From major law firms to legal aid teams, CoCounsel is putting generative AI to work, making research a fast and natural process
The disruptive potential of generative AI is being debated across every industry. In the legal profession, where vast volumes of written information sit at the heart of many projects, the opportunity is tremendous.
Laura Safdie is the vice president of artificial intelligence at Thomson Reuters, and was previously the chief operating officer and general counsel of Casetext, the maker of CoCounsel, an AI legal assistant in use at more than 10,000 law firms and corporate legal departments in the US. Thomson Reuters recently launched CoCounsel in Australia.
"It's been interesting watching the legal community navigate this moment of change almost on hyperdrive," Safdie says.
A practising attorney in the US, Safdie identified an opportunity to build a product that would address a resource disparity. She co-founded Casetext in 2014. For many lawyers, especially those with their own solo practice, the ability to produce quality work efficiently is affected by the quality of tools at their disposal. CoCounsel bridges that gap.
"At the time, we didn't call it AI," Safdie says. "It was machine learning, natural language processing and legal informatics. We had been working with OpenAI under NDA [non-disclosure agreement] for some time, but our vision always superseded the state of the science. That changed with GPT-4. When we saw it, we pivoted the entire company in a week."
Through this vastly improved natural language interface, users could ask CoCounsel to perform tasks based on reliable sources of data, such as litigation records and contract databases. The use of reliable sources is now the top point of difference between CoCounsel and ChatGPT. Safdie says consumer-grade tools such as ChatGPT are inappropriate for professional legal use.
"We thought we would have to educate people about how powerful this technology is," she says. "But with ChatGPT they've seen it and now the problem is their understanding of AI and what is appropriate in different contexts and solutions."
CoCounsel uses large language models (LLMs) as an "engine" at its core, but builds a complete product on top, with a precision focus on legal use, with all the security, privacy and confidentiality the industry requires.
"CoCounsel gives you results in a way that lets you easily verify sources and actually go to them," Safdie says. "We have a battery of more than 1,000 tests that run every night to avoid issues like model drift. There's so much that goes into doing this properly."
Safdie points to examples of CoCounsel being used by major US law firms to deliver fast e-discovery feedback to clients on thousands of documents in a single day instead of taking four weeks. A California legal aid service that didn't have adequate staff to keep pace with application reviews cut its appeals backlog in half with CoCounsel.
Now owned by Thomson Reuters, CoCounsel arrived in Australia this year, and law firms have been piloting the product to see how it fits into their workflows. Sarah Jacobson, the director of knowledge management at MinterEllison, says CoCounsel has been highly anticipated.
"I'd seen it before and it was the most sophisticated legal AI tool in the market," she says. "We are taking a leading-edge approach to generative AI and CoCounsel's use of retrieval augmented generation is gold for law firms, because it means we can leverage all the quality documents we have in our own repositories."
At MinterEllison, that means tapping into a database of about 10,000 knowledge documents. Jacobson is particularly impressed by CoCounsel's "true" natural language interface.
She says: "Lawyers are curious, but they often don't have time to spend learning prompt engineering skills. CoCounsel doesn't only answer questions effectively, it also asks clarifying questions to iteratively step you through what you're looking for."
In the pilot of CoCounsel, Jacobson noted the effectiveness of the "review documents" feature, particularly in streamlining the summarisation of witness statements and pinpointing consistencies and inconsistencies. Additionally, the "timeline" skill received positive feedback for its ability to distill complex information into a clear chronological sequence, which has been invaluable for legal professionals handling intricate data.
"What we're really excited about is the forthcoming integrations within the Thomson Reuters ecosystem, with CoCounsel connecting to Practical Law and Westlaw Precision, which will really enhance the platform's capabilities," Jacobson says.
While traditional discovery methods involve searching for specific words or phrases, CoCounsel will search based on an idea. For example, you can ask to find "anything that could suggest misconduct" and CoCounsel could flag anything to match this highly nuanced concept for the human team to review.
Jacobson believes CoCounsel has great potential to improve the speed of resolution in disputes and expand e-discovery capabilities, achieving faster time to first insights so lawyers can get into detailed analysis sooner.
Client reactions to AI cover the full range of positions, Jacobson says, from those who do not want it used on their work to others who are demanding the law firm adopt the technology to deliver faster results and better value. There's an ongoing dialogue between law firms and clients as everyone discovers how these tools can improve how the industry operates.
Safdie says: "We feel 2024 is the year of AI strategy. My advice is to make sure you ask the right questions to adopt the solutions built for the use cases you need. It's certainly the moment to get educated on these tools if you haven't yet."
