AI in Legal Technology

Alright. It is great to be with you gentlemen. Thank you for the invite, looking forward to the conversation.

And and when I talk to friends, you know, introduce myself, I always say that I'm one of those overly educated lawyers, one degree from overseas to from here, to make it interesting, I guess, my background, from Georgia, and, and, I'm sure, listeners would reflect on my accent and realize it's not, Georgia, Georgia here in the US. Georgia overseas in Eastern Europe, grew up there, former soccer player, former, foreign affairs, person, was an interpreter.

One of those, past lives did work for various international companies trading in commodities eventually moved to the US. Here, I was trained and started practice as an international trade lawyer. And then, like, many of us in early two thousand found myself too. I discovered it discovery. Let's put it that way.

Is this something that people are asking you about They are.

It's interesting. And and the genesis of that is in terms of where we are with respect to the stage of AI. And many have alluded to the fact that it's confusing for the audience in general, and whether that you're talking about in the legal industry or otherwise what we mean by AI. And I think at some point, we need to start as an industry, at least within legal to define what we're talking about.

We go back quite a bit. And growing up overseas and started with record player, cassette player, and etcetera. So if you think of that aspect of it, let's say a song, Right? Here are the five categories the way I think of, the technology and AI, which is speeding up content is one.

Organizing content is two. Exploring known content three, exploring unknown content four, and then generating content is five. So if you go back to that song, you recall, in order for you to listen to something really particular, and you wanted a, an artist song, and he said, I Wait. Wait.

Wait. Wait. Wait for this word. This is a really cool words in this song. You had to wait forever.

Now whatever streaming device you're using can say, hey, whatever, play this song and pick up at this juncture. That's your speeding up content, and we've been doing this. Right? Then now if I say I love Rachel Charles, I can have Intel Digital Library of Rachel songs organized.

And then I can explore that content, say, play old Ray Charles songs or Holiday right now. We're in the holiday season. Play old Ray Charles Holiday music. But then, and there's a fourth category to the unknown exploration.

I can also say a lovely Charles, and here are the songs. Can you suggest a new artist that I'm not aware of? I think concept truly. Was it Einstein who said that it just become, what is it, poeingly obvious that our technology has exceeded our humanity is the fifth category.

Which is generating content. And this is where we're struggling to wrap our heads around it because now it's not speed up and deliver music and explore unknown music. It's a now row write music.

So I think this is where that has really notionally completely turn things upside down because there is in current concern have we pushed the limits of technology and AI where it's really crossing the concepts of humanity.

So, absolutely right. When you go back to thinking about technologies and how it was advancing and, setting aside the legal industry, you know, playing checkers and then playing Garic Asparov, right, Watson, and then now Chinese Go, which is much more difficult, right, because of the all the sentiment aspects and the way it has to replicate human thinking. And now now we computers beat gold, I think when you ask me the question, how why is this different for legal professionals? This is my personal opinion, obviously. Look at it from this angle. Jared said was going back two thousand eight.

Was discovery a sexy topic? No. Was it for a legal professional, a way to advance in any way, they're in a company on a law firm, etcetera. No. It was just the role where it was created because of necessity. And there were not many folks who I wanna grow up to be any discovery professional, whether it's the legal side or the technical support side. It was not in the legal industry, the use of hard and early in any way, disturbing the balance of fact finders and merits, right, if you were gonna think about the discovery council, merits council.

Again, generative AI is hitting a very different topic. It is not extracting just the facts to get them quickly to you. Which was left more to discover the professionals. Now it can draft motion for summary judgment.

Now it can do legal research and on and on and on and on. It's attacking every aspect of legal profession. And I think this is kind of has woken up the, legal industry from that perspective. Wait a minute.

What do our attorneys in the future look like? What does legal education and law schools look like? How do you prep new generation of folks who are coming into the industry, whatever the role is to interact with the data.

And then so that's one. I specific to the legal industry, it's gained the core marriage aspect of it is impacted by the gen AI. The other one that gets just simple is If I sat down with my children and said, I that they probably still would say, I don't understand what my father does in terms of the discovery. Now when they hear me talk about Janette, they're, like, Oh, yeah.

I okay. There you're dealing with that. I get it. So there is that more every person at a dinner table so to speak.

Conversation. So you take those two worlds, whether the micro legal or macro in the world, it's fascinating in everybody.

That's a great question.

A lot of things are similar.

Let me name a few.

One is lack of predictability.

It was that it was there then. It it's here now, which is the traditional model was I have hundred thousand documents if you put twenty people per documents, hours, etcetera, I'll finish in twenty days, whatever it is.

That is for traditional lawyers, something that They can discern it easily. With the early tard, we supposed to stay locked. I know it's gonna be more efficient. It's gonna be less expensive, but I don't know exactly when I'm gonna be done.

That uncertainty is not something because of us as lawyers and legal industry being risk averse, something that people want to tolerate, and it's hard for them to wrap their mind around it. But and you're having exactly the same conversation now. That's kinda similarity. The other one, and this is I have spoken in other menus is to say folks, let's not repeat with the JNAI, the same mistake we made with TAR.

Which is confused folks either oversell or undersell what it actually does. That this is not some magic button, you still have to have a lot of different skills as an attorney as a technical professional to interact with that. So there are a lot of those conversations. In terms of what's different, I think we have at least reached the stage where there are enough outside, like, if I'm talking from the kind of internally law firm perspective, that I can point to a lot of examples where this has been done.

It has been done successfully and alleviate some of those concerns.

We're not there with the GenAI yet. Right? You still need those There are people who are doing case studies, how does it compare to contract attorney review versus the internal team review versus typical predictive and according models versus Zela Lam. So I think where we are slightly better positioned is it is more accepted.

Last trepidation about it, but a long way to go. And again, I wanna stress. We couldn't be exactly what we were with Todd. How many years now we've been doing Todd and there's so many barriers still yet.

Let's again, let's not repeat the same mistakes. That's my key here.

It's another powerful tool in your tool and people always kind of ask me that question. And when you dig in into these databases, look, I got eighty five things here. You know, social net, what can now list his clusters, email domain. I mean, there are a lot of things. Right? And my first question to them, what are you trying to accomplish? Because At the end of the day, let's not change tire with a screwdriver.

So the way I explain it, it's an extremely powerful new tool that we have to be cautious how we use it. But at the end of the day, it's about the result and validating it. Stop obsessing with the process. I think this is the mistake we made with Tar. Even though you had to do several more decision and even says, nobody cares about what the black box is, they still fight about the black box.

So the way we I try to explain is that you may have a case, not every case where these can be deployed and get you to the facts even faster than you would have done with the other tools in your toolbox.

I so that's interesting. I there are so much that one can elaborate there. Right? First of all, it depends on the language model.

Is the language model backed up by federal and statute and case law, etcetera. Because if you were to ask, was there, if you took a data and said, what's the insider trading? Well, is this the insider trading the definition in English language? Is the definition legal is a definition international.

Right? So I think I completely agree with the prompts are, and the queries are gonna be a debaters at work product. Depending how you frame it, how you ask it. And, the danger is also somebody coming in and saying, well, now I wanna sit down in front of your gen AI tool and ask questions.

Right? Is that something that one would want to permit? So, yes, to some extent, those conversations are not different. You're getting faster, but think about and this is where I think notionally, I'm having a little bit of a challenge as an industry.

We have written a lot of best practices about Tar. I, again, and I'm sure I'm could have missed this, but I yet to read somewhere. Anyone saying that let's say you have hundred thousand documents. How many can you review to say I'm done?

What if I have three case scenarios and in one unreview five hundred documents in one five thousand in one, only five documents? Is five documents or not. Right? And we don't talk about it.

What is not to or, you know, go to the federal rules here, but there is we operate under the standard of reasonable ink It's not perfection. It's reasonable inquiry. So it's five documents versus five thousand reasonable inquiries. So now put that on steroids.

And go to GenAI and say, you know what, I'm not even gonna review documents. I'm just gonna say, why don't you tell me everything in this database, what is relevant, and what is not is that gonna meet the reasonable inquiry standard? So I think this is where it's a bit different because with Todd, you're still having this negotiation, such terms, no such terms, and validation models, what did you review, richness, all of those? Those all those things, depending where we go and how fast LLMs got accelerated and how fast those l and then models get actually for lack of a better expression industrialized into the platforms.

We could be having very different conversation in terms of what a reasonable inquiry is.

Well, we had this conversation recently at a conference, and, these were all the things that, among many. I talked about your product. What about evidence? How what evidence?

How are you gonna be substantiating this? We have experts who testified to, right now, to Tar. Who's gonna be the expert? Because a lot of these models are being leverage by some of the large corporations have already developed.

So right now, many of the large models sit with a few. Unless you have large model, who are you bringing in as an expert? And and that yet to be developed. I don't think we have yet enough experts right now in the industry to say, know, when I'm about to go use JNAI and to do my case, can you please come and testify or help me authenticate?

And that's I mean, that that's gonna be issues that judges and and are going to grapple with.

No. That's that's totally fine. And I think more conceptual answer wise and in terms of how we do legal analysis. I, both overseas and in the US, a student of civilian jurisdiction.

For me it's about having a statutory approach and have a rule and apply to all the scenarios versus the common law is motivated. Okay. We need the case and we need the president, and then let's go by. Right?

So I'm answering with our answering that conceptually it's going to depend. So there is a school of thought about how do you authenticate, for example, deepfakes? It's like, and that is this whole notion of Euryn's court that is not it's a small court in the jurisdiction that doesn't have all the resources. And folks have this fake video, who do you deploy to authenticate it?

So one side can say exactly what you're mentioning. Oh, we are going to this dark side of things, we've got it's gonna be impossible to work it. On the other hand, those at Humota's traditionalism said, just use the simple rules and concepts and how we evolved. Denotion of authenticating signatures, hand signature is centuries old.

So we developed and there are experts who do that. So how is We just have evolved. Right?

Somebody said technology is something that wasn't there where you were not born. Well, so it's a new thing, and we were gonna figure out. But a side note, I agree with you, but I don't believe federal rules currently are designed to handle modern data, and not just the it's not there.

I talked about that we are changing the notion we with in this industry thought of document, custodian, in search terms. What do those mean these days? Is a Slack channel communication for third three days. Is it, one day?

Is it three days? Is it five segments? You have a case that says I'll give you a relevant hit plus five exchanges before and after. What is a document?

Now relevant does data does not necessarily reside in custodians. It might reside in channel. So As far as federals are concerned, we need to really consider updating them to catch up. Now having said that, considering the pace, contrarians would say, well, we will never do that.

Consider process it takes to update the rules. And then by the time they are implemented outdated. Yes. That is a there there's a fair point, but we still need to update them.

But I'm not worried as much about how we gonna your example of how we're gonna authenticate, you're gonna look at the trails and we're gonna develop the tools the same way that it is right now race between plagiarism tools and that those who detect Right? It's gonna be a race technology detecting wrongdoing.

I think we agree on that, calling me more of an old school. I don't think, there was this research I think Pew Research Group did it, which was asking a lot of data scientists and, in the industry, what would be the role of human agency And I think in terms of controlling decision, decade, decision making, and I think, like, fifty five or fifty six percent said, come twenty thirty five, machines will control the decision making process. And then and out of, you know, whatever forty five percent said, no. Humans will still be in charge of this stuff.

Mhmm.

I disagree. I think inherently the I don't think. And then then I don't know who said this. If I knew I would properly attribute, but AI is not replacing lawyering or lawyers.

AI is gonna replace lawyering and lawyers who do not use AI. This is gonna be an augmenting force, not, replacing force in my opinion.

Now having said that, the the crux of that statement is what we consider modern lawyering.

Because that itself has evolved, in terms of how we go about things. In the past, I think, when you when you think about eDiscovery and it importance, in fact, in the fact finding process and all of that. A lot of segments and low pretty much all segments of lawyer were considered to be art.

We come to realize that a lot of now, a lot of those parts can be processed, and they can be automated. So by definition technology already changed in the last twenty years, how we lawyer. But if we're strictly focusing on this gene AI and what it can do and I don't think it will eliminate again that human agency assuming that's what the quote kind of implied. I think it's just gonna augment. And it's a supercharge, our ability to the wire.

So, Jared, so you asked Z about what are the roadblocks for AI, but how are you defining AI for lawyers. Is it just the general sense of using conceptual searching and TAR? Is it gin AI? Because that I think that some of the challenge the roadblocks for for lawyers is really just defining. What does that even mean when you say AI?

And I think I'm gonna go yeah. I think I'm gonna go back to where I started, on our conversation, which is, when I would consult, and I, as I do on a daily, I'd ask them, are we speeding up, organizing, exploring, or generating. Right? I think there are different types of roadblocks. If you're talking about TAR where we still have it, we are fighting about, well, can search terms and TAR to be combined, we quote, you know, studies that are from seventies when it comes to searching grades. I think we need to do a little bit better.

We are more worried about, particular models rather than saying, like, at the end of the day, I've reached the position and recall and everything it looks good from the validation perspective. Those are the top challenges. I think it's gonna be very different barriers with respect to the gen AI type because you're talking about here the hallucinating aspects of it. You're talking about bias aspects of it.

I think clients are continuing to express concern to what extent we rely on those. I think for now it's gonna become Again, using TAR analogy, much more initially prioritization tool. It's much more fact finding and accelerating tool. Rather than people turning on saying up down ten queries, I have everything where it needs to be and let me turn it over.

So, I think it depends I completely agree with the idea of which which aspect of AI you're talking about, when it comes to barriers.

In terms of the quote that Daniel provided about loiring, I don't think loiring will change, but I do think the notion of offshore outsourcing for Brutforce first level reviews gonna take a significant hit because of GenAI.

And I think where, you know, as I'm sure you're following, there are some efforts with respect to additional kind of amendments, federal rules, and one of them areas is, privilege and what type of, you know, we talked about regular logs and categorical, etcetera. Approval is still remains a very expensive part of discovery.

I think we are going to because how the agent AI works in terms of actually creating that content based on what's already there that are gonna be pre coded and identified as work product for various reasons, I think that's second area where we're gonna see tremendous savings using GenAI.

Yes. I think the notion of I mean, we will still need your typical statistical validation and sampling and to do those things. However, because of the first those two things I mentioned, which is the, the, the substantive linear review as on prerogelog generation, we need to train our folks how to quality control that. This is not your go just look at the document, but understanding trends.

Why did particular model classify attorney work product this way, but attorney client privileged. That way, why it's interpreting third parties to break privilege here, but not there. Those nuances will require different type of quality control in real time. And I think gas time needs to be spent as these things are deployed to shift from more of a brute force to qualitative analysis on higher level reviews.

Yes. Absolutely.

Well, I think well, it is it too soon from the capability or is it too soon from defensibility and barriers. I don't think it's too soon from technology. It's there. And I, I mean, again, this is why my first two things I mentioned, I think that's coming.

Now where it's too soon is for the rules to deal with that. Going back to when I said, is it okay to review five documents versus five hundred to say I have done my reasonable inquiry in your hypothetical Daniel, if you wanna gonna do that, is that, meaning the reasonable inquiry criteria? I just loaded up in the, you know, the SI protocol And, Whatever JNAI, how does this shake up? And then I have what I have.

I don't think we're ready for it.

Yeah. I yes. I do think we are from the technology. The other element, and I was talking to a colleague who's, sharing this article is, will we ever reach the point I don't think this is next year, but, there are some articles written where the will will we have memory chips?

Because, Daniel, you're talking about taking discovery platforms and discovery technology and going out, so to speak, to hit the models, to help and integrate those things. My question is, will the models ever come to your pro via laptop? Right? And the colleague was kinda mentioning that in the past, If you needed a particular graphics, you had to have a different chip and driver and all this installed, but now you get one computer with all this memory, all this capacity, and all these things that in the past were at odds all integrated.

So will my whatever the company laptop in three years come in with loaded genetic aspects.

That, to me, is, I think, you know, it's a fascinating aspect.

But in terms of GenAI coming to discovery platforms, I think we're already there.

Yes. Absolutely. I mean, and is interesting. Wow. Office is, we moved in our, DC office, and I was looking at some of my old materials.

And I said, oh, wait a minute. There is that two thousand seven early, quote unquote, TAR with it, so, brought smile to my face, but it reminded me that when folks start saying today, Wow. You know, AI is here. It has been here for a long, long time.

So what do you think about six sigma? The way kind of you look at it, you lean and then you have six sigma, which is if you have any particular process twenty steps. Can we reduce twenty to eight? Great.

Now that we're down to eight, this part two is now can we make these eight as most efficient as possible. The legal project management is more of a watch today. A lot of client driven that, not only breaking down in, like, what they call work breakdown but understand your burn rates and because clients in twenty twenty three of these post financial crisis, obviously so are not only looking at lawyering aspect of it, but being fiscally disciplined and being efficient. So marrying these two together, when you think about the project, I mean, let's take the Lean Six Sigma.

It's important in the Mary with Little Project Management if you quoted x number of dollars for reviewing million documents. First of all, let's start applying link six sigma. Do you really need to review a million docket? Can we reduce it?

Now we're down to two hundred. What's the best way now to reduce that two hundred? Those principles very much help being efficient and disciplined. Now that's one element of it.

But then you also quoted the client, a million documents, whatever it is, and whomever, with all the resources would cost x number of money. Are you monitoring where the spend is that resource needs to be shifted? You know, some risk benefit assessment needs to be made. All of those are what I talked earlier.

Realization that there is in the entire legal product and legal service delivery There are portions that can be a project. I know a lot of people hate this word, but commoditize portions of the project. And then if they can be commoditized, then you can process it, that's where those disciplines come in.

Project management, absolute I mean, I the well, first thing you learn just generally when you start practicing, and you and those of us kind of law school. Right? You you learn in legal research, the the irac or rule, issue rule analysis conclusion. What is the first thing that you work at a firm or any new company, etcetera.

The first thing to say, what's the conclusion? What's the bottom line? Right? So there is a little bit of, I think, disconnect between coming out from law school understanding theoretical aspects of law and then actually practice of it.

And I do think print, like, a business principles project management principles do breach that gap of theoretical to practical?

Yeah. Jared knows better. Jared knows better. He's gonna he's gonna control me.

Oh, he's definitely Jared. He's definitely Jared.

Thanks again, Z thank you for having us and, hope the audience enjoys it.