Will law firms fully embrace generative AI? The jury is out | The AI Beat

As the daughter of a lawyer, I’ve been fascinated by the steady march of generative AI into the legal industry. Memories of my Dad digging through ever-growing piles of paper, rummaging through books in the law library, beholden to the billable hour — will those all become relics as law firms fully embrace gen AI tools? 

The jury is still out on that. On one hand, generative AI is shaking up the legal industry, with companies like Everlaw adding generative AI options to their product portfolio, while Thomson Reuters integrates with Microsoft 365 Copilot to power legal content generation directly in Word. 

On the other hand, lawyers tend to be a conservative bunch — and in this case, attorneys would likely be wise to be cautious, with headlines like “New York lawyers sanctioned for using fake ChatGPT cases in legal brief” going viral. Another problem is that their clients may not feel comfortable with law firms using generative AI — a new survey found that a third of consumer respondents said they’re against any use of gen AI in legal.

>>Follow VentureBeat’s ongoing generative AI coverage<<

But while many law firms may be old school, the legal landscape is also ripe for generative AI disruption, because the legal system generates a huge and ever-increasing amount of data to sift through, while there are a boatload of repetitive tasks legal professionals have to deal with — such as in document production — that generative AI could help tackle. 

“The law lives and dies on text…and the haystacks [of data] people deal with are huge,” AJ Shankar, who founded Everlaw in 2011 and has a UC Berkeley computer science PhD, told me in an interview a few weeks ago. “It is just so pervasive, and it is the strength of these large language models — it is what they do particularly well.” 

Legal e-discovery is a rich field for generative AI 

Shankar said the legal process of e-discovery — which involves taking a huge corpus of potential digital evidence in a case, from emails to docs, and distilling it down into the tiny percentage of smoking-gun data that really matters — has been a rich field for AI over the past several years. 

But with Everlaw’s new generative AI, which is available in a beta program, lawyers can go beyond just clustering data at the aggregate level to querying, summarizing and otherwise extracting details from documents to get what they need. For example, the company says that while it typically takes hours for a legal professional to compose a statement of facts, it can now happen in about 10 seconds, delivering legal teams a rough draft to edit and fact check. 

“It’s yet another very deep way to interact with individual documents, and then there’s also generating work products and synthesizing evidence across multiple documents,” he said of Everlaw’s generative AI tools, which use OpenAI’s APIs for GPT-3.5 and GPT-4.  “That’s something that it does that our prior tools did not do at all.” 

Still, Shankar admits that “the jury’s out” on whether and how lawyers will actually use the tools in their day-to-day workflows. “I would say there’s a lot more enthusiasm as opposed to skepticism when we just show them what the thing does,” he said. “This is kind of the litmus test.” 

A legal copilot 

For Thomson Reuters, which includes software platforms like Westlaw, Practical Law and Document Intelligence, and recently pledged to invest $100 million annually in AI, getting lawyers on board with generative AI is all about trust. 

“There’s generally a question around trust when generative AI tools start trying to answer questions about legal information or finance — they often make stuff up which isn’t acceptable,” said Kriti Sharma, chief product officer, legal tech, at Thomson Reuters. She told me that the company’s recently-announced partnership with Microsoft 365 Copilot — a plugin that is currently in a closed review beta process but will be fully rolled out by the end of the year — is about “how do we inject trusted legal information into Copilot?” 

For example, if lawyers are producing drafts in Microsoft 365, they could ask the Thomson Reuters plugin to validate the legal data, offer a variation of a clause, summarize a precedent, or tell them which are the right points to negotiate. 

“Think of it as legal intelligence coming into Copilot, where you work,” she said, adding that Thomson Reuters has “a number of experiments and tests ongoing right now, because the stakes are high — we’re working through getting a pretty high bar of accuracy before we can say this is now ready for general availability.” 

The tool, she claimed, is “fundamentally going to change how lawyers work. And I think they can’t wait.” 

But while there is excitement, there is nervousness as well, she added. “The nervousness comes from challenges around trust and accuracy, like how do you know this is correct? If you have to check every single thing the machine has done, that’s problematic.” 

That is the problem Thomson Reuters, Everlaw, and any other generative AI tool in the legal space has to solve. 

“If we can bring in trusted information which is grounded in legal facts, and we can turn that into a high-quality work product that saves people time, that’s when you’ll start to see a big shift,” said Sharma. “Right now it feels like a lot of curiosity and excitement. But now, we need to turn it into real work.”