WEBVTT
00:00:00.160 -->
00:00:03.839Hello and welcome to the Emerging Litigation Podcast and webinar.
00:00:03.839 -->
00:00:05.839I'm your host, Tom Hagy.
00:00:05.839 -->
00:00:07.599So today we're doing something a little different.
00:00:07.599 -->
00:00:17.760This is not going to appear just on major podcast platforms like Apple, Podcasts, Spotify, and on and on YouTube as a video, but it's in in many other places.
00:00:17.760 -->
00:00:22.719But it's also going to appear on the Cerify Legal Edge platform as a CLE program.
00:00:22.719 -->
00:00:36.240It's going to be a comp there, it will be accompanied by presentation materials, PowerPoint, et cetera, for attorneys seeking practical guidance on one of the more significant advances in technology affecting litigation and trial practice today.
00:00:36.240 -->
00:00:41.679And so that's what we're going to talk about is AI as litigation infrastructure.
00:00:41.679 -->
00:00:47.920So we'll talk about building better case strategies, workflows, simulations, client reporting.
00:00:47.920 -->
00:00:55.920A lot of what we're going to talk about too is getting ready for trial, as one of our guests was doing and as we did this, so we interrupted that for him.
00:00:55.920 -->
00:01:06.879And uh so joining me are two lawyers from Reed Smith's who bring both litigation experience and hands-on experience applying artificial intelligence in the legal practice.
00:01:06.879 -->
00:01:14.879Adam Masaro is a partner, an intellectual property and corporate litigator who handles complex high-stakes disputes nationwide.
00:01:14.879 -->
00:01:26.319His experience includes quite a number of jury trials, trade secret litigation, software and internet technologies, class actions, complex appeals, and a broad range of commercial disputes.
00:01:26.319 -->
00:01:30.879This is his second appearance on the podcast, which is now a podcast webinar.
00:01:30.879 -->
00:01:32.159I'll just keep saying that.
00:01:32.159 -->
00:01:34.799Dan Brown has a unique title.
00:01:34.799 -->
00:01:40.079He's an innovation lawyer at Reed Smith, and he's focused on applying artificial intelligence.
00:01:40.079 -->
00:01:49.040Before moving into legal innovation, he practiced as a litigator, launched his own firm, built a book of business, and represented parties on both sides of the litigation.
00:01:49.040 -->
00:01:58.959So today he focuses on helping attorneys make practical and responsible use of this incredible, incredible and advancing technology AI solutions.
00:01:58.959 -->
00:02:02.560So again, welcome, welcoming Adam back to the program.
00:02:02.560 -->
00:02:09.520And he'll explain in a moment that a great deal has changed since he and I spoke over a year ago.
00:02:09.520 -->
00:02:18.879When he and I first talked about AI and litigation, a lot of lawyers were still evaluating whether these tools even belonged in the legal practice at all.
00:02:18.879 -->
00:02:22.560A lot of people are wondering whether it belongs in our lives at all.
00:02:22.560 -->
00:02:28.159But right now, what we're doing is we're embracing the power of it and leaving the rest to everybody else.
00:02:28.159 -->
00:02:42.719Most importantly, we'll talk about how users can use these technologies not simply to work faster, and you certainly will, but to think more clearly, prepare more effectively, communicate more transparently, and in the end, serve clients better.
00:02:42.719 -->
00:02:47.199And with that, here's my conversation with Adam Massaro and Dan Brown of Reed Smith.
00:02:47.199 -->
00:02:48.479I hope you enjoy it.
00:02:48.479 -->
00:03:03.520So Dan Brown and Adam Massaro, welcome to the Emerging Litigation Podcast.
00:03:03.520 -->
00:03:04.800I appreciate you doing this.
00:03:04.800 -->
00:03:13.520The um uh so Adam, you and I talked uh before on a previous podcast, and I'll uh and I'll link to that uh as well.
00:03:13.520 -->
00:03:21.039But you know, then we we were focusing on ways that AI could help a lawyer prepare for trial, like organizing all the all the materials.
00:03:21.039 -->
00:03:22.240You and I were talking about that.
00:03:22.240 -->
00:03:30.560Like I don't know how you guys hold all this all this stuff in your head, but um, you know, developing cross-examination, looking for weaknesses.
00:03:30.560 -->
00:03:39.120So so now when we talk about AI and talking to you, we describe it as a as a litigation operating system.
00:03:39.120 -->
00:03:43.919So can you describe what that means in in practical terms?
00:03:45.599 -->
00:03:47.039Yeah, I certainly can.
00:03:47.039 -->
00:03:49.599I think the best way to capture is the evolution.
00:03:49.599 -->
00:03:58.479So last year, when we really went all in on AI, it was really using the tools to make the litigation process smoother, faster, smarter.
00:03:58.479 -->
00:04:10.719And that all happened last year, but it's still very separate where you know, put a brief into an AI tool like Harvey, have it get it, you know, better, identify issues, things like that, and then pull it out.
00:04:10.719 -->
00:04:16.240And this year, what we really started to think about was putting the entire ecosystem together.
00:04:16.240 -->
00:04:18.399And that's the infrastructure you're talking about.
00:04:18.399 -->
00:04:29.120And the way we see the practice going is where AI is at its best, is when it's plugged into each part of the litigation phase so that we get information through the process quicker.
00:04:29.120 -->
00:04:36.879So, an example of that, what I mean now is you can use any LLM as sort of your core vault to have the core case information.
00:04:36.879 -->
00:04:49.600But now it's thinking about how do we start to run processes over that AI throughout the entire litigation process so that, for example, when a new fact comes up in the case, we can quickly assess: is this significant?
00:04:49.600 -->
00:04:50.879Is it not significant?
00:04:50.879 -->
00:04:55.199Is there a new legal issue that was not addressed earlier in the case that now we can address?
00:04:55.199 -->
00:05:09.839So the idea is putting all the different pieces together in one infrastructure so that our lawyers, when they're building cases from the very beginning, are doing it in the system so that we're getting information about the case quicker and now integrating the different AI tools together.
00:05:10.079 -->
00:05:14.639You know, and Dan, I don't know if you have anything to add, just just just jump in.
00:05:15.040 -->
00:05:32.399Yeah, I think to dovetail on Adam's point, the the notion is that it's gonna be model agnostic and to a certain extent uh harness agnostic because the operating system is really just an expression of the state of play in litigation.
00:05:32.399 -->
00:05:44.240So you have a bunch of claims, you have a bunch of defenses, you have factual propositions that line up with those claims and defenses, and then you have individual facts that compose into those propositions.
00:05:44.240 -->
00:05:50.720And all of that is not possible to cram into the context window of a single AI session.
00:05:50.720 -->
00:05:57.839So you have all of that information organized and accessible to the LLM, but outside of the specific software.
00:05:57.839 -->
00:06:11.120And a lot of the AI-powered litigation workflows involve kind of managing that massive information to make sure that it's accurate, updated, properly governed, reflects disputed facts and tensions that exist in the record.
00:06:11.120 -->
00:06:24.879And that way, both for the humans operating the AI and the agents spawned from the AI, they get access to that that corpus of information and can push the litigation forward based on the state of play.
00:06:25.439 -->
00:06:26.079Gotcha.
00:06:26.079 -->
00:06:36.319So are so when you're talking about this, uh are we talking about a a um a single platform or are we talking about integrating different platforms?
00:06:37.920 -->
00:06:41.360So I'll I'll frame it at a high level, and then Dan can get a little more precise.
00:06:41.360 -->
00:06:56.319The the tool that we've been working on internally and testing is set up so that it basically, as Dan mentioned, is agnostic, meaning you can pull from different, you know, models, Harvey, et cetera, to get the outputs.
00:06:56.319 -->
00:07:03.360And then we basically marshal them in one system that we've created so that we have the core facts.
00:07:03.360 -->
00:07:09.279But to Dan's point about having too much information, it'll overload and then you'll get information fidelity problems.
00:07:09.279 -->
00:07:14.319So the key is knowing where the systems will work and where they'll break you down.
00:07:14.319 -->
00:07:20.480But think of this like basically a general that now has the core case theory and the core facts.
00:07:20.480 -->
00:07:33.120But some of the strategic pieces we did was come up with ways to slice the facts thin enough and the case law thin enough and the information about the matter thin enough that we can actually get real good output as a result.
00:07:33.120 -->
00:07:38.879And then that overlay, the marshaling of all this information, that's really where it all comes together.
00:07:38.879 -->
00:07:45.600And now we have a repeatable sort of infrastructure that can start to allow us then to expand out and things like that.
00:07:45.600 -->
00:07:59.120But yeah, the best way to think about it is just sort of a nerve center that has the core information built in a way that we know is gonna give us strong, credible output without some of the errors we've seen in the past using AI.
00:07:59.120 -->
00:08:00.000Okay.
00:08:01.040 -->
00:08:09.199And then yeah, it's gonna, I mean, harnesses and models and all of that associated software, it's gonna come and go.
00:08:09.199 -->
00:08:11.600The market is relatively frothy right now.
00:08:11.600 -->
00:08:14.160There are new entrants and a lot of investment.
00:08:14.160 -->
00:08:28.959And so what Ab and I are talking about is building something that exists outside of any particular company's software or system, but then is accessible to or usable with any particular company's software or system.
00:08:28.959 -->
00:08:37.120It's basically a way to collect, store, categorize, and tag a very large volume of information.
00:08:37.120 -->
00:08:44.000And then you can select the bits and pieces of that information to bring into context to then work with AI.
00:08:44.000 -->
00:08:50.559And then in that work, you produce additional outputs that then get added back into that ball of information.
00:08:50.559 -->
00:08:53.440So that's why I mean I was saying state of play.
00:08:53.440 -->
00:08:58.080I essentially see a lot of litigation management as managing a complex state.
00:08:58.080 -->
00:09:07.679And so since you're managing a complex state, the best way to do that is to actually compose and collect the entirety of that state.
00:09:07.679 -->
00:09:11.519And then that state will be updated as the litigation progresses.
00:09:11.519 -->
00:09:23.519And then both the human operator, the lawyers, and the AI agents can access bits and pieces of that state, bring it into context, work, put, produce output, and that output composes back into the state.
00:09:23.840 -->
00:09:24.240Yeah.
00:09:24.240 -->
00:09:29.919So that that gets the the I think you're already answering kind of the next thing I wanted to talk about.
00:09:29.919 -->
00:09:32.799My experience is, you know, I'm not a trial lawyer.
00:09:32.799 -->
00:09:35.120I'm I'm mostly just a writer.
00:09:35.120 -->
00:09:47.200But um I found early on experimenting with different chatbots, um I was seeing, yes, they do hallucinate, and then one was one was just trying to make me happy and give me an answer that I wanted.
00:09:47.200 -->
00:09:54.399It found some, it found some antitrust cases somewhere in some remote part of uh some county in Texas.
00:09:54.399 -->
00:09:59.440Well, that's not where they would be, but it was trying to like look it was trying to satisfy me, and I was looking for these things.
00:09:59.440 -->
00:10:09.039It's like uh and then after a while I started learning how to train and which which you know I happen to use Copilot because it relies on my enterprise data.
00:10:09.039 -->
00:10:17.519So so something I learned just as a writer was if you can point it at information you already trust and then work with it.
00:10:17.519 -->
00:10:20.879Um that's where it becomes really powerful.
00:10:20.879 -->
00:10:23.600And so I I love it for all of those reasons.
00:10:23.600 -->
00:10:33.759So so that's what you're it seems that's what you're talking about when people think um, oh, I'll use I'll use a chat bot and I'll ask a general question and and get a broad answer.
00:10:33.759 -->
00:10:36.399What you're describing is kind of a reverse.
00:10:36.399 -->
00:10:39.679You're building that that body uh of information.
00:10:39.679 -->
00:10:52.480And so I guess that's your answer to how you build a reliable case-specific body of information that the team can use without worrying about necessarily hallucinations or pulling things that are incorrect.
00:10:52.480 -->
00:10:54.879Is that I guess that's what you've just described.
00:10:54.879 -->
00:10:55.679Is that right?
00:10:56.480 -->
00:10:57.519I I think so.
00:10:57.519 -->
00:11:20.000I think in the to your point about the credibility of the information, part of what we've designed is a function to check the credibility of a fact, check the credibility of the legal inputs to make sure that it's a high enough level of accuracy that we then feel comfortable basically closing off that issue for the time being in the matter.
00:11:20.000 -->
00:11:23.440So that we're not constantly just sweeping the ocean for the same issues.
00:11:23.440 -->
00:11:38.399So if you think about it that way, if you think about if you're gonna build a case on the plaintiff's side, you're trying to get as many concrete facts and legal principles so that you have a strong foundation, then you build the architecture around it, and then you have the thematic aspect of the case for why it feels good you should win.
00:11:38.399 -->
00:11:41.120A beautiful case has all of those things.
00:11:41.120 -->
00:11:46.240But you're building something while at the same time, the other side is trying to tear down your building.
00:11:46.240 -->
00:11:51.120And so we have to be realistic that that is the push in the pull of litigation.
00:11:51.120 -->
00:11:56.159But as cases go on, certain facts become largely undisputed in the case.
00:11:56.159 -->
00:11:59.519And in that situation, then we have a high level of confidence.
00:11:59.519 -->
00:12:05.679So the way that we've built the tool that we're working through now is to really know, like, okay, these facts look very rock solid.
00:12:05.679 -->
00:12:07.039We have admissions on them.
00:12:07.039 -->
00:12:09.840Those facts have not changed over three depositions.
00:12:09.840 -->
00:12:15.039At that point, we have a high enough probability to say that fact is likely going to sit there and be good.
00:12:15.039 -->
00:12:16.960The net effect of that is two things.
00:12:16.960 -->
00:12:28.799One, now we're not overloading the either harnesses or the LLMs with extra extra information that just isn't necessary to do the specific task because now we know those facts are largely rock solid.
00:12:28.799 -->
00:12:39.039The related point there too is we don't need to then do discovery on every last fact in the case if we have a high probability that that case is gonna be landing and accurate.
00:12:39.039 -->
00:12:41.919So we can really focus on what's gonna win or drive the case.
00:12:41.919 -->
00:12:47.919Because at the end of the day, there's probably 10 documents, 10 facts that swing a case from winning to losing.
00:12:47.919 -->
00:12:51.840There's probably two to three legal issues that push the case from winning to losing.
00:12:51.840 -->
00:12:59.519And so rather than spend an inordinate amount of time on all the things collectively, it's refocusing that attention.
00:12:59.519 -->
00:13:10.480And we've also built the tool in a way that the flip side is if something does verge where everyone thought this was the contract, and then you know, magically another version shows up.
00:13:10.480 -->
00:13:13.679At that point, then it's going to flag it as a disputed issue.
00:13:13.679 -->
00:13:15.279And we may have to revisit it.
00:13:15.279 -->
00:13:24.879But the idea is if we're working collectively, partners and associates together, we should be building something that we have a high level of variability in before trial.
00:13:24.879 -->
00:13:28.080So that by the time we get to trial, we actually know what we're fighting over.
00:13:28.080 -->
00:13:29.919And the same is true on the defense side.
00:13:29.919 -->
00:13:38.080We're trying to build up the core facts to win the case and get as many of those core facts locked in early so that then we're focused on the key issues.
00:13:38.080 -->
00:13:46.399So I think that's really the difference is sort of re-envisioning the litigation process rather than expansive discovery, expansive exploration.
00:13:46.399 -->
00:13:49.679Like let's know what we actually need to know to drive the case.
00:13:49.679 -->
00:13:53.440If we do, the two benefits are A, cases move faster.
00:13:53.440 -->
00:13:55.440Faster cases are less expensive.
00:13:55.440 -->
00:14:03.759And if we know the target we're hitting for trial well before trial, we're also going to be able to control spend from a staffing perspective as well.
00:14:03.759 -->
00:14:07.759And those are really the hardest variables in litigation, is when things go off the rails.
00:14:07.759 -->
00:14:18.480And keeping on track, if your case starts from you know one theory and retains that theory all the way through trial, you've got a really good shot of winning and presenting a defensible case.
00:14:18.480 -->
00:14:22.960But losing the end result is oftentimes the hardest part about a trial.
00:14:22.960 -->
00:14:24.879Gotcha.
00:14:25.279 -->
00:14:25.759Damn.
00:14:26.639 -->
00:14:33.840Yeah, I mean, I draw uh a fairly bright line between intelligence and knowledge.
00:14:33.840 -->
00:14:39.440And I do not count on an artificial intelligence system to know things.
00:14:39.440 -->
00:14:41.840Uh so I don't treat it like a Google search.
00:14:41.840 -->
00:14:50.639I don't want it to tell me what it recalls from its training data, and I don't want it to make inferences about what is true or not true.
00:14:50.639 -->
00:14:56.000And so um, you know, we talk about closed universe versus open universe.
00:14:56.000 -->
00:15:02.399Like if you open up uh ChatGPT and just treat it like a Google search, it's open universe.
00:15:02.399 -->
00:15:12.080It's gonna maybe search the web, it's gonna maybe think through uh what it recalls from its training data and give you an answer that it thinks you expect to see.
00:15:12.080 -->
00:15:16.879And that is a recipe for disaster in the context of litigation.
00:15:16.879 -->
00:15:28.480Um so a lot of what we're doing here is trying to match the intelligence that the system provides with the knowledge that we bring to the table.
00:15:28.480 -->
00:15:31.600And that includes, you know, things as simple as case law.
00:15:31.600 -->
00:15:38.159I don't want the artificial intelligence system to tell me what cases it thinks may apply.
00:15:38.159 -->
00:15:53.360I want to bring the cases that I have understood to apply and that I believe govern the case, and then have the uh AI system analyze those cases, maybe summarize them, extract bits and pieces from them, apply them to facts, et cetera, et cetera.
00:15:53.360 -->
00:15:55.360And same thing with the facts of the case.
00:15:55.360 -->
00:16:04.000Obviously, in a piece of you know, private civil litigation, there should be nothing in the system's training data about that particular dispute.
00:16:04.000 -->
00:16:18.879And so all of the facts that are relevant to the case are gonna be things that I bring to the system and show to the system, which is why that the sort of corpus of information, that state that I was talking about before, needs to exist outside of the system.
00:16:18.879 -->
00:16:22.879Uh, because it has to be sort of a permanent human auditable record.
00:16:22.879 -->
00:16:38.799And then the bits and pieces that are relevant to a memo that I'm working on, a motion that I'm working on, an issue that I'm trying to research and analyze, I bring those pieces of the state into the system and then have the system um, you know, review, analyze, and provide associated output.
00:16:38.799 -->
00:17:13.279And then to Adam's point as well, it not only allows you to sort of focus on what matters, because if you can look at, you know, at a glance from a 30,000-foot view, the entirety of what's happening in a piece of litigation, and then sort of zoom in where you need to to make sure you understand the granular facts, um, you will not only be able to focus your time, effort, and money on what is actually going to drive the case, but it's going to point out any sort of deficiencies that you might have papered over without really thinking about it as you head towards trial.
00:17:13.279 -->
00:17:23.039Maybe there's the simplest fact in the world that would not be difficult to prove at trial, but you just didn't pull in that piece of paper or you didn't authenticate that record.
00:17:23.039 -->
00:17:32.960Having the state examinable and something that you're looking at constantly as the litigation proceeds is going to shine a spotlight on that hole in your record.
00:17:32.960 -->
00:17:42.319Whereas if you don't have some sort of representation of the state of the matter, you might miss that and only realize it on the eve of trial.
00:17:42.319 -->
00:17:50.079And then suddenly your settlement posture is very, very different than it would have been if you had your act together and you knew the entire state of play going into trial.
00:17:50.079 -->
00:18:03.920So it both allows focus and I think uh has an error prevention mechanism as well, because the entirety of what matters in a litigation is visible, auditable, navigable, and easy to understand.
00:18:04.240 -->
00:18:17.759You didn't want it to tell you which cases maybe uh were relevant or make inferences that uh uh in in my world, what I ended up doing was um I I use it for like for a uh for first draft.
00:18:17.759 -->
00:18:20.720I'll give it a bunch of my own documents and then we'll do a first draft.
00:18:20.720 -->
00:18:25.599And then I'll then I'll say, okay, let's let's look out on the web too and see what might relate.
00:18:25.599 -->
00:18:34.400And then uh uh but sometimes what it'll do if I'm writing about a case, uh it'll start off the first draft, it'll say, in a landmark case.
00:18:34.400 -->
00:18:39.920Um it's like, okay, it's not a landmark case, uh, and it's not for you to say.
00:18:39.920 -->
00:18:46.240It may not even be for me to say yet, um, but it it I would after a while I would just say, for example, don't use adjectives.
00:18:46.240 -->
00:18:48.079Let's let's just let me do that.
00:18:48.079 -->
00:18:50.319If we're gonna screw up, I'd rather it would be me.
00:18:50.319 -->
00:18:51.599So that was one thing.
00:18:51.599 -->
00:19:06.480But I would I would think too, and and um and I'll use it differently, but uh I would think too, when you have massive amounts of uh volumes of testimony, uh finding contradictory statements, um this must really be useful.
00:19:06.480 -->
00:19:08.720AI must really be useful there.
00:19:10.240 -->
00:19:11.519Yeah, without a doubt.
00:19:11.519 -->
00:19:21.680Um I mean, one of the things that I talk about quite a bit uh when I'm coaching people on how to use the technology effectively is that one of AI's superpowers is speed readers.
00:19:21.680 -->
00:19:28.640And it you again, you have to sort of trust but verify uh when something is speed reading for you.
00:19:28.640 -->
00:19:53.039But just the the raw power of taking hundreds upon hundreds of pages of text that would take you hours to go through and being able to within minutes really pinpoint the bits and pieces in that corpus that matter is uh I mean it people talk about the efficiency play a lot because the efficiency play is fairly obvious.
00:19:53.039 -->
00:20:19.920But setting aside efficiency, if you don't have to spend six hours combing through, you know, 1,500 pages of testimony, and instead you can do it in six minutes, then you do have all of that time to think through different, more creative, more innovative ways to use that testimony, as opposed to having to, you know, clear out your glasses, take a deep breath, and dive back into the page over and over again.
00:20:19.920 -->
00:20:28.880And I mean, Adam and I joke, because people that talk about like, oh, you know, AI is gonna take away, like, you know, uh young legal professionals' jobs.
00:20:28.880 -->
00:20:30.559We haven't seen any of that yet.
00:20:30.559 -->
00:20:38.559And like, oh, AI is sort of it is potentially like a laziness enablement machine because the associates can just have the AI do the work.
00:20:38.559 -->
00:20:50.880What what what we see happening more and more, and that we think will happen more and more, is that these young attorneys' jobs is getting more difficult because the AI is doing all of the easy stuff.
00:20:50.880 -->
00:21:02.319You don't get that sort of 20-minute break to not really think hard while you're trying to navigate through a pile of folders and find the right document or the right clause in the contract or what have you.
00:21:02.319 -->
00:21:04.240That's delivered to you instantly.
00:21:04.240 -->
00:21:21.440So you have to go from difficult judgment thought-provoking task right into another difficult judgment-heavy thought-provoking task, which I think is is is more challenging and will force attorneys to develop sharper skills faster and earlier in there.
00:21:22.240 -->
00:21:22.720Yeah.
00:21:22.720 -->
00:21:26.400I can I can relate to what you said uh extremely well.
00:21:26.400 -->
00:21:29.920Um I I've been a legal writer all of my life.
00:21:29.920 -->
00:21:33.119I'm almost 70 now and in my 20s.
00:21:33.119 -->
00:21:42.319Um, I literally had a deposition that I wanted to write about because it was about it was a really core issue in this area of litigation.
00:21:42.319 -->
00:21:45.599This was the guy who actually wrote the clause that everybody was arguing about.
00:21:45.599 -->
00:21:47.440So I really wanted to pay a lot of attention to it.
00:21:47.440 -->
00:21:50.079But 1500 pages, it was days.
00:21:50.079 -->
00:21:51.680I was days doing that.
00:21:51.680 -->
00:21:59.759And um, and you know, of course, we would read a lot of opinions and things too, and we'd have to do summaries, but then we'd have to talk to the lawyers and get context and write.
00:21:59.759 -->
00:22:05.440Stories around them, but that initial reading of a 150-page opinion, you got fat.
00:22:05.440 -->
00:22:06.720I mean, you got very fast.
00:22:06.720 -->
00:22:07.519You're on deadline.
00:22:07.519 -->
00:22:08.799You got quick at skimming.
00:22:08.799 -->
00:22:10.799You didn't read certain things, you know.
00:22:10.799 -->
00:22:13.359But it's it's almost embarrassing now.
00:22:13.359 -->
00:22:16.799I'll say I'll I'll give it 150 pages and it's in seconds.
00:22:16.799 -->
00:22:30.079The thing I like about it, I'll get a summary, but um what I also like about it is is that the way I use it, it's not uh over-paraphrasing anything and it's telling me exactly where it got what it got.
00:22:30.079 -->
00:22:37.680So then, okay, well, now I'm just an editor of the thing and I can go back and verify, trust but verify is what you were talking about.
00:22:37.680 -->
00:22:42.000Anybody doing legal research, you know, they're they're accustomed to search terms.
00:22:42.000 -->
00:22:49.279And I mean, uh, I remember learning bullion, and certainly then it became more freestyle when you went to things like Google.
00:22:49.279 -->
00:22:59.039But but now um the prompt prompting, and Dan, I'm if you're teaching this, I'm sure this is a part of what you teach, is you know, effective prompts.
00:22:59.039 -->
00:23:03.920So, so what what how would you describe effective or strategic prompting?
00:23:03.920 -->
00:23:07.920How would you say it's also different from ordinary legal research?
00:23:08.559 -->
00:23:19.440So it's it's funny because uh in the same way that you described, you know, early uh legal research using Boolean and then things kind of moved on to natural language.
00:23:19.440 -->
00:23:30.960We've had an accelerated version of that in artificial intelligence, where earlier on, 2024, 2025, prompt engineering was important.
00:23:30.960 -->
00:23:41.839And if you wanted to get high quality output from an AI system, you needed to put a lot of effort into figuring out what the prompt should be.
00:23:41.839 -->
00:23:46.160Uh the systems have accelerated quite a bit.
00:23:46.160 -->
00:23:57.119And now uh I'm not gonna say that prompt engineering is dead, but I think that prompt engineering is less important than it was even six or eight months ago.
00:23:57.119 -->
00:24:06.799And now when I think about prompting uh an artificial intelligence system, I think about providing an assignment to a colleague.
00:24:06.799 -->
00:24:14.240Um so I'm not gonna over-engineer or over-prescribe every single little needly detail.
00:24:14.240 -->
00:24:18.720I think uh earlier prompt engineering did require that if you wanted good output.
00:24:18.720 -->
00:24:22.400You had to say not just where you want to go, but exactly how to get there.
00:24:22.400 -->
00:24:35.839Like each turn, each lane change, every single landmark that you would see along the way has to be described in order to provide the AI system with sufficient information to complete the task.
00:24:35.839 -->
00:24:41.359Now, uh, generally speaking, I just describe where I want to go.
00:24:41.359 -->
00:24:46.799I provide the relevant context that any sort of similar colleague would need to get there.
00:24:46.799 -->
00:24:54.559And then I provide guardrails about things to definitely avoid for sure, and then things to definitely do for sure.
00:24:54.559 -->
00:24:59.279But then I don't over-prescribe how exactly to go about getting there.
00:24:59.279 -->
00:25:06.400Because especially on complex tasks, the way that the AI system might complete it might be a little bit different from how I would complete it.
00:25:06.400 -->
00:25:14.000And I and I literally don't want to micromanage, I just want to explain the destination as clearly as I as I can I can make it.
00:25:14.000 -->
00:25:20.640So I'm I'm a big fan of uh like voice features now, and I tend to dictate my prompts.
00:25:20.640 -->
00:25:22.720Uh, I speak faster than I type.
00:25:22.720 -->
00:25:29.839I'm used to speaking, I'm used to giving assignments to colleagues orally uh and and sort of having a conversation.
00:25:29.839 -->
00:25:42.079So I will often just you know hit the dictate button, speak for 90 seconds, a couple of minutes, um, look at the transcript to make sure that it's about right.
00:25:42.079 -->
00:25:49.039On some of these systems, uh they have like a way to improve the prompt that kind of can take something a little bit messy and structure it up a little bit.
00:25:49.039 -->
00:25:51.680Sometimes that's helpful, sometimes it's not.
00:25:51.680 -->
00:25:56.640Um and then and then from there I let it go and then I iterate.
00:25:56.640 -->
00:26:02.079So, you know, I'm not gonna dump a prompt in, get the output, and then move on.
00:26:02.079 -->
00:26:14.400I'm going to assign a task, see what it comes back with, and then ask follow-up questions, press on things that don't make sense, and try to collaborate with the machine essentially.
00:26:14.400 -->
00:26:25.759Because fundamentally, what I'm trying to do is just try try to make the system understand as clearly as I possibly can what it is that I want and what it is that I'm trying to accomplish.
00:26:25.759 -->
00:26:27.759I call it the mind meld, right?
00:26:27.759 -->
00:26:45.920And you try to do that with colleagues as well, because if you're working with a junior lawyer and you're trying to get them to write a memo, write a brief, et cetera, et cetera, you want them to sort of understand your intention and understand what good looks like and the guardrails of how to get there and then the context of what they need to accomplish the task.
00:26:45.920 -->
00:26:48.559And I go about prompting in much the same way.
00:26:48.960 -->
00:26:49.440Yeah.
00:26:49.440 -->
00:26:51.680Well, also very relatable.
00:26:51.680 -->
00:26:56.880And you you hit on one thing that um you press on things that don't make sense.
00:26:56.880 -->
00:27:07.599Um that that is something that I think um that is where experience uh coupled with AI becomes very powerful.
00:27:07.599 -->
00:27:16.079Um because you so I've been in, I've been an editor, I've been an editor on off and on for you know 45, God, 45 years.
00:27:16.079 -->
00:27:26.640And uh so you kind of you kind of know when a when a a newer writer is giving you something, you you you look at everything as if, you know, with a skeptical eye or critical eye.
00:27:26.640 -->
00:27:30.720So and it's the same with it's the same with what I get back from AI.
00:27:30.720 -->
00:27:35.279You know, because I've been doing this for so long, I couldn't have done this when I was 25.
00:27:35.279 -->
00:27:41.920But now that I've seen a lot of really good writers, you know, are you, you know, is this really backed up what you're saying?
00:27:41.920 -->
00:27:44.160Um is this the right way to say it?
00:27:44.160 -->
00:27:46.240This just doesn't smell right sometimes.
00:27:46.240 -->
00:27:51.839Um, but so that's that's where the uh experience, I think, the human experience comes into play.
00:27:51.839 -->
00:28:04.559Now maybe AI will catch up with that someday, maybe it already has, but it just seems like you're gonna know stuff, um, you're gonna have an intuition or an experience that you you can apply to the to what what the uh product is.
00:28:06.240 -->
00:28:16.960And the only thing I'd add is I think what we're seeing is the delta between good and great lawyers is getting broader.
00:28:16.960 -->
00:28:24.319Because if you're harnessing tools and you're building off your years of experience, you are amping up your practice.
00:28:24.319 -->
00:28:31.200You know, just in the just recently I had six depositions over uh a seven-day period.
00:28:31.200 -->
00:28:46.319And my ability to flip from one to the other and build aggregated theories has just gotten that much stronger because I'm ingesting and using my internal thought processes to get to the core issues that I know matter.
00:28:46.319 -->
00:28:52.480But my ability now to ingest information and be strategic about it, I think is really what's changing.
00:28:52.480 -->
00:29:02.640So if you're leaning in and you're already, by definition, creative because litigation is a creative game that we all look at ways that we can better position, because the facts really are the facts.
00:29:02.640 -->
00:29:05.359You got to figure out how you can better frame and address them.
00:29:05.359 -->
00:29:22.720So I believe that as I sort of see this process going forward, it's it's really valuing like what is the augmenter that someone can do while tapping the right tools and being thoughtful, but tracking where they can be deficient, but then also using them for the best purposes.
00:29:22.720 -->
00:29:27.839But when you combine those two, the delta, I think is gonna be more pronounced over time.
00:29:27.839 -->
00:29:31.680And that's really what will be interesting to see sort of what that progression looks like over the law.
00:29:31.680 -->
00:29:34.880Because look, we're and the one thing we've left out is the judges.
00:29:34.880 -->
00:29:39.200The judges are also using AI in different forms and fashions.
00:29:39.200 -->
00:29:40.799And that's not a bad thing.
00:29:40.799 -->
00:29:42.400Judges are overloaded.
00:29:42.400 -->
00:29:52.319So I think when you see everybody using it, you're gonna see an elevation of the bright people that really embrace it and build off that historical knowledge.
00:29:52.319 -->
00:30:00.799And the judges will be more crisp, they will not get as confused by bad arguments, and that's all just a net benefit for the justice system as a whole.
00:30:00.799 -->
00:30:07.440And that's what I think we're gonna see more of now that we've sort of passed through this wave of the hallucination issues and things like that.
00:30:07.440 -->
00:30:14.640Now it's really like how are the best practitioners, both on the bench and in private practice, using it to go forward.
00:30:15.920 -->
00:30:22.319Well, and then a funny corollary to that too is uh it's not just judges using AI, it's also pro se plaintiffs.
00:30:22.319 -->
00:30:31.839And uh pro se plaintiffs are using AI to great effect uh because it can generate persuasive seeming text very, very quickly.
00:30:31.839 -->
00:30:46.640Uh so it'll be interesting to see how how the justice system handles uh, I think, just the output that AI allows from both uh attorneys and uh and pro se litigants alike.
00:30:47.119 -->
00:30:47.440Yeah.
00:30:47.440 -->
00:30:49.119Yeah.
00:30:49.119 -->
00:30:55.039And and Dan, you did say one thing too about you know associates not not losing their jobs but having more to do.
00:30:55.039 -->
00:30:56.240I think that's absolutely right.
00:30:56.240 -->
00:31:04.160If I um it's already been my experience that say I'm I want to write an R you know four articles a week or something.
00:31:04.160 -->
00:31:11.039Well now I I can write twenty, but now I gotta now I really have to examine all of them carefully.
00:31:11.599 -->
00:31:11.839Yeah.
00:31:11.839 -->
00:31:30.559Well, and and I mean I I I I say this a lot, is like so you know, intelligence becomes cheap and analysis becomes cheap, and the bottleneck becomes evaluating the quality of the intelligence and analysis that is uh coming at you at at high speed and at volume.
00:31:30.880 -->
00:31:31.519Yeah, yeah.
00:31:31.519 -->
00:31:33.519It's it's a necessary bottleneck.
00:31:33.519 -->
00:31:38.319You get so involved in a case and you become familiar with your own theories and arguments.
00:31:38.319 -->
00:31:49.119And I'm wondering um if AI can be used to help, you know, challenge your theory uh and see whether it's clearly supported and kind of push back on your theory.
00:31:49.119 -->
00:31:50.640Is it used that way?
00:31:51.920 -->
00:31:53.279The answer is absolutely.
00:31:53.279 -->
00:32:01.200And I think that's the the best example I can give is AI is imperfect.
00:32:01.200 -->
00:32:04.720Juries and judges are imperfect as well.
00:32:04.720 -->
00:32:12.559And that to me is actually a net benefit where you know juries are having information overload, judges have information overload.
00:32:12.559 -->
00:32:22.559So when I'm stress testing filings now, part of what I'm always trying to do is see, even with imperfect information, what is the net outcome?
00:32:22.559 -->
00:32:39.200And you know, some of the best ways to test that, for example, would be when you're going through a briefing process, as you get far enough on where you have your arguments and the other side's core arguments before you finish the briefing, have the AI model draft an order and be very neutral about it.
00:32:39.200 -->
00:32:46.319But that's telling me information about what's getting lost, what's not getting lost in the process and showing where my weaknesses are.
00:32:46.319 -->
00:32:55.039I think the some of the best ways I've seen it is as we get enough information in the case, and as Dan talked about, then we have a lot of volume.
00:32:55.039 -->
00:33:05.359The the infrastructure we just talked about that we're building to have high fidelity, that's great for purposes of making sure we have really, really good information for our case.
00:33:05.359 -->
00:33:22.480But if we're talking about challenging our arguments, sometimes imperfect information is actually the best because we've all seen that, you know, in cases, a sort of ancillary theme or theory that really isn't the strongest, it can still win the day because it lands with the right people that are deciding the case.
00:33:22.480 -->
00:33:27.839So I think it's all the more imperative to really use the predictive aspects of it.
00:33:27.839 -->
00:33:32.720And that's just building sort of the appropriate level of information so you're getting good feedback.
00:33:32.720 -->
00:33:39.200So for instance, you know, if you're asking just the models to tell you what's going to happen in the case, you're not giving enough context.
00:33:39.200 -->
00:33:44.160You're gonna have to build around, you know, who is the decider, what is the demographic, all those things.
00:33:44.160 -->
00:33:54.480And then separately, you know, don't be afraid to push it to its logical conclusion, the sense of, you know, if I didn't include these facts, would the outcome change?
00:33:54.480 -->
00:34:05.839You know, start to stress the sensitivity there because the arguments that you think are your top three may not actually be your top three once you see how somebody that's not as embedded in the case is going to think about it.
00:34:05.839 -->
00:34:13.039So I think the real magic is everyone knows now it's really good at being adversary if done appropriately.
00:34:13.039 -->
00:34:16.400But it's how do you really test those sensitivities?
00:34:16.400 -->
00:34:26.239That is, you know, I think part of what we're thankfully doing a lot of, which is, you know, we're trying to break tools in the process, figure out what works, and then you know, kind of share that common wisdom throughout the firm.
00:34:26.239 -->
00:34:39.199You know, and so we're really trying to now do sort of breakdowns on what worked in a particular case from soup to nuts, share that information so that everybody can kind of enhance the adversarial review, you know, rather than dealing with in silos.
00:34:39.679 -->
00:34:40.000Okay.
00:34:40.000 -->
00:34:41.360Dan?
00:34:42.719 -->
00:34:50.639Yeah, I mean, the unflinching criticism that AI will provide can be quite useful and insightful.
00:34:50.639 -->
00:35:01.760And as you mentioned, and as Adam has mentioned multiple times to me, you know, and I've experienced this, you get you you litigate a case, you fall in love with your own arguments, and you get a little bit lost in the sauce, right?
00:35:01.760 -->
00:35:14.239And then you walk into court and you feel like the wind got knocked out of you because the judge sees something or opposing counsel says something that you didn't see coming because you fell in love with your own arguments and you got lost in the sauce.
00:35:14.239 -->
00:35:18.719And, you know, AI only knows what you give it.
00:35:18.719 -->
00:35:27.360And so you can give it, you know, the pile of evidence that, you know, relates to your case, you can give it the briefing, and you can say, what am I missing?
00:35:27.360 -->
00:35:36.960And maybe it'll find something, maybe it won't, but that insight can potentially be quite valuable, can at least allow you to brace yourself before you get the wind knocked out of you.
00:35:36.960 -->
00:35:40.320And it's not something that you can easily replicate with your colleague.
00:35:40.320 -->
00:35:56.559You know, if I have a like another partner whose um, you know, insight and acumen I trust, I can't ask them to sit down and read seven hours worth of legal background in order to give me their view, right?
00:35:56.559 -->
00:36:02.719I they'll maybe read a few pages, but they they can't necessarily bill for it, and I can't ask them to spend that kind of time.
00:36:02.719 -->
00:36:03.280Right.
00:36:03.280 -->
00:36:15.760And so AI might not have the same acumen judgment and intuition as one of my partners, but the AI is not going to be bothered by having to read 300 pages and then and then ask um be asked its opinions about it.
00:36:15.760 -->
00:36:22.159So it's a really useful tool that before this type of technology was available, we just didn't have access to.
00:36:22.480 -->
00:36:22.800Right.
00:36:22.800 -->
00:36:23.280Right.
00:36:23.280 -->
00:36:29.039So, Adam, as as we record this, you're getting ready to go to trial.
00:36:29.039 -->
00:36:35.599And and so I guess I wanted to talk a bit more about some of the nuts and bolts around witnesses and experts.
00:36:35.599 -->
00:36:49.760Um, what what can AI how can it help you with additional opportunities like hypotheticals or simulating an opposing expert or testing uh how you cross-examine.
00:36:49.760 -->
00:36:55.679So how are you using it to prepare for cross-examinations or or test an expert witness, say?
00:36:56.480 -->
00:37:05.119I think experts is a right category because we we oftentimes are going to know what they're gonna say because they have to disclose to a certain degree their opinions.
00:37:05.119 -->
00:37:16.800Also, in corporate cases, experts are really often driven, especially either forensic accounting or other financials, by the Excel and the other corporate work papers.
00:37:16.800 -->
00:37:22.719Excels are files that are very easy digestible for an LLM.
00:37:22.719 -->
00:37:40.880So while lawyers who are typically not technically trained and have typically not the appropriate accounting background, using it to build across for an expert is particularly valuable because it's gonna fill in that gap where you weren't classically trained as an accountant.
00:37:40.880 -->
00:37:44.079You're not gonna understand the concepts well enough to even know what to look for.
00:37:44.079 -->
00:37:54.000Like, give me a thousand-line Excel with 25 tabs and ask me to figure out, you know, where is the deferred income being shown?
00:37:54.000 -->
00:37:55.440Like, no chance.
00:37:55.440 -->
00:38:04.000But, you know, help me get along through an LLM to get there, and I can then build the functional knowledge to perform the cross.
00:38:04.000 -->
00:38:13.519So from a baseline perspective, it starts to elevate the lawyer so that they're no longer at a position, I think, of uh a weakness with respect to a trained expert.
00:38:13.519 -->
00:38:14.480Number one.
00:38:14.480 -->
00:38:18.000Number two is hypotheticals are great.
00:38:18.000 -->
00:38:26.480You know, that is where you can oftentimes make your most damage on an expert, but you don't know how they're going to respond to it because you've oftentimes not done it.
00:38:26.480 -->
00:38:30.239And you typically don't want to preview that hypothetical before the cross, anyways.
00:38:30.239 -->
00:38:38.400So running hypotheticals through a large language model is a fantastic way to see what would be the various ways they could answer this.
00:38:38.400 -->
00:38:46.000Give me a 13 different creative ways they could address this issue, and then let me sort of see the end product ahead of time so I know where to go.
00:38:46.000 -->
00:38:48.320Because a really good cross has two things, right?
00:38:48.320 -->
00:38:56.320It's very simple with simple premises that leave you to an inevitable result that is very uncomfortable for the other side.
00:38:56.320 -->
00:39:01.119And so when you can forecast that and then you can do it multiple ways, then it's great.
00:39:01.119 -->
00:39:04.800And I'd say take it to the next level, which is run your cross.
00:39:04.800 -->
00:39:10.639As Dan talked, I mean, talk to type is obviously very easy to do with large language models now.
00:39:10.639 -->
00:39:22.320So, you know, put the materials into the LLM that you are the other expert will rely on, and then start crossing live through the LLM and see how they respond.
00:39:22.320 -->
00:39:29.519Or run parts of your cross through and ask the LLM to give you all the different ways that you could deflect on those issues.
00:39:29.519 -->
00:39:36.400But, you know, sort of previewing how you're gonna address the cross is critically important and getting up to speed is really helpful.
00:39:36.400 -->
00:39:51.760Several of our colleagues actually had a very uh complex arbitration that they recently did, and their ability to turn over crosses from day one to day two during the trial and so forth was really driven by using LLMs because they could really dial in the examination.
00:39:51.760 -->
00:40:02.000And I will say one thing that was oftentimes gone out of out of vogue was we used to get dailies for trials all the time, but then they got really expensive.
00:40:02.000 -->
00:40:11.599And the utility of them was like, great, you get a daily for eight hours of trial testimony, but like no one can actually read it fast enough to then use it for the cross the next day.
00:40:11.599 -->
00:40:13.519And now these act opposites happen.
00:40:13.519 -->
00:40:32.639So using dailies either during a deposition or a cross so that you're building it so that when you get back to your hotel room that night, and then you can ingest all the testimony from that day, build it into the cross for tomorrow, you're at such a more advantage than either the witnesses or the opposing side in that respect.
00:40:32.639 -->
00:40:37.280So it's just being smart about using it and knowing the times that have the most difficulty.
00:40:37.280 -->
00:40:40.159Like trial is complicated, trial is variable.
00:40:40.159 -->
00:40:45.280So anything you can do to simplify your life during trial is super helpful in that respect.
00:40:45.280 -->
00:40:52.079And then really just stress testing it because, you know, oftentimes I think lawyers draft their first cross and that's their only cross.
00:40:52.079 -->
00:40:57.199But if you're doing it well, you should probably be on version eight or nine by the time you get to trial.
00:40:57.199 -->
00:41:02.159But that's either too time consuming or too costly in a normal litigation world.
00:41:02.159 -->
00:41:04.800We're like now we've eliminated a lot of those issues.
00:41:04.800 -->
00:41:08.000Dan, anything to add to that?
00:41:08.719 -->
00:41:15.199Uh yeah, I mean, I think LLMs are very good at adopting a persona.
00:41:15.199 -->
00:41:31.679Um, I mean, that's fundamentally like what a big part of the breakthrough was recently is we know when you talk to Claude or ChatGPT, right, the two leading um AI products, uh Claude is taking on the persona of Claude.
00:41:31.679 -->
00:41:37.519And then the model is using token prediction to say, like, okay, Claude is a helpful assistant.
00:41:37.519 -->
00:41:42.079So as a helpful assistant, I predict that this is what the output should look like.
00:41:42.079 -->
00:41:44.719And it's it's all based around a persona.
00:41:44.719 -->
00:41:57.360And so if you give an LLM a bunch of information to take on a different persona, it's gonna try really, really hard to accurately create the persona that you've assigned it to.
00:41:57.360 -->
00:42:25.039Um so you know, if that's your star witness, if that's an expert, if that's someone that you're going to cross-examine, um, or for you know, our younger associates, uh, we encourage them to do it before they take a deposition for the first time, is try to, you know, plug in a bunch of information about the person you're going to depose and practice having the LLM adopt the persona of the witness and ask questions, see how the witness tries to weasel out of things.
00:42:25.039 -->
00:42:38.159And it's not going to, you know, perfectly predict what the answers are going to be, but it gives you an opportunity to try to ask the questions, ask follow-ups, pin the witness on a certain point.
00:42:38.159 -->
00:42:54.400And even if the dialogue live with the actual witness doesn't go the same way, the fact that you had an opportunity to work with it is going to, I think, make you more comfortable with it, make you uh less surprised by the answers that you get.
00:42:54.400 -->
00:43:04.800And I think it it I think Adam talks about putting in reps, you know, before you actually do the live performance, you know, cross-examining a witness at trial or what have you.
00:43:04.800 -->
00:43:12.639And prior to having the LLMs, again, the best you could do was just ask a colleague, like, hey, can you mock this with me?
00:43:12.639 -->
00:43:15.119And lawyers are not particularly good actors.
00:43:15.119 -->
00:43:25.519And asking a colleague of yours to spend several hours reading a bunch of background material so that he or she can effectively adopt this persona is a pretty heavy lift.
00:43:25.519 -->
00:43:27.440It's expensive and it's a big ask.
00:43:27.440 -->
00:43:36.559Uh, but the LLMs can read all of that material in seconds and will gladly play that role for you so that you can try and get some of those reps in before the big day.
00:43:37.039 -->
00:43:46.480So you can yeah, so so you can kind of create uh I don't know if an avatar is the right word, but uh the persona, I guess is what you're saying is of of a witness.
00:43:46.480 -->
00:43:52.480You could plug in things that they've said before, maybe things they've written, and then you can say things about their attitude.
00:43:53.920 -->
00:43:54.159Yep.
00:43:54.159 -->
00:43:59.440And and I mean, and you can you can direct the AI system to behave in different sorts of ways.
00:43:59.440 -->
00:44:03.199So You know, uh have a terrible attitude and give up nothing.
00:44:03.199 -->
00:44:05.840Or be professional but circumspect.
00:44:05.840 -->
00:44:17.039And again, it's not going to perfectly predict what the actual witness would do, but it's going to give you the opportunity to play with something that is a simulacrum of what you're about to do.
00:44:17.039 -->
00:44:19.119Uh so it will make you more What?
00:44:19.440 -->
00:44:20.239What was that word?
00:44:20.239 -->
00:44:21.360Jesus Christ.
00:44:21.360 -->
00:44:22.000What is it?
00:44:22.000 -->
00:44:22.559Simulacrum?
00:44:22.559 -->
00:44:24.719What is simulacrum?
00:44:25.039 -->
00:44:27.440I've played too many RPGs, I think.
00:44:28.719 -->
00:44:29.679Simulacrum.
00:44:29.679 -->
00:44:30.880Okay, I like words.
00:44:30.880 -->
00:44:31.760I'm writing that one down.
00:44:31.760 -->
00:44:32.719Simulacrum.
00:44:32.719 -->
00:44:34.159Gotta see what that means.
00:44:34.159 -->
00:44:36.719Um so uh no, that's good.
00:44:36.719 -->
00:44:39.280But it just you also used the word weasel earlier.
00:44:39.280 -->
00:44:42.159So what you might say, is this person gonna be a weasel?
00:44:42.159 -->
00:44:43.280This person's a weasel.
00:44:43.280 -->
00:44:44.480Act like a weasel.
00:44:44.639 -->
00:44:49.599Uh listen, I've taken enough people's depositions, and there are plenty of weasels in the world.
00:44:49.599 -->
00:44:55.519And listen, if I ever get deposed, I'm gonna be as weasely as I possibly can within, you know, within my ethical boundary.
00:44:55.920 -->
00:44:56.079Yeah.
00:44:56.079 -->
00:44:56.960So yeah, okay.
00:44:56.960 -->
00:44:59.119Well, so I guess that would that flips it though.
00:44:59.119 -->
00:45:02.159That also would help a witness prepare, yeah?
00:45:02.159 -->
00:45:04.480Or help you prepare a witness.
00:45:05.280 -->
00:45:05.760Absolutely.
00:45:05.760 -->
00:45:09.679I mean, it it it it it goes it goes both ways, right?
00:45:09.679 -->
00:45:20.559So, you know, you can dump in the materials related to the deposition you're about to defend and say, like, you know, what are the most uncomfortable questions that opposing counsel is gonna press my witness on?
00:45:20.559 -->
00:45:35.440And again, I can think of as many things as I like, but the AI is gonna provide a fresh perspective, you know, because again, I've gotten maybe too close to my materials, I've fallen in love with my own arguments, I believe in the veracity and character of my honorable witness.
00:45:35.440 -->
00:45:45.679And uh an opposing counsel is gonna see the world a very different way, and the AI system can help you see the world through the lens of your opponent.
00:45:47.119 -->
00:45:47.440Okay.
00:45:47.440 -->
00:45:52.079The um and so c kind of moving on to another type of persona.
00:45:52.079 -->
00:45:56.000So you we started talking about mock, let's talk about mock juries.
00:45:56.000 -->
00:46:12.159Um so I'm I'm wondering how how you can use this uh in a use AI to to create a mock jury, because that's those are those gotta be fun, um, but they're also probably very expensive to do and a lot of logistics too.
00:46:12.159 -->
00:46:19.039But if you're using AI to do a quick kind of a mock jury, do you um set up uh personas there?
00:46:19.039 -->
00:46:21.199So I've got, I don't know, say six jurors.
00:46:21.199 -->
00:46:26.000This one's uh this one's a welder, this one's a musician, this one's this one's a weasel.
00:46:26.000 -->
00:46:26.559I don't know.
00:46:26.559 -->
00:46:27.360Do you do that?
00:46:27.360 -->
00:46:29.760Tell me about mock juries in AI, I guess.
00:46:30.079 -->
00:46:48.639Yeah, I'm glad you asked, because you know, this is an area that in what I kind of call like version one, which was you know late 24 to 25, was like really my focus because there's so much that you gain from a jury experience, but only when you have those experiences.
00:46:48.639 -->
00:46:53.199And they're fleeting even for those of us that try more cases than others.
00:46:53.199 -->
00:46:58.000And so it seems so obvious to me that there has to be a tool for these types of things.
00:46:58.000 -->
00:47:04.320But getting there was tough because I think it was sort of the end product in the sort of the litigation cycle.
00:47:04.320 -->
00:47:06.719But there are tools out there now.
00:47:06.719 -->
00:47:10.000We're currently piling one that we really like.
00:47:10.000 -->
00:47:12.239And it is doing exactly what you're talking about.
00:47:12.239 -->
00:47:20.639It, you know, they have the capacity to identify demographics in where you're gonna try the case, which is a, you know, the first start of it, right?
00:47:20.639 -->
00:47:25.760Because every demographic across the across the country is gonna have a different type of persona.
00:47:25.760 -->
00:47:36.719And then within those demographics, you're gonna have a complete mixture of life experiences, you know, all over the educational spectrum, life experiences, all those things.
00:47:36.719 -->
00:47:44.800So the way that they're able to create effectively these jurbots at this point is they're creating their own personas for them.
00:47:44.800 -->
00:47:52.559They're building life histories for them that would be replicative of the actual um environment where you're gonna try the case.
00:47:52.559 -->
00:47:55.840And then they're very good at a couple things.
00:47:55.840 -->
00:48:01.679One, typically the way you do these now is you do a video um mock presentation.
00:48:01.679 -->
00:48:06.719So you and somebody else in your you know firm, you'll represent one side, they'll represent the other side.
00:48:06.719 -->
00:48:15.679We record them, and we do introduce several you know, pieces of salient information, key facts, and then key evidence.
00:48:15.679 -->
00:48:18.159And then we give them a verdict form as well.
00:48:18.159 -->
00:48:23.599Uh the difference is when we used to do historical mocks, we had one day with the jurors.
00:48:23.599 -->
00:48:33.519We'd have you know three panels, present the evidence and have them all deliberate separately, it all gets recorded, and then whatever we get from it, we get from it, and that's it.
00:48:33.519 -->
00:48:34.960We can't go back.
00:48:34.960 -->
00:48:37.440The difference, though, is two things.
00:48:37.440 -->
00:48:52.480One, you know, in those situations, if we have a couple difficult jurors in that mock examination where it's a human version, they may drive the entire deliberation, maybe good, or may just completely take it off the rails because they know it's not real.
00:48:52.480 -->
00:48:58.320The flip side with a simulated jury is that one, they don't know it's not real.
00:48:58.320 -->
00:49:08.400Number two, what's really interesting is when they're deliberating, they're not just doing it once, they're running the simulation 10,000 times.
00:49:08.400 -->
00:49:10.800So you're getting different variations.
00:49:10.800 -->
00:49:23.760And the coolest part about it, I think, is now, let's assume, for example, you did one in the first quarter of the case to get a general vibe on where you are based off what you know.
00:49:23.760 -->
00:49:33.920You can then replant those same mock jurors, wipe their memory if you wish, and then present a more refined version in the second quarter.
00:49:33.920 -->
00:49:36.159Do one in the third quarter as well.
00:49:36.159 -->
00:49:39.280And you can also ask them questions, you can query them afterwards.
00:49:39.280 -->
00:49:41.039If you would have known this, would it change this?
00:49:41.039 -->
00:49:45.440And they may say no, it wouldn't, and they'll explain why from common sense purposes.
00:49:45.440 -->
00:49:50.480So it's a going to give you a lot greater information set.
00:49:50.480 -->
00:49:54.079It more tracks the reality of the case, which you have multiple times.
00:49:54.079 -->
00:49:59.440And oftentimes in a real mock, they're so expensive that you're doing one at best.
00:49:59.440 -->
00:50:10.079And you're typically doing it in the last third of the case, if not towards the end, which is great for a trial prep, but anything you missed when discovery is closed, like that door is shut.
00:50:10.079 -->
00:50:14.320So the idea of doing them throughout the case is very, very appealing.
00:50:14.320 -->
00:50:17.840The other reality, too, is you know, back to the reps concept.
00:50:17.840 -->
00:50:22.960Part of why trials don't go well is because you didn't prep ahead of time.
00:50:22.960 -->
00:50:35.440So forcing the lawyers to run a mock within the first quarter of the case, run another mock halfway through the case, we're already getting better because we're hearing the words come out and some are landing, some aren't.
00:50:35.440 -->
00:50:41.199But by the time under this structure, like you've already done the dance four times before you get to the real thing.
00:50:41.199 -->
00:50:45.599And that's a huge competitive advantage when you think about it appropriately.
00:50:45.599 -->
00:50:49.519And the one thing obviously we've left out is the cost differential is crazy.
00:50:49.519 -->
00:50:54.960I mean, it's like one-tenth of the cost to do it because it's really a cost of compute component of it.
00:50:54.960 -->
00:50:59.119So, you know, the the pilot we're looking at now, we've been very pleased with.
00:50:59.119 -->
00:51:06.719Um, we've seen some really good outcomes comparing it to real life experience and seeing some of the same things that we'd see in a regular jury, how they'd rule.
00:51:06.719 -->
00:51:17.679And I think that area of law is letting it way better because predicting trials is by far the most difficult thing because we have a lot of humans who we have no control over in the process.
00:51:17.679 -->
00:51:20.159And we have a very, very compressed time period.
00:51:20.159 -->
00:51:34.159So, from my perspective, when we sort of look at like where things are going, the more we can dial in predictive outcomes for juries, we can be so much better at deciding which case you should go, which case you shouldn't, and actually give the client, I think, much more grounded information.
00:51:34.159 -->
00:51:39.760So that's an area that's still very evolving, but is only going to get much better with time.
00:51:40.639 -->
00:51:48.400Yeah, the the part you talked about having um being able to use mock juries during the actual trial, that's fascinating.
00:51:48.400 -->
00:52:00.000Um, because you can give it, as you said, you can give it new information or something maybe that just happened during the during the trial, or give them fresh and you can eat, like you said, you can wipe their memory clean or or start over.
00:52:00.000 -->
00:52:02.079I think that's that is wild.
00:52:02.079 -->
00:52:06.480Um Yeah, and the cost, I I've never I don't even know what mock juries cost.
00:52:06.480 -->
00:52:17.760I just know from doing um um market research, just getting focus groups together, you you can easily run into tens of thousands of dollars uh or more.
00:52:17.760 -->
00:52:20.480Um so I'm assuming it's it's similar to that.
00:52:20.480 -->
00:52:25.840But being able to do it over and over and over again and then adjust it, I think that's that's wild.
00:52:26.800 -->
00:52:27.679Yeah, it's very exciting.
00:52:27.679 -->
00:52:37.679And I think I mean to your point, I'm mock jury is a traditional way, like you're in the six figures, just on cost before you tack on the lawyer time to get ready for it, all those things.
00:52:37.679 -->
00:52:41.519Um so you know, they've gotten wildly expensive.
00:52:41.519 -->
00:52:50.880And some cases warrant it, but every case, you should know to some degree what is likely to happen before you go to the real thing.
00:52:50.880 -->
00:52:58.880And the idea that, oh, just because the case isn't big enough, we can't tell you what's likely to happen with any level of certainty, I think is a real detriment to a legal system.
00:52:58.880 -->
00:53:11.280So the idea that cases of any size now, we can start to give you some level of objective predictability on top of our lawyer judgment is just making the whole process better for everybody.
00:53:11.280 -->
00:53:14.480Dan, anything to add to that?
00:53:16.639 -->
00:53:17.760And you don't have to.
00:53:18.079 -->
00:53:21.440No, I mean, uh I guess two two two small things.
00:53:21.440 -->
00:53:21.760Yeah.
00:53:21.760 -->
00:53:24.800One is with with with the reps piece, right?
00:53:24.800 -->
00:53:29.199There's maybe a nearly infinite way to do an opening statement.
00:53:29.199 -->
00:53:30.800You only get one shot.
00:53:30.800 -->
00:53:40.880The idea that you can try 15 different ways before a mock jury and then put together what's actually most effective based on the feedback you're getting is remarkable.
00:53:40.880 -->
00:53:43.119And I think it changes the game quite a bit.
00:53:43.119 -->
00:53:57.199And then so much of what a trial is, is taking a huge body of information and trying to tell a very clear story by emphasizing just small bits of that information.
00:53:57.199 -->
00:54:00.480And a lot of a trial is pruning and curation.
00:54:00.480 -->
00:54:02.960And again, you only get one shot at it.
00:54:02.960 -->
00:54:15.440And if you can try and do it in different ways, you know, putting emphasis here, putting emphasis there, maybe we have this attorney depose this witness and this other attorney depose that witness because we think the optics are better.
00:54:15.440 -->
00:54:23.920Being able to test those things and get feedback before you actually go to the big show is uh, yeah, completely game-changing.
00:54:23.920 -->
00:54:31.119And uh I think is going to increase the quality of trials overall.
00:54:31.119 -->
00:54:37.440I think it'll make people more comfortable with trial, and I hope that it will increase the chances that more cases go to trial.
00:54:37.440 -->
00:54:43.280Because I think Adam will agree that trials are so risky and expensive that they just don't happen very often.
00:54:43.280 -->
00:54:48.000I think it's less than 2% of uh civil cases in the US go to trial.
00:54:48.000 -->
00:54:59.119And a lot of times it's a real shame because you have a winner on your hands and it's a matter of persuading the client that the that the juice is worth the squeeze, that the risks are worth it.
00:54:59.119 -->
00:55:09.199And having the data associated with mock trial feedback can start to instill that confidence so that people feel that they should have their day in court for real.
00:55:09.199 -->
00:55:22.239And that if something, you know, something bad happened and they're looking for compensation, that they should be uh, you know, they should have the courage to sort of step in and actually fight it out in court, which just doesn't happen often enough from my perspective.
00:55:22.880 -->
00:55:23.199Okay.
00:55:23.199 -->
00:55:24.800All right.
00:55:24.800 -->
00:55:27.039Um see, so moving on.
00:55:27.039 -->
00:55:31.440Uh so Adam, I think we you I think I think you used this.
00:55:31.440 -->
00:55:41.280Maybe it was something you said or something you wrote, but talking about um using um having a litigation command center and uh and a client dashboard.
00:55:41.280 -->
00:55:45.280So can you can you kind of elaborate on what that's about?
00:55:45.280 -->
00:55:54.639Um and uh and what would genuinely uh what would a genuinely useful litigation command center uh show to lawyers and to the clients?
00:55:55.599 -->
00:56:07.760Yeah, I'm glad you raised it because I I think we've if you talk about the other sort of major advancement that we see coming is uh the communication with the client and communicating differently.
00:56:07.760 -->
00:56:16.719So the tool we've created contemplates this, and we've seen other AI tools that do something similar to this.
00:56:16.719 -->
00:56:28.719But the idea is that the client is gonna have greater supervision of how lawyers practice going forward, which I think is a positive for two reasons.
00:56:28.719 -->
00:56:34.159One, I think that at the end of the day, the client is entitled to know how we're proceeding.
00:56:34.159 -->
00:56:45.679Because, you know, even though we're the ones that have the experts in litigation and they're the outside observer in certain respects, they're ultimately the ones that are the client paying for the services.
00:56:45.679 -->
00:56:54.079So it's vital that they know what's going on in the case and can also supervise at a greater degree along the way.
00:56:54.079 -->
00:56:57.280So the accountability piece is really helpful there.
00:56:57.280 -->
00:57:11.119And for example, the way we've envisioned it is that we've created ways that basically the case can be pulsed in whatever time period the client wants, call it twice a month, weekly, depending upon the status of the case.
00:57:11.119 -->
00:57:20.000That pulse then tells the client, here's where we are in the case, here's where we feel strong on, here's the facts that we have locked in, here's where we're headed next.
00:57:20.000 -->
00:57:33.119But rather than just telling the client, you know, all is well, what we've envisioned, what we've built is a way for the client then to basically check the quality of our analysis and to make sure that there is support behind it.
00:57:33.119 -->
00:57:35.360So, number one, it now gives the client to do.
00:57:35.360 -->
00:57:45.519And also, you know, all clients, right, they see a lawyer call with two or three lawyers and they're just trying to get through the information, and all they see is just, you know, the multiplying dollars.
00:57:45.519 -->
00:58:03.519So the idea that the client, whether they're an in-house legal individual or somebody that's a non-legal, you know, client rep, the idea that they can, you know, likewise interrogate our analysis is super valuable and on their time, not on ours, so that they actually have the time to sit with it, come up with their questions, things like that.
00:58:03.519 -->
00:58:05.840So I see it as hugely valuable there.
00:58:05.840 -->
00:58:12.159And the last thing is, you know, all clients want to see a cost benefit from AI and totally fair.
00:58:12.159 -->
00:58:16.880Like we can't be championing all these efficiencies and then have the cost go up.
00:58:16.880 -->
00:58:19.199Like there's just a massive disconnect there.
00:58:19.199 -->
00:58:24.239But one of the ways we do that is by working in parallel with the clients.
00:58:24.239 -->
00:58:30.639So when I talk about these command centers, it's helping the client know where we are in the case and where the focus is.
00:58:30.639 -->
00:58:43.280And for example, that client who knows their industry better than we will ever know it can look at our work product and look at our pace on the case and work with us to identify gaps, things like that, and then also participate.
00:58:43.280 -->
00:58:56.400So, you know, part of it is especially with sophisticated in-house counsel, you know, having them play a more vital role in the process not only helps make it more efficient, but they can also sort of zero in on some of the core issues and things like that.
00:58:56.400 -->
00:59:08.320So that ability to share freely information and allow them to really have the access into our thinking, I think is a real value point because that's just gonna spawn additional thoughts, processes.
00:59:08.320 -->
00:59:15.039And then they may be able to get from point A to point B faster because of their historical knowledge of the industry, of the client, all those things.
00:59:15.039 -->
00:59:19.360And so that's really the hope is that with that added level of communication, it's helpful.
00:59:19.360 -->
00:59:22.000And also, look, sometimes the client just doesn't want to talk to us.
00:59:22.000 -->
00:59:23.519They just need an answer fast.
00:59:23.519 -->
00:59:25.840Maybe the CFO just called them.
00:59:25.840 -->
00:59:32.000And, you know, every tension is even when you're trying to be respectful of the client, like we can't be everywhere all the time.
00:59:32.000 -->
00:59:36.960So the idea that that CFO, if they just need to know, because the CEO is like, where are we on this case?
00:59:36.960 -->
00:59:38.079What's going on?
00:59:38.079 -->
00:59:39.360You know, how far are we?
00:59:39.360 -->
00:59:46.480The idea that the CFO can just pulse the reports we've given to them and say that immediately, you know, makes their lives easier as well.
00:59:46.480 -->
00:59:49.119Because we all know that business doesn't just like wait for the law.
00:59:49.119 -->
00:59:51.440You know, ultimately they want to make a decision fast.
00:59:51.440 -->
00:59:56.880The CEO is like, I have two minutes to discuss this case, and the in-nounce council is you know, scrambling for it.
00:59:56.880 -->
01:00:00.320You know, now we're putting in a position that they have the information they need as well.
01:00:00.320 -->
01:00:04.000So it's really sort of addressing that pain point, I think, is the last point there.
01:00:04.639 -->
01:00:09.199I think Adam, I think I think you're one of these guys, you could be funny without even trying.
01:00:09.199 -->
01:00:12.400Um clients, clients just don't want to talk to us.
01:00:15.599 -->
01:00:17.039Dan, was that funny?
01:00:17.039 -->
01:00:17.440I don't know.
01:00:17.440 -->
01:00:18.000It was funny.
01:00:18.000 -->
01:00:25.840I mean, I it it matches my experience that sometimes sometimes I feel like all they want to do is talk to me.
01:00:25.840 -->
01:00:31.920And and and other times it's like they like they they hear the meter running as I'm speaking.
01:00:33.840 -->
01:00:34.800Okay, good enough.
01:00:34.800 -->
01:00:39.840The remaining questions I had, we've touched on quite a bit, so I don't want to belabor them.
01:00:39.840 -->
01:00:47.199Uh we're talking about the model changing economics of litigation and the work of the teams and and then what associates can do.
01:00:47.199 -->
01:00:51.599Is there anything additional say about any of that before we uh go on to my final questions?
01:00:52.079 -->
01:00:59.599One thing that I think is happening, right, is there's constant downward pressure on fees, and then and then you know, law firms are always putting upward pressure on fees.
01:00:59.599 -->
01:01:09.280Um but it it can put law firms in a difficult position if AI is making you more efficient and you're paying for the tool.
01:01:09.280 -->
01:01:22.239Now you have paid money to bill fewer hours on the same matter, which is a pretty bad double whammy uh, you know, if you don't get a handle on it.
01:01:22.239 -->
01:01:35.519Um and I think uh as you know everyone is trying to get a handle on how to manage this technology, use it, and deploy it, uh there's going to be a lot of pressure for efficiency in in particular.
01:01:35.519 -->
01:01:52.079And that's why I think um alternative fee arrangements are, I mean, maybe, maybe finally AI will be, maybe not the death of the billable hour, but hobbling the billable hour, making taking away the the dominance of the billable hour.
01:01:52.079 -->
01:02:04.320Uh because what alternative fee arrangements offer for clients is predictability and steady pricing so that they know what the spend is going to be.
01:02:04.320 -->
01:02:11.199Litigation is particularly unpredictable, and the pricing can vary wildly depending on how a piece of litigation goes.
01:02:11.199 -->
01:02:22.480And so law firms have an opportunity to say, I will take the risk of though that variability because I believe in the efficiency and productivity that I can get using this technology.
01:02:22.480 -->
01:02:26.559So it puts the onus on the firms to figure that out, but I think that that's a good thing.
01:02:26.960 -->
01:02:35.920Yeah, the hourly um I've never I'm not a lawyer, haven't, but the times when I've had to chart uh had to track my hours, I just found it maddening.
01:02:35.920 -->
01:02:49.519Um and I also always think too about you know, you get a a lawyer with a hundred years experience who can whatever resolve an issue in five minutes or uh I don't know, get somebody out of the soup in five minutes.
01:02:49.519 -->
01:02:52.960It's just like, what do you charge five minutes for five minutes?
01:02:52.960 -->
01:02:56.639I I So there's got to be something between the value, I think, you know?
01:02:56.960 -->
01:02:57.119Yeah.
01:02:57.679 -->
01:03:15.360I mean, there's there is something a little bit the incentives seem wrong that you have the lawyers have an incentive to just grind out the hours and have their associates, you know, billing overnight every night 24-7, whereas like a lot of the value comes in like a 10-minute conversation.
01:03:15.360 -->
01:03:23.360And um, and so I think that moving away from the billable hour model and AI facilitating that will be good for the profession.
01:03:23.360 -->
01:03:24.639At least I hope so.
01:03:24.960 -->
01:03:25.199Yeah.
01:03:25.199 -->
01:03:27.039Adam, anything to add to that?
01:03:27.440 -->
01:03:47.679You know, your point about the lawyer with 45 years of experience who has seen so many paths go wrong, that they lead you down the right path and maybe even get to the point of, you know, telling you to do something hard that hurts, but you're gonna save hundreds, if not millions, of dollars of doing it.
01:03:47.679 -->
01:03:56.400I think what's interesting when we sort of look at the future of that category is how do firms capture that wisdom in the process?
01:03:56.400 -->
01:04:09.679So, you know, I think when we had this conversation a year, Tom, I think what we're gonna be talking about next is capturing the institutional knowledge of law firms in a way that then we have some real value.
01:04:09.679 -->
01:04:22.880Because from my perspective, you know, we have litigators that have 30, 40 years of experience, you know, doubling my years of experience, that will have more wisdom than I can have at this point, no matter what.
01:04:22.880 -->
01:04:36.960And so I think that's where you see firms going is collecting that knowledge, figuring out how to collect it in a way that is usable, both from work product and from the what's in, you know, in between your ears.
01:04:36.960 -->
01:04:51.679But when we start to figure out how to harness that and do post mortems on cases in a more thoughtful way so that we're building on every both success and failure, that's really when we have something new to offer to the client.
01:04:51.679 -->
01:04:55.039And I don't think we're as far off as you know, some may think from that.
01:04:55.679 -->
01:05:01.280Confidentiality, privacy, uh, accuracy, you know, professional responsibility for the client.
01:05:01.280 -->
01:05:05.039Are there safeguards that you're putting in or you would recommend?
01:05:05.039 -->
01:05:07.360I'll let Dan go ahead first.
01:05:07.760 -->
01:05:14.880Yeah, I mean, don't be putting confidential information into public AI systems that it's just a terrible idea.
01:05:14.880 -->
01:05:26.079Even if you click that button that says don't train the model on it, um, it still is uh it it's it's not enterprise uh grade security, right?
01:05:26.079 -->
01:05:33.519So you can get enterprise grade security through substantially all AI vendors now, and it's just it's just a matter of doing it.
01:05:33.519 -->
01:05:43.199So you can have ChatGPT with enterprise grade security, you can have Claude with enterprise grade security, and then the host of legal specific products, Harvey, Lagora, et cetera, et cetera.
01:05:43.199 -->
01:05:44.880All of them offer that.
01:05:44.880 -->
01:05:59.039Um and once once you're there, as long as you're complying and your vendor is compliant with appropriate uh the applicable laws and regulations, then I I I don't think we're in very much of a different world from any other technology that we're using.
01:05:59.039 -->
01:06:04.880You know, I have A lot of people say, like, well, is it okay for me to put this into an the AI system?
01:06:04.880 -->
01:06:07.280And I ask, Are would you send it in an email?
01:06:07.280 -->
01:06:09.119And they're like, Well, yeah, I received it in an email.
01:06:09.119 -->
01:06:11.199And I was like, All right, so what's the difference?
01:06:11.199 -->
01:06:19.519Microsoft has access to that in the same way that you know our friends at Harvey, Lagora, OpenAI, or Anthropic will have access to it.
01:06:19.519 -->
01:06:24.079And they're governed by the same contractual obligations, the same laws, et cetera, et cetera.
01:06:24.079 -->
01:06:41.760So it's really just about getting your ducks in a row with um with you know the right vendor relationships and the right terms in your master services agreements to have things like zero data retention and the appropriate information security uh you know certifications and systems in place.
01:06:41.760 -->
01:06:54.320But once once you have those in place, I think it you know you should have the same degree of comfort using those systems as you do, you know, saving a document into your firm's you know document management system.
01:06:54.719 -->
01:06:56.719Or or same level of discomfort.
01:06:56.719 -->
01:06:57.280Yeah.
01:06:57.280 -->
01:06:58.159Or yeah.
01:06:58.159 -->
01:06:59.599As a case may be.
01:06:59.599 -->
01:07:07.679You know, I was in I was chatting with co-pilot yesterday, and I was asking what I thought were some pretty clever questions based on my many years of experience.
01:07:07.679 -->
01:07:16.239I said, you know, you're not using this, you're not gonna go out later, and some guy like me is uh gonna ask you these questions and you're gonna use what you learned from me, are you?
01:07:16.239 -->
01:07:19.199And it copilot promised me it was not going to do that.
01:07:19.360 -->
01:07:22.480So Well, and the AI would never lie to you, right?
01:07:22.480 -->
01:07:24.079Why would Copilot lie to you?
01:07:25.599 -->
01:07:35.840I we call her a she, my wife and I both use her because we work together sometimes um and we share you know personal things we're doing, like you know, having a granddaughter or whatever.
01:07:35.840 -->
01:07:40.639Sometimes now Copilot, when she talks to it, it makes fun of me.
01:07:40.639 -->
01:07:45.199So I don't know I don't know if that's healthy.
01:07:45.199 -->
01:07:47.360Anyway.
01:07:48.159 -->
01:07:50.880Yeah, uh who knows what your wife is saying to the system.
01:07:50.880 -->
01:07:52.800That's uh you that kind of output.
01:07:52.800 -->
01:08:01.119Yeah, I'm getting a hint of what it says because she tells me the perils of giving these systems uh permanent memory.
01:08:01.840 -->
01:08:06.239We know Tom got up late today and didn't have his coffee yet, so don't ask him this yet.
01:08:06.239 -->
01:08:08.000I mean it's stuck literally like that.
01:08:08.000 -->
01:08:09.360Um I don't know.
01:08:09.360 -->
01:08:14.400Adam, did you want to add anything to that around confidentiality or professional responsibility and use?
01:08:14.880 -->
01:08:17.760Yeah, I think there's two points that are worth mentioning.
01:08:17.760 -->
01:08:22.159Number one is we have to also be wary of the client.
01:08:22.159 -->
01:08:35.279You know, part of our jobs as that trusted advisor, even a pre-AI world, was helping them resist their temptations to do things during the litigation that actually makes their situation worse, even if they think they're helping.
01:08:35.279 -->
01:08:43.279That's always been the case in litigation, because they always want to do something that you're like, well-intentioned, wrong execution.
01:08:43.279 -->
01:08:56.720However, now that they also have access to at least a street-level legal degree through AI, even sophisticated clients are ripe to do things that are outside a protected environment.
01:08:56.720 -->
01:09:07.439And, you know, in particular, because obviously some of the early cases about, you know, client-driven AI legal research have come up as not protected, as a baseline, there's a real concern there.
01:09:07.439 -->
01:09:20.640So talking with clients early on about understanding what is appropriate AI use versus not is significantly important because for whatever reason, you know, we all seem to tell a lot of secrets to AI.
01:09:20.640 -->
01:09:24.960And we seem to have a high level of confidence in it and some more than others.
01:09:24.960 -->
01:09:32.880So, you know, trusted environments are dangerous when we're in the middle of a litigation and we're trying to be very thoughtful about what we do.
01:09:32.880 -->
01:09:34.239So that's point one.
01:09:34.239 -->
01:09:44.000I think when you talk about the professional responsibility piece, I have somewhat of a different take, I think, than perhaps sort of where other folks have gone.
01:09:44.000 -->
01:09:50.399Everyone, you know, has real concerns about using AI and all the, you know, risks of loss prevention and malpractice.
01:09:50.399 -->
01:10:00.880I I often see it at the flip side, which is I think if you're not using AI thoughtfully, you're missing out on a bunch of things to make sure your work is done well, thoughtful.
01:10:00.880 -->
01:10:12.319And also sometimes, even if you're in a spot where you're trying to assess, you know, my instinct is telling me one thing about how to handle a thorny issue, but my instinct is driven by my own subjectivity.
01:10:12.319 -->
01:10:15.039Like I need a third resource here.
01:10:15.039 -->
01:10:24.640I actually think that AI in the appropriate system is going to give you some pretty harsh feedback about when you're making a really tough judgment call on how to act.
01:10:24.640 -->
01:10:29.680Because we have external pressures, we have client demands, we have financial pressures, all those things.
01:10:29.680 -->
01:10:43.199And sometimes asking the right question to something that is not financially and you know pressure driven, you'd be surprised that sometimes how it would make you rethink your thinking on potentially, you know, the challenging issues.
01:10:43.199 -->
01:10:48.000And litigation by definition is about making choices and dealing with consequences.
01:10:48.000 -->
01:11:02.000So thinking through thought, thought processes, using AI to help out with professional responsibility as well as other pieces, I think only makes you more informed in the process than just saying I relied on my judgment.
01:11:02.000 -->
01:11:05.279I thought it was a perfect thing at the time when, you know, your judgment is fallible.
01:11:05.279 -->
01:11:06.159We all are.
01:11:06.159 -->
01:11:13.520And, you know, that's why we need help in the process, especially when we're making very, very important decisions that we know are going to be scrutinized.
01:11:14.079 -->
01:11:14.560Right.
01:11:14.560 -->
01:11:15.119Yeah.
01:11:15.119 -->
01:11:20.399I mean, you wouldn't turn in a document without spell checking it or checking citations, right?
01:11:20.399 -->
01:11:23.199So it's uh it's getting in that direction.
01:11:23.199 -->
01:11:27.199So we've talked a lot about the the the safeguards and things like this.
01:11:27.199 -->
01:11:44.479Um it it just seems to me, you know, managing attorneys now is is gonna be uh different and training attorneys is gonna be uh very different because these these tools are just so when used right, they're just so powerful.
01:11:44.479 -->
01:11:47.119Um when used wrong, they're just just not.
01:11:47.119 -->
01:11:49.279Um they're the opposite.
01:11:49.279 -->
01:12:00.800Just tell people who are listening to this that they want to embrace what you all are talking about, using AI as as a as a tool for trial and and and all of this.
01:12:00.800 -->
01:12:10.479So what would you each of you say um there's some key things that firms should just should do right away if they want to move into this area, besides talk to you and listen to this.
01:12:10.479 -->
01:12:11.920What should they do next?
01:12:11.920 -->
01:12:13.439Dan, what you want to go first?
01:12:13.439 -->
01:12:14.479Go ahead, Dan.
01:12:15.119 -->
01:12:15.520Sure.
01:12:15.520 -->
01:12:28.960Um I mean, I think most of what I learned about how to use AI was just by throwing myself at it and being able to experiment, being able to play.
01:12:28.960 -->
01:12:32.880And and I I I I found a lot of enjoyment out of it.
01:12:32.880 -->
01:12:37.600Uh, you know, I think a lot of lawyering is so monastic typically.
01:12:37.600 -->
01:12:41.439It's just you and some documents and a computer screen.
01:12:41.439 -->
01:12:55.600And I've in AI, I found not a person, because if you anthropomorphize it, I don't think you'll you'll you'll get the output that you're looking for, but it is an intelligence and you can treat it like a colleague.
01:12:55.600 -->
01:13:07.119And I've always been what my sister describes as an external thinker, meaning I like to say it out loud, I like to get people's reactions, I like to talk about what I'm thinking about.
01:13:07.119 -->
01:13:14.319And AI sort of allows me to do that without draining someone else's resources and wasting their time.
01:13:14.319 -->
01:13:23.039And um, and it turns out that that sort of interaction pattern, I think, teaches you how to use AI more effectively.
01:13:23.039 -->
01:13:41.359You start to develop intuitions about what's going to work, what's not going to work, about how to get high quality output, even in an extraordinarily complex situation, by staging out the work and doing it, um doing the workflow in a way that is going to maximize the tool's capabilities.
01:13:41.359 -->
01:13:53.760And the only way that you can really develop that intuition and really understand how to make the technology do what you want is with time spent using the technology.
01:13:53.760 -->
01:14:01.359Um, like our director of applied AI, Rich Robbins, is fond of saying that AI is not a spectator sport.
01:14:01.359 -->
01:14:10.880And what he means by that is that unless you try to make the technology do things for you, you're never going to learn how to make the technology do things for you.
01:14:10.880 -->
01:14:17.119So I think firms should be acquiring tools, giving people access to basic training.
01:14:17.119 -->
01:14:24.960But if they really want AI use to become sophisticated and effective, they have to give people room to experiment with the technology.
01:14:24.960 -->
01:14:36.239So I think you know, having uh you know, hours of experimentation credited towards a billable hour requirement is a great way to start, for example.
01:14:36.239 -->
01:14:43.439You're putting your money where your mouth is saying, I do want you to experiment and I'm willing to pay you full freight to spend that time.
01:14:43.439 -->
01:14:54.399And I think that those types of investments will yield significant dividends as you develop sort of an army of power users with this technology.
01:14:54.399 -->
01:14:59.600Uh, because I think the technology just provides almost like a superpower.
01:14:59.600 -->
01:15:13.520I think uh a partner at a DLA that I was uh uh watching on a panel last week said that uh it's like steroids for lawyers, except now everyone is allowed to take the steroids and encouraged to take the steroids.
01:15:13.520 -->
01:15:23.039And so if if you're able to harness the technology effectively, then it's going to make you as a practitioner a lot more effective.
01:15:23.039 -->
01:15:32.720And if you multiply that across an entire firm's worth of attorneys, then I think that you have a competitive advantage that will be difficult to match without a similar investment.
01:15:34.399 -->
01:15:36.159Adam, what about you?
01:15:37.119 -->
01:15:40.640You know, the change management is super challenging, right?
01:15:40.640 -->
01:15:43.520So I think you got to start from the bottom up.
01:15:43.520 -->
01:15:49.119One of the things we did very strategically at the end of last year this year was really identified champions.
01:15:49.119 -->
01:15:52.960The one nice thing about these tools is we can see who's using them.
01:15:52.960 -->
01:15:54.960Um we get those output reports.
01:15:54.960 -->
01:15:58.720And so we've definitely identified champions and power users.
01:15:58.720 -->
01:16:11.760And, you know, you got to put resources where those power users are because they're the most forward-thinking and adaptive, and then you know, try to get the information out of them to then disperse to the masses is number one.
01:16:11.760 -->
01:16:26.800On those champion power users, we also were really thoughtful to basically try to identify one per office location in the US in particular, because we know that this is sort of a grassroots effort while it's catching fire.
01:16:26.800 -->
01:16:34.960And so having sort of somebody that can really, you know, sell the story of the value add at the office level, they're a trusted source.
01:16:34.960 -->
01:16:36.319They're gonna reach that person.
01:16:36.319 -->
01:16:38.239So I think you got to go from the bottom up.
01:16:38.239 -->
01:16:41.039It doesn't necessarily have to be a partner, it can be an associate.
01:16:41.039 -->
01:16:44.800And this is a weird world where associates teach partner skills and vice versa.
01:16:44.800 -->
01:16:46.800It's not just a one-way street anymore.
01:16:46.800 -->
01:16:48.159I think that's really important.
01:16:48.159 -->
01:16:51.600The last thing is the firm to Dan's point has to invest.
01:16:51.600 -->
01:16:54.960The the bill the billable credit for it is very important.
01:16:54.960 -->
01:17:02.399But the related port is just, you know, firms that reconceptualize that RD is a part of our budget going forward.
01:17:02.399 -->
01:17:12.159Like we're not just, you know, real estate leases, um, cost for salaries and some overhead allocations for Westlaw anymore or Lexus Nexus.
01:17:12.159 -->
01:17:16.399Like this is a big part of our line item for any budget going forward.
01:17:16.399 -->
01:17:21.760And you quantify that in real dollars as well as um dollars spent by lawyer time.
01:17:21.760 -->
01:17:31.760But if you have a real RD budget, you will drive it and recognizing that this is the one world where we as law firms like to have everybody in little segments.
01:17:31.760 -->
01:17:38.000But you need lawyers that are technologists and they're not separate any longer.
01:17:38.000 -->
01:17:43.039And so if they think of themselves as innovating, it'll become self-fulfilling.
01:17:43.039 -->
01:17:52.720But you really have to set that tone at the top of the law firm that this is important and that we expect to see structural change at every level in that respect.
01:17:52.720 -->
01:18:06.880But that really does come down with having like deliberate budgets for this stuff going forward and giving enough grace to have tools that work, don't work, and just recognizing that's part of our business going forward.
01:18:06.880 -->
01:18:10.239We're gonna have tools that will work, ones that won't work.
01:18:10.239 -->
01:18:20.000But, you know, you know, I think law firms often are like, we want to direct ROI on something, whether we do this conference, we want to see whether it turns around in the dollars.
01:18:20.000 -->
01:18:33.600I think there has to be a greater uh willingness to let experimentation happen because some experimentation will fail, but the ones that really work are going to advance the firm and advance the legal industry.
01:18:33.600 -->
01:18:40.880So I think it's just a different way we think about practicing law going forward and how we manage the business of law on top of it.
01:18:42.399 -->
01:18:42.720Okay.
01:18:42.720 -->
01:18:44.239Well, we'll leave it there.
01:18:44.239 -->
01:18:48.159So uh Dan Brown, Adam Massaro, thank you very much for talking to me today.
01:18:48.159 -->
01:18:49.199This was excellent.
01:18:49.199 -->
01:18:50.239All right, thank you, Tom.
01:18:50.239 -->
01:18:52.239Hey Tom, it's a pleasure as always.
01:18:52.239 -->
01:18:53.760Yeah, same here.
01:18:53.760 -->
01:19:01.520The Emerging Litigation Podcast is a production of Critical Legal Content, which owns the awesome brand HB Litigation.
01:19:01.520 -->
01:19:05.039Once again, I'm Tom Hakey with Critical Legal Content at HB Litigation.
