAL and the Stanford Center for Legal Informatics, OpenJustice officially became open source, meaning users can now revise and change the code to make it suitable for their needs. Over 40 custom models have been developed by OpenJustice users since then.