Shortly after ChatGPT landed, Matthew Peters, a partner and the national innovation leader at McCarthy Tétrault LLP, paid close attention to the novel technology. He read about the problems that some users said they’d encountered with the generative-AI chatbot—that it writes robotically, fails to safeguard data and makes up facts wholesale. Like everybody in the legal industry, he heard the story of Steven A. Schwartz, a New York lawyer who, in federal court, submitted an AI-generated brief that was full of imaginary caselaw. (In December, a similar incident took place in B.C.) But Peters believes deeply in experimentation. When a disruptive tool hits the market, he reasoned, you can’t just ignore it.
In the summer of 2023, under his supervision, the firm embarked on two pilot projects: one with CoCounsel, a GPT-backed digital legal assistant offered by Thomson Reuters; and another with ChatGPT itself, albeit the premium version, GPT-4, which is more powerful than the free program and can run on a secure Microsoft platform that protects client data. (In his personal life, Peters planned a family trip to Dubai and Egypt with the help of ChatGPT.) For the CoCounsel pilot, 60 lawyers at the firm experimented with the technology and attempted to integrate it into their workflow. For the GPT-4 pilot, a 60-person cohort of lawyers, law students and support staff performed a similar test using the now-famous chatbot.
The results were promising. ChatGPT was decent at drafting simple texts like emails and client pitches. CoCounsel was adept at a range of basic legal tasks, such as reviewing and summarizing contracts or brainstorming questions that lawyers might ask during a discovery. As AI becomes more powerful, says Peters, the impact will be profound: “If a certain job took 10 hours in the old days, by which I mean nine months ago, it will likely take seven or eight hours in the near future.”
Consider the steps involved in preparing a legal memo. You have to research the topic, connect those findings to the matter at hand and write the document. In the next few years, an AI chatbot will likely be competent enough to generate a solid first draft on its own. You would then double-check the law, fine-tune the analysis and edit the prose, a far less labour-intensive undertaking. Peters estimates that generative AI will reduce the workload for many legal tasks by 25 percent.
Which raises an obvious concern. If law firms can achieve the same output with 25 percent fewer hours, will they seek a commensurate reduction in staff? To put it another way, is a massive job loss just around the corner?
Peters and his colleagues at McCarthys aren’t the only lawyers who’ve started to assess the potential of modern AI. Lisa Feldstein, a sole practitioner specializing in family health law, is a regular user of the paid version of ChatGPT. Importantly, she understands the potential limitations of the technology. She never feeds confidential information into the bot. (Instead, she manually adds it to the final product.) She never tells it to produce a court document, such as a factum or affidavit. And as for looking up caselaw? She’s not going to bet her career on that risky gambit. “I’d never rely on it for research,” she says.
What the technology can do, says Feldstein, is help out with relatively minor tasks: emails, student reference letters, blog posts. In some cases, she prompts ChatGPT to generate a draft, which she can correct and refine. Quite often, though, she’ll compose a rough version of the document and ask ChatGPT to play the role of editor. It’s great, she’s found, at managing tone. Informal language in correspondence, like an email, can be rephrased in an instant. “I can write in a lazy, sloppy way,” says Feldstein, “and tell ChatGPT to make the work more professional.”
For now, Feldstein has restricted her use of ChatGPT to fairly low-stakes work. But experts predict that generative AI tailored to lawyers—like the CoCounsel product—will soon be able to handle substantial legal tasks. Michael Litchfield, the director of the business law clinic at the University of Victoria, has been looking into the impending impact of AI on the profession. (In his academic research, he’s investigating how AI might assist self-represented litigants.) The secret to building a truly powerful legal chatbot, he says, is to train it on the right data. At present, that’s an obstacle: the justice system doesn’t consistently capture all court records. “There isn’t some website listing oral decisions delivered from the bench,” says Litchfield. “That data will not make its way into generative AI unless somebody captures it.”
Yet there’s no reason it can’t be done. Litchfield can imagine a future in which legal-tech entrepreneurs build libraries of currently unpublished material and license them to AI companies. If this forecast comes to pass, law firms could tackle casework alongside armies of specialized bots with first-rate research and drafting capabilities.
Perhaps that’s when the AI revolution in law will really get underway. Abdi Aidid is a law professor at the University of Toronto and the co-author of the book The Legal Singularity: How Artificial Intelligence Can Make Law Radically Better. He predicts that even non-lawyers will be able to use generative AI for legal purposes. “When people are mistreated, they know why they’re mad,” he says. “But they have a hard time expressing their anger in language that’s legible to our system.” In a matter of years, generative-AI tools, available for a modest subscription fee, could help people transform a mere grievance into a viable court document.
The nightmare scenario for lawyers isn’t hard to picture. If law firms can rely on specialized bots to conduct sophisticated legal research, why would they continue to hire so many associates? And if members of the public can draft credible claims with AI, why would they retain an actual lawyer? Surely, the job market for legal services is on the precipice of disaster.
Aidid isn’t convinced. In his view, clients will value the wisdom of human lawyers long into the era of generative AI. He also disputes the often-posited theory that ChatGPT-like tools will turn lawyering into a less-lucrative career path. It’s true that lawyers, armed with productivity-boosting AI, will soon complete many tasks at a speedier clip. And in an industry dominated by hourly billing, it won’t be possible to collect as much revenue on a per-task basis. But if legal work becomes faster and cheaper, the profession will benefit from at least two new business opportunities. First, lawyers will be able to offer existing clients the option to spend the same amount of money in exchange for more legal advice—an attractive deal. Second, lawyers will have a chance to market their talents to a currently untapped pool of customers: people who, today, are entirely priced out of the legal marketplace. That’s a large slice of the population. In Canada, 71 percent of people with a legal problem never contact a lawyer. “Think about the amazing misallocation of resources in the legal industry,” says Aidid. “Most people who need a lawyer can’t afford one.”
At McCarthys, Peters foresees no job losses at all, even though his firm has begun rolling out ChatGPT-4, in phases, to all of its lawyers. (In time, CoCounsel will also be in use firm-wide.) His assessment is rooted in the same logic that underlies Aidid’s analysis. Generative AI will cut into the number of billable hours that lawyers spend on tasks like caselaw research, contract review and due diligence. But it will allow the firm to both increase its level of service to current clients and land new ones, thereby maintaining—or perhaps adding to—its total revenue. Best of all, says Peters, with AI handling the more mundane aspects of casework, associates will devote a greater portion of their time to advanced legal strategy. “We’re not reducing the number of students we’re hiring, and we still have as many associates as ever,” he says. “Quite frankly, I think their experience is going to be better in three years than it is now.”
So is the future rosy? There is one final issue to consider. Romesh Hettiarachchi—founder of Signal Lawyers, a solo shop specializing in real estate, commercial litigation and business law—certainly appreciates the upside of AI. He’s trained a unique version of ChatGPT-4 on a limited corpus of texts that are germane to his practice, making the bot far less likely to go rogue and hallucinate facts. He uses it on straightforward legal jobs, from drafting claims based on a template to sourcing basic legal rules. But he’s afraid that a major downside of generative AI lies ahead.
Today, Hettiarachchi explains, the vast majority of legal grievances don’t make their way into the justice system. And yet, our courts are still chronically backlogged. What if, thanks to AI, the number of legal claims doubles or triples? He shudders to think about this possibility. “Our justice system is already under-resourced,” says Hettiarachchi, who also hosts a podcast about his journey as a sole practitioner. “Unless there’s an investment in the system, things will collapse in on themselves. Change has to happen at the governmental, regulatory and professional level. As a legal community, we need to make decisions that can meaningfully influence the trajectory in a positive direction.”
Here’s the good news: if you’re a lawyer worried about losing your job to robots, you can probably relax. But a tsunami of claims could be coming. Can the justice system find a way to handle it? That’s another question entirely.