Speaker 1: Hello and welcome everyone. I'm Hannah Smith and today we're here to discuss AI, more in particular how AI is changing the legal landscape. And we'll also hear more about CoCounsel, a first-of-its-kind AI legal assistant tool which is powered by GPT-4 and which will be utilized by D. L. A. Piper in the future. I'm delighted to be joined by my colleague Bennett Borden, Chief Data Scientist and AI partner in our Washington D.C. office. Absolutely great to have you over with us today, Bennett. So we're here to discuss AI, more in particular how it's changing the legal landscape. Is AI coming for us lawyers? Is it here to eat our lunch, right? And before we dive in, I just want to refer to an article I've recently read in the New York Times, actually an article you were quoted in. And there it says, so it says that a study of researchers at Princeton University, University of Pennsylvania and New York University concluded that the industry most exposed to the new AI was legal services. And then another research report by economists at Goldman Sachs estimated that 44% of legal work could be automated in the future. So Bennett, this begs the question, are lawyers

Speaker 2: an endangered species? Only in some ways. What's really interesting about this technology is how it empowers how we practice law. If you think about every practice area that a lawyer can work in, we fundamentally are dealers in information. We gather it, we analyze it, we add our legal acumen to it, and then we deliver it. And so any kind of tools that make any of those steps easier is going to have an impact on how we practice law. What's interesting to us is that it's not that AI is going to replace lawyers. It's that lawyers who use AI are going to replace lawyers who don't. So what we need to think about is how do we take advantage of this technology without it disrupting how we do business?

Speaker 1: Exactly, exactly. Because you can think that obviously it will make us lawyers more productive, but there are some things that a robot can't do, which is like building client relationships, right? Exercising judgment when it comes to really complex matters. Are there any other kind of things that you say, well, a robot will never be able to replace that?

Speaker 2: Yes, it really is the human relationship and the strategy. The interesting thing about this technology is that it is embedded in what we've done in the past. So all the data that these algorithms and other AI systems are built upon are what happened before. And so the risk is that the answers that these things spit out are based upon what we've done before, but what we've done before isn't always the best thing. Lawyers, very often they want to find the answer, but sometimes the answer needs to change, or we need to progress in the law, or we need to be able to push against these boundaries that we've had in the past. And these systems aren't going to be able to do that. There are certain things that it's tremendous for. So they're very good at describing things that have happened. So think like an internal investigation or a litigation. We are pouring through massive quantities of data to try to figure out what happened and why. They aren't so good at predicting what's going to happen, which is something that lawyers are hired for all the time, right? If my company does this, how is the market or the regulators going to react? So there is a requirement now for us to figure out what is the essence of being a lawyer and how we even charge for that work. Historically, we charge by the hour, but I'm not so sure that in every case, the time that we spend on something equates to the value of that legal work or legal solution. And that's part of what we're trying to figure out now is how do we capture the essence of what it means to be a lawyer, now enabled by technology, and what is the proper price for that?

Speaker 1: Exactly, exactly. So actually on that, so the new AI is clearly challenging the status quo of traditional law firms. So do you actually think that business models might change going forward?

Speaker 2: Absolutely. They simply must.

Speaker 1: In the short term, long term, what are your views?

Speaker 2: I think because DLA Piper is so forward-thinking on this, that we are going to be driving that change in model. And as we should. The problem with the legal profession generally has been it has been resistance to technology that gains efficiency because if we build by the hour and we become more efficient, we're just handing that efficiency off to the client or the market. And that doesn't inspire anybody. So what we're trying to do and what we must become successful in doing is how do I capture some of that efficiency gain for ourselves? Basic Adam Smith theory of markets, right? And so what we're looking at now is particular kinds of legal products and services that we can easily transpose into some other form of billing than by the hour. So whether it's fixed fees or per unit or per task, and that's where we're putting most of our attention.

Speaker 1: Exactly. It's all about keeping the value for a client, right? Going forward. Totally makes sense. So now going over to the CoCounsel tool, which is an AI legal assistant tool that DLA Piper will be using in the future. You are part of the team testing this tool. So you're the perfect person to ask Bennett, what is CoCounsel and how does it actually work?

Speaker 2: So CoCounsel is a product by a company called Case Text, and it is a large language model, like ChatGPT or BARD or some of the other systems we're seeing come out. It's actually based on GPT-4, which is an open AI product, the same that powers ChatGPT. But it is trained on the world, but then especially onto legal content. So think all the federal and state cases in the United States, some of the secondary sources, things that we would find in online research tools. So what that has done is taken the elementary school kid of GPT and then sent it to secondary school in the law, right? So it is better at legal things than the typical GPT models. What it is really, really good at is identifying and summarizing information. And so there's a lot of applications that it does really, really well. It also has the limitations that these models do. The way that GPT models work is they're just predicting the next most likely word in the prose that they are spitting out in its answer. It does it so very cleverly and in a way that makes us think that it's intelligent because it's so good at that. The downside is it actually doesn't understand the meaning of the words that it's spitting out. And that's why you get these hallucinations, which is a very kind way of saying it just flat out lies to you. And that's a problem for lawyers. If you ask this thing a question like, what are the events that triggered the duty to preserve documents in the city of New York or the state of New York, if that answer is wrong, that's a problem. And so when trying to put guidelines in use and especially kind of quality control measures, understanding the limitations of the tech, that's really

Speaker 1: what the key piece is and how we can use it. Okay. So how much time will it take before the system is actually trained, the software has learned? Will we see any kind of immediate benefits? Will there be long-term benefits? How does that work?

Speaker 2: It is astonishingly, amazingly, immediately beneficial. Wow. Amazing. It's really what we're working on now are one, what are the good use cases for it? For example, I uploaded a complaint in a case of the document that you start a case with saying you done me wrong and here's why. And this was a very complicated, I purposely put this, very complicated. There were many third parties. It described 50 years of an environmental case of effluents and chemicals and all sorts of things. And I asked it a question and said, imagine that I'm the CEO of this defendant, what deposition questions can I expect? And it churned for four or five minutes, which was an eternity for these things and spit out 104 deposition questions that were astonishingly good. Really understood the different chemicals, the different parties, the different relationships. And so it's a good head start in creating a legal product. It will never replace lawyers, mainly because we really do have to verify everything it does. And even the deposition questions it spit out, they're very good, but the nuance of what do I ask first and how do I set a layer of concepts to my deponent to get them to think in a certain way? Like all that is very much the art of being a lawyer.

Speaker 1: Yeah, absolutely. That's a relief, right? Fantastic. Maybe one last question. I can imagine that some clients, some of our viewers are thinking about data protection, client confidentiality. How do we safeguard those with a tool like CoCounsel?

Speaker 2: And that's one of the most important things we're looking at. One of the benefits of working with Case Text and their product CoCounsel is the terms of use that we have with them are ironclad. And so it protects our client confidentiality and also the secret sauce that makes us lawyers. And that's very different than just using something off the shelf like ChatGPT, where if you look at those terms of use, those protections aren't there. The next thing we're looking at is you can actually connect into GPT-4 directly without necessarily going through a third party like Case Text. And so one of the things we're looking at is designing our own systems for particular purposes, training these large language models on DLA data so it has a DLA voice. And so we're at the very beginning stages of this utterly transformative and disruptive technology, but it is terribly exciting technology

Speaker 1: to be a part of. Absolutely, absolutely. Exciting times ahead of us. And thank you so much for this conversation, Bennett. I'm sure we'll speak soon on other topics related to this. Thanks very much. And to our viewers at home, if you're interested in AI and you want to read more, we have a dedicated AI resource center on our website. So have a look at dlapiper.com and look for Focus on Artificial Intelligence. Thank you and see you again.