Artificial intelligence as a concept is exciting science fiction. It conjures up images of robot lawyers and know-it-all computers that easily solve the industry’s most complex problems—doing anything that a person could, but faster and better.
However, firms that have taken the plunge and are actually using and experimenting with legal AI technology tell a much different story. Their reality is not science fiction. It’s much more prosaic and programmatic. And that’s a great thing for law firms, because while space-age hype makes headlines, practical benefits encourage adoption and make AI more accessible.
Benefits coming into focus
Fresh off launching AI Hub and a document classification engine, plus forming new partnerships with AI pioneers Kira and Leverton, HighQ has initiated a research study aimed at understanding how legal AI is being adopted, evaluated and deployed. While data still is being collected through an industrywide survey and personal interviews, a clearer, far more functional picture of AI is emerging.
Andy Neill, manager of the AI product line at HighQ, sums up the early impressions: “What we’re finding is that almost 80 percent of the value is in that first 20 percent of the effort—a huge amount of insight from a small amount of intelligence.
“Clients who are actively using AI have given use case examples like, ‘I’ve just been given thousands of files to review, and I have to tell my client next Monday how big this job is going to be and which ones they need to be concerned with.’”
To accomplish this without AI, a firm would need to pull together a team to look through the thousands of documents manually (maybe tens or hundreds of pages each) and mark whether they require attention—likely a group of 10 or so people working for a solid 72 hours.
“Without an awful lot of science-fiction-type levels of intelligence, without anything fancy around natural language interfaces where you’re able to talk to the computer and it talks back, firms can get a tremendous amount of value from the type of AI that when it’s presented with a bunch of files, it creates a filter so you only work on the ones that matter,” says Neill.
Firms putting AI to the test
Eversheds Sutherland Ireland is one such firm that has taken this practical approach to AI—streamlining a tedious, time-consuming process with some smart but not overly complicated logic. And the early returns are very positive. The firm was able to complete a change of control contract review of 1,400 contracts in less than 48 hours using HighQ OCR and AI functionality.
Similarly, Addleshaw Goddard is using a mix of HighQ and Kira AI capabilities to analyze documents to identify potential issues, consolidating the results efficiently into one place in HighQ’s AI Hub. As a consequence, a review of 100 property management agreements took AG just 20 minutes per contract instead of an hour. In another exercise, staff reviewed 100 leases in half the time of a traditional, manual review.
“That’s where we are receiving the most feedback,” says Neill. “People are expressing that in a real, practical sense, AI is useful to them and valuable to their clients, because they are getting the lawyer’s advice in a meaningful turnaround time at a meaningful cost.”
Because due diligence and contract review can require a lot of people and resources to turn around requests in a short amount of time, an AI tool that assists in the review process and performs even simple document triage—so firms can focus their human intelligence on more valuable work—is where the value is right now.
Subtlety is a necessary feature
Another takeaway from the document-identification-and-review use case is the need for an AI solution to be focused and flexible. Each firm has its own way of doing things, and their documents have distinct differences. So while it may seem obvious, firms must be able to easily train an AI solution to teach itself how to recognize different things based on their unique requirements.
“From a vendor perspective, knowing everything a client is going to want to put into the AI system is almost impossible,” says Neill. “So while we’ve trained our AI engine to recognize an employment contract, for example, we’ve also given our clients the ability to either improve that intelligence or discard it and train their own employment contract classifier.”
“We’re basically empowering our clients to be data scientists,” adds Neill. “They can find internal examples of an employment contract, put them into a folder, and tell the AI engine, ‘This folder contains employment contracts. Go train yourself.’”
In fact, it’s even more subtle than that. An organization’s employment contract includes mostly standard content, but differs on salary amount, the date it was signed, and the name of the person. Otherwise, it’s a pro forma agreement that’s used time and again. But sometimes, there are different paragraphs or subsections added in. There also might be new language that’s been negotiated—potentially only negotiated with executives.
“So you could quickly surface how many employees are on a standard contract and how many are on special contracts that a client needs to be concerned with,” explains Neill. “For due diligence, it helps to be able to sort them into piles, but with AI, they’re able to sort them quite subtly and know, ‘These are normal. We can do minimal due diligence on those. These are special, we need to do a higher level of due diligence on those.’”
The ability to deploy and train AI to recognize subtleties in documents and classify them without enlisting a bunch of coders, developers or other expensive IT resources ensures that firms can get results from AI quickly and cost-effectively. Again, it’s not the stuff of science fiction, but it delivers a measurable benefit that’s sorely needed in the legal industry—the more efficient delivery of legal services.
While it’s natural to get excited about technologies that show promise to transform the status quo and push the boundaries of what’s possible, it’s important that firms invest in the reality of AI and not a marketing-driven fantasy, at least for now.
