There are two ways to look at the data maelstrom churning within modern enterprises everywhere: it’s either a profound liability or a profound opportunity.
Richard Finkelman, managing director at Berkeley Research Group (BRG) and a 2024 AI Visionary, has both feet firmly in the latter camp.
“Those of us who understand e-discovery understand the importance of clean data,” Richard told us after being named an AI Visionary earlier this year. And in a generative AI world, he said, “the people who know how to manage data will also become the people who know how to do prompt engineering. This should create great opportunities for our whole industry.”
A broad goal to derive deeper legal insights (and, for that matter, larger business lessons) from data caches of all kinds is already beginning to define the next era of e-discovery and other document review endeavors. Indeed, opportunities abound on the AI horizon in the legal field—ready to improve everything from contract analysis and litigation support to the practice of law and more (including, as it turns out, little things like wedding vows).
Prompt Engineering as a Specialty—and a Solution
“Prompt engineering is, in this golden age of our industry, going to look something like a litigation support person” does now, Richard told us.
As lit support professionals help enable stronger collaboration between attorneys and technologists, prompt engineers—perhaps those same lit support professionals, should they choose to specialize in this emerging field—will provide an essential connector between case strategists and the generative AI tools that are already helping to move document reviews forward.
“One of the biggest problems with generative AI is hallucinations,” Richard explained. Many generative AI systems, when asked for something they don’t have enough training to give—such as a 1,000-word bio on a person about whom they only know some 500 words of background information—may simply invent the rest to meet the prompt, often without any indication that half of the result is fabricated.
“If you keep asking questions, it’ll keep answering without telling you: ‘look, I’ll keep answering but I don’t have anything to go on here,’” Richard explained. “Prompt engineering is how you solve that problem. A data scientist or a litigation support person who’s technically qualified can spot that, go back, and reengineer it with new prompts until they get something satisfactory.”
With the right data science training, Richard said, people can approach this problem armed with knowledge on how to meet technology where it is—ready to be mindful of its limitations, and well- prepared to overcome them in the interest of facilitating the best possible results for a legal team.
Putting Data Science into Practice in the Legal Realm
Richard noted that “There aren’t many data scientists in this industry”—yet.
“You can’t take a class and expect to really learn how to use Relativity in document production until you’ve actually done it,” he said. “You learn by doing.”
Likewise, legal professionals who’ve been in this space for a long time won’t pick up on data science without experience—and data scientists won’t pick up on the nuances of legal work without experience, either.
“There’s learning that needs to be done on both sides,” he said. And, often, an intermediary familiar with both specialties needs to serve as an intermediary between lawyers and data scientists, ensuring everyone has the information they need to collaborate effectively. These intermediaries already exist within the ranks of e-discovery professionals.
Richard sees two key barriers to the broader adoption of AI in the legal world: education and the accessibility of training data.
“Education seems obvious, and it was the number-one identified issue to adopting AIML technologies in the 2023 AIML report from BRG, Relativity, and ACEDS,” he said. “Courts need education and attorneys need education—and the education has to include their own use of AI.”
Mutual pursuit of education between data science experts and legal practitioners can help close these gaps and make everyone more involved with this particular application of artificial intelligence in the real world.
And training data? That will take some mutual pursuit, too.
“A lack of reusable training data is something that will hamper the pace of AI innovation. When people train AI models with litigation or compliance data, they are not interested in donating any part of the technology to future AI models,” Richard observed. “In fact, usually it is precluded by protective orders and settlement agreements. This means destroying models and rebuilding them for the next case. Other industries, like the medical field, are able to make larger advances because of the reusability of data.”
Getting creative about training models, and deploying them for bespoke projects as well as at a broader scale, is key in this space. And it requires a lot of collaboration between experts, which also means it requires a lot of strong advocacy from those in the know.
“I feel that those of us involved in generative AI have a responsibility to explain how to use the technology as safely and effectively as possible,” Richard said.
Moreover, he told us, “this is the best industry to do AI from.” The data sets—some of the cleanest that can be found—are incredible training tools, and the lessons learned in this space can improve AI architecting for legal applications and beyond.
Improving Careers, Relationships, and Communication between Firms and Clients
When all is said and done, Richard believes that AI and its many applications are setting the stage for all kinds of big wins in the legal industry.
This includes, as we’ve mentioned thus far, the deliverables and insights gathered from e-discovery and other document review projects. But it doesn’t stop there. The impacts on careers will be huge.
“Professional services industries are likely to benefit the most from the explosion of AI technology—and that means they also are likely to see more radical change in how work gets done,” he told us. “Industries like legal will see people who master the skills of the ‘new’ way to do things benefit exponentially while those who do not will perish.”
Jumping in now, as early adopters, will give team and individuals a head start on strategizing for the future state of legal.
“As someone who was already building and using AI technology before generative AI, I see benefits from being an early adopter of generative AI technology” as an individual, Richard said. “I pay for ChatGPT 4 and Claude Pro and I use them regularly to help me with customers or to educate them on how the technology works. I receive a lot of positive feedback from clients when I share my experiences.”
And more practice with AI begets more creativity in how it can be applied to address new and stubborn challenges.
“What excites me even more is the ability organizations have to create their own AI. Original AI applications can be created that help organizations solve unique problems. Professional services organizations in particular stand to benefit developing AI for specific client problems,” he continued.
For example, he said: “At BRG, we are building private language models for specific client use cases.”
This doesn’t have to mean crafting new platforms from scratch; custom, add-on applications of AI can be just as powerful. Actually, according to Richard, the enhancement of existing, reliable platforms with imaginative uses for AI is one of the areas he’s most excited to watch.
“Microsoft’s Copilot is a good example of this; OpenAI’s integration with the Microsoft Office suite will undoubtfully unleash productive booms,” he said. “So will augmenting applications like Relativity with generative AI capabilities.”
To take advantage of these benefits, Richard said it’s essential to just jump in—much sooner than later—and give AI a try.
“I would encourage all organizations to adopt some AI, even if it is just within a working group or a series of pilot projects. Pick an area where augmented AI can help a group, like Copilot from Microsoft,” he advised. “The problem with not doing anything is that augmented AI technology will become pervasive much faster than people realize, and that means losing competitive advantage for not adopting some AI.”
Recommended for you
