Artificial Intelligence & Robotics
How lawyers can take advantage of ChatGPT and other large language models disrupting the legal industry
ChatGPT and other conversational artificial intelligence built with large language models could radically change how the legal industry operates, giving early adopters a major advantage.
For instance, conversational AI can automate routine tasks such as document review and contract analysis for lawyers, and help them perform legal research and writing much more efficiently. Law firms also can use it to create blog posts and social media content for marketing purposes.
“If you are a lawyer who is not using this stuff, your opponents are. They are going to do better work than you,” says Noah Waisberg, an entrepreneur, former corporate lawyer and co-author of the book AI for Lawyers. “For most people, it’s going to give them an edge, and I don’t know why you wouldn’t take an edge.”
OpenAI’s Generative Pre-Transformer models, called GPT, are the most well known of the large language models, which can understand, process and respond to human language. Google also recently launched an experimental preview version of its conversational AI tool, Bard. In addition, Microsoft’s BingAI search engine has OpenAI’s GPT technology embedded in it. In February, Meta introduced LLaMA, which stands for Large Language Model Meta AI. It is available to academics, policymakers and others who apply for a noncommercial license.
Daniel Martin Katz, a law professor at Illinois Institute of Technology’s Chicago-Kent School of Law, says even more large language models are in the works. He says we’re in the early stages of this technology. Overall, the AI hardware and services market is expected to grow to $90 billion by 2025, up from $36 billion in 2020, according to UBS.
“There’s been a material increase in the capabilities of these tools, of these large language models, particularly with GPT but just in general, and that does bear on the type of work that lawyers do,” Katz says. “This is important for lawyers because we have technology that’s finally pretty good at language, and that has always been a challenge.”
A LexisNexis survey from March revealed 57% of consumers are aware of generative AI tools such as ChatGPT. But awareness was significantly higher among lawyers at 86%, and half had already used it in their work or were planning on doing so.
According to the survey, “84% believe generative AI tools will increase the efficiency of lawyers, paralegals or law clerks,” In addition, “It’s not just the practice of law; 61% of lawyers and 44% of law students also believe generative AI will change law schools and the way law is taught and studied.”
Conversational AI is a tool that will create new opportunities and free up lawyers from tedious tasks, says Waisberg, who founded Kira Systems, a legal tech startup that used AI to help lawyers conduct contract review, in 2011 in Toronto. He sold the company in 2021 to Chicago-based Litera. He then became CEO of Zuza, a contract-analysis tool.
“There are some problems we’ve been working on for a long time that you can see it was solving very quickly,’ Waisberg says. “But pulling data from contracts, it wasn’t right on some stuff.”
ChatGPT is good at solving some very difficult problems, Waisberg says. But the technology still has trouble pulling data from contracts accurately and still needs someone to double-check the information.
GPT-4 isn’t ready as a stand-alone approach for contract analysis if predictable accuracy matters, he says. “I don’t have any idea when it’s going to be right or wrong, and that’s kind of troubling,” Waisberg adds.
Not a replacement
With the increasing capabilities of large language models and neural network technology, there is increasing concern about the future of lawyers. But AI chatbots like ChatGPT will not cause massive layoffs of lawyers, says Daren Orzechowski, a partner at Allen & Overy. He is co-head of the firm’s technology practice and is based in Silicon Valley.
“This is an efficiency tool, and I don’t think it’s replacing people,” Orzechowski says.
Orzechowski adds that the technology might even open up new areas for lawyers and specialties. For example, he notes that technological advances like desktop PCs changed how people worked but didn’t significantly reduce the number of lawyers in the industry. Conversational AI tools, however, will change how lawyers practice law.
In February, Allen & Overy integrated Harvey, the AI platform built on a version of OpenAI’s GPT technology focused on legal work, into its global practice. Harvey will enable more than 3,500 of the firm’s lawyers across 43 offices operating in multiple languages to access and generate “legal content with unmatched efficiency, quality and intelligence,” according to a press release. During its Harvey trial run, firm lawyers asked around 40,000 queries regarding their day-to-day client work on contract analysis, due diligence and regulatory compliance.
Harvey.AI isn’t the only startup using AI and large language models in the legal space. Others include Logikcull, LawGeex and DISCO.
In early March, Logikcull launched a suite of AI features and ChatGPT integrations that aim to reduce the time spent on discovery by up to 90 percent, says Andy Wilson, the company’s CEO and founder.
Logikcull, with nearly $40 million in venture capital funding, created a command line tool for searching documents using ChatGPT. It lets users command all kinds of actions on a paper like “summarize this document” or “translate into English,” Wilson says.
“It’s just a massive breakthrough enhancement for legal productivity,” Wilson states.
Logikcull’s customers spend over 65,000 hours monthly using its software to search through data they upload, like Slack messages, emails and audio files. Then they’re reading it to find relevance, Wilson adds. What conversational AI in these large language models are enabling them to do now is to do that reading for them at scale.
“The bigger issue the legal system is going to face is their business model is based on inefficiency and billable hours,” Wilson says. “For people using fax machines and printing their emails out, this will be mind-blowing.”
Technology is getting smarter every day. The more people interact with the technology, the smarter it gets in a process called machine learning. For example, GPT-4 in March passed the bar exam with a score of 75, higher than the 68% average and placing it in the 90th percentile, Katz says. In a previous test, GPT-3 scored 50 and only passed two multiple choice portions of the exam, placing it in the 10th percentile.
And in just four years, GPT has progressed from 0% on the Multistate Bar Exam for GPT-2 to its score of 75, Katz adds.
“Large language models can meet the standard applied to human lawyers in nearly all jurisdictions in the United States by tackling complex tasks requiring deep legal knowledge, reading comprehension and writing ability,” Katz wrote in his paper published on March 15.
With conversational AI, lawyers also must consider the ethical obligations of attorney-client privilege. ChatGPT holds onto all the queries it receives, says Renée McDonald Hutchins, dean of the University of Maryland Francis King Carey School of Law. Ultimately, she posits, law firms will probably build their in-house systems with proprietary data centers.
The large language models powered by AI are only as good as the data fed them, Hutchins adds. Some of that data can contain embedded errors, she says. And if the data set the AI is trained on is biased and has embedded privileges in them, the AI will recreate those results.
“Sometimes we don’t want the law to replicate existing hierarchies,” Hutchins says. “If we rely on it too heavily, we risk replicating oppression in a way we don’t want to.”
For example, suppose a judge uses conversational AI and large language models as tools in the criminal system for sentence recommendations relying on arrest records. In that case, they might replicate a racial bias, Hutchins says, because police arrest brown and Black people at a higher rate than whites.
“[They] can be very powerful tools,” Hutchins says. “We have to be very mindful of [their] limitations.”
Another potential problem is that ChatGPT and other conversational AI and large language models might face regulatory barriers if a nonlawyer is using ChatGPT to provide legal advice and assistance to laypeople; that might be considered unauthorized practice of law, Hutchins warns.
“We need to make sure we are graduating students who are familiar with and skilled in technological advancement, but we also need to be mindful of human experts’ need to guide clients through the legal system,” Hutchins says.
Laura Lorek is the publisher of Silicon Hills News, a regional technology publication based in Austin, Texas. She is also the host of the Ideas to Invoices podcast. Previously, she was a senior writer at the San Antonio Express-News, a senior writer at Interactive Week Magazine, a specialty writer and a technology columnist at the Sun Sentinel.