The Legal Data Intelligence Podcast with David Cowen (Episode 32)

In this episode of Careers and the Business of Law: The Legal Data Intelligence Series, host David Cowen speaks with LDI Architects Melina Efstathiou and Matthew Hamilton. They discuss "From Assessment to Intelligence: Operationalizing Early Data Strategy (EDS) in the Age of Generative AI," a new guide published by the LDI Disputes & Investigations group. They talk about how the combination of EDS and generative AI could enable general counsel to ask plain-language questions of their own data and make faster, better-informed decisions on cost, counsel, settlements, and more.

This guide was jointly created by Architects Kevin Clark, Melina Efstathiou, Ross Gotler, Matthew Hamilton, Tristan Jenkinson, Glenn Melcher, Daniel Miller, and Linda Sheehan.

Listen to the full episode and read a partial transcript below.

David Cowen: What can the tech do now that it couldn’t do earlier? What can it do now when you point it toward unstructured data?

Matthew Hamilton: The advent of GenAI has changed the game in terms of how we can query the data. Before, we had the data, but we would need to hire an engineer, a programmer, or someone to write scripts to make it usable and try to find some patterns, some meaning in it. One of the main strengths of the large language models is that they can take this big mountain and find patterns, and they can make logical leaps, or illusory logical leaps.

So we don’t have to describe a concept with just one search term. We can give it some breadth. They can go out and find meaning in things where, before, you would have had to use very precise query language. It could be a real challenge, in a sort of undifferentiated dataset where things are in 17 different formats, to come up with consistent, reliable answers. The generative AI tools tend to make that last part just frictionless.

Melina Efstathiou: Yeah. What I would add to that is that the whole natural language interaction with the models has been a game changer. The reason I say that is that Matt and I know, for lack of better terms and being fully transparent with the audience, how painful it can be to lead that mindset shift.

You have lawyers and attorneys from all sorts of generations, and not everybody wants to just jump on the tech train. It was really, really difficult to articulate even search syntaxes back in the day and what they would deliver, and, more to the point, why they did not deliver what people expected. So the introduction of generative AI, with a human-language back and forth of prompting and the ability to understand just plain Jane questions, as you progress and interrogate the results, has been like an absolute tsunami in legal technology.

But there’s also the provision of rationale. Like Matt knows, the way we used to interrogate data was very, very linear. It was forensic. It was Boolean search-led, and it was led predominantly, if not exclusively, by the information we had in the metadata and, of course, the extracted text.

Now we ask it one question. It delivers particular results, and it also tells us why, what the rationale is behind them. If we disagree, and it is very healthy to disagree with the machine, at least through the first few rounds of prompting, that back and forth has been completely enlightening, even for subject matter experts like Matt and me. We get to see live the interaction between the data and the model, and how a change in our language delivers a different response, a different realm of results.