U.S. Legal Support recently hosted a webinar on the power of generative AI in the legal profession. We partnered with Merlin Search Technologies, a pioneer in the use of AI and large language models (LLMs) to optimize legal processes for attorneys’ and clients’ benefit.
Our Senior Vice President of Sales Operations, Brian Pauley, was joined by two experts:
- John Tredennick, CEO and Founder of Merlin Search Technologies
- Dr. William Webber, Chief Data Scientist at Merlin Search Technologies
Together, they talked about the ways gen AI benefits all stakeholders in the legal profession.
Demonstrating the Power of Generative AI for the Legal Profession
Treddenick began the presentation talking about an exercise he conducted to test gen AI’s capacity in the legal profession. He explained a memo assignment that he gave to “Jennifer AI,” a composite associate. The task involved a hypothetical case in which a large HOA in Florida was fighting back against expropriation of its land for commercial use (i.e., a shopping center).
The request went out at 4:00 PM on a Tuesday, with no response expected for at least a week.
However, within 15 minutes, “Jen AI” produced an insightful response detailing important considerations for Treddenick and his partners. They sent a follow-up request asking whom to depose and for what reasons, which was also met with detailed responses within 15 minutes.
The upshot is that any associate trained in using this kind of technology can produce sophisticated legal insights that would normally take at least a week within mere minutes.
Understanding the Underpinnings: What is Generative AI?
Tredennick then turned to Dr. Webber for an explanation of what gen AI is and how the underlying technology works. Webber explained that gen AI is a relatively new technology based on neural networks that attempt to imitate how the human brain learns things.
The way this works is that a model will be trained on a huge amount of data, such as texts, images, videos, and more so that it can internalize the underlying logic between them. The immediate goal, being able to guess a word or otherwise (re)produce information, leads to a byproduct of the model picking up a ton of general and specific knowledge along the way.
This led to a back and forth between Tredennick and Webber about a potential weakness in LLMs: namely, they can’t remember or pass on information, and their knowledge is limited to the specific documents and other information they were trained upon and nothing else.
That is, except for any information users put into a model’s context window.
Digging Deeper into Language Processing and Context Windows
To demonstrate how useful gen AI is in a legal setting, Webber provided an example of how context windows are used to feed new information to a model and have it produce insights.
In the example, GPT is asked a very simple question: “What is this about?”
However, the model is also provided with a text to study, which in this case was publicly available correspondence with Governor Bush regarding the HOA and land expropriation questions from earlier. By providing this, GPT can read and understand the problem at hand.
Treddenick pointed out that a major advancement in gen AI has been the expansion of this whiteboard space, which was much more limited in earlier versions of GPT and other models.
Security Considerations and Preventing Gen AI “Hallucinations”
Another major consideration around gen AI and its application to law is how secure it is and whether clients’ confidential information can or should be exposed to LLMs. Treddenick explained how companies such as Microsoft protect users with commercial licenses guaranteeing information will not be shared or even hosted beyond mere milliseconds.
Licensed, legitimate uses of these AI products thus confer a reasonable expectation of privacy.
However, another concern when using LLMs for any purpose is accuracy and fidelity or avoiding “Hallucinations.” Webber explained that these are instances in which an LLM produces content that is plausible but doesn’t exist in actuality—like especially dangerous falsified legal citations.
Fine-tuning and careful use of context windows are the best defenses against this issue.
How Gen AI Powers eDiscovery and Deposition Summaries
Treddenick and Webber went on to provide a demonstration of how Merlin Search Technologies uses gen AI to power their eDiscovery solution. Key takeaways from their presentation include the best practice of treating gen AI in discovery and research as a holistic tool rather than a simplistic search engine.
In practice, Merlin’s approach to using gen AI for eDiscovery comprises:
- Summarizing and analyzing documents
- Creating accurate and practical timelines
- Synthesizing and reporting on information
- Answering questions across all documents
- Reviewing documents for accuracy and fidelity
One thing that sets Merlin’s approach apart is their use of summaries to allow the LLM to function more efficiently and effectively, maximizing AI’s benefits for all stakeholders.
Sneak Peek: U.S. Legal Support’s DepoSummary Pro™ in Action
Another context in which summaries are crucial to efficient legal operations is in deposition summarization and processing. The last part of the presentation before the Q&A was a sneak peek at U.S. Legal Support’s own AI-powered solution, in collaboration with Merlin: DepoSummary Pro™.
Pauley explained how our tool is able to process thousands of pages of information and create a concise depo summary by leveraging the same technologies Treddenick and Webber covered.
What sets DepoSummary Pro™ apart is a combination of features including but not limited to:
- Multiple layers of summarization
- High-level organization in abstract form
- Accurate and explicit statements and references
- Full editability upon delivery, for further customization
Treddenick showed viewers exactly how DepoSummary Pro™ works in practice, highlighting how it empowers attorneys to spend less time on manual review and more on strategic case activities. Ultimately, AI in deposition speeds up workflows and maximizes efficient access to justice.
And the good news – this deposition summarization service is now available! Learn more about DepoSummary Pro here.
Leverage the Power of AI to Optimize Your Legal Practice
The speakers ended the webinar by answering viewers’ questions about the efficient and secure use of gen AI tools and best practices for minimizing hallucinations, among other issues.
We’re proud to have hosted this event with our partners and friends and Merlin Search Technologies. We’re also thankful to all participants who watched and posed questions to Treddenick, Webber, and Pauley. If you’d like to learn more about gen AI and its use in the legal profession, or if you’d like to see U.S. Legal Support’s AI in action, watch the webinar here.