AI, E-Discovery, and the Future Lawyer

Thanks so much, Daniel, for that introduction. I'm excited to be here and have this conversation with you and Eric today.

That's an excellent question, Daniel. As I look at what is possible with eDiscovery right now and particularly with the growth of data volumes and the variability of the types of data and collections, I think that information governance one would be a major blind spot. But what I would say is when you look at an eDiscovery project where the most value can be brought is lawyering it better, getting better agreements, getting better scoping, understanding the universe of your company, and drilling down to what's really important and negotiating for that in the context of litigation so that you can find what's proportional and do what's right for the litigation, but being mindful of proportionality and how that applies in that lawsuit and in that data environment just takes really good sophisticated on the ground lawyering of that problem. And I think that that's the biggest blind spot that we see with cases that seem to go right.

Those are all excellent points, Eric. And before I address that question directly, I wanna talk a little bit about, like, what I see as removing of barriers for the more senior lawyer. So historically, you've had an army of people in between you and the data as the senior lawyer in the case, and you tend to get curated sets of things or curated ideas based upon instructions that are very, as you said, linear that are gone out and found and reported back. This just removes all of those barriers to the data for the senior lawyers, for the substantive experts, for the trusted advisers to directly query data.

Once you learn how to prompt, you're talking with the dataset, and that is net new. I think that that as well, and I'd love the idea of the information loop changing and it becoming more of a keep informing us. I think that's a very valid insight and I think you'll see we'll start to create stories, but then you're constantly testing those stories as you get new endpoints, as you get, say, for instance, third party productions or you get additional deposition data. It's constantly revising that story just naturally as that information comes in.

So the challenge for both the lawyers within the law firms as well as for the clients to directly answer your end question really comes down to understanding the potential to unlock entirely new ways of solving the problem. So those people that use AI to basically automate what they're already doing and just replicate what they're doing are only gonna capture a fraction of the value of AI. It's going to be completely unlocking litigation in a different way because now we have these new tools that are gonna allow us to do new things and solve old problems with entirely new methods that just were not possible under the old extent billing hour.

So it's just going to open up and unlock all kinds of new potentialities for how we solve these problems and how we advise our clients.

In my private time, I use ChatGPT a lot too and I was actually struggling with that very question. Right? Because the beautiful thing about an AI engine is its conversation, is the back and forth, the querying, the thinking through new ideas, having it present to you new thoughts that you would not have thought about were it not for that conversation. But at some point, you have to extract it out of that back and forth q and a into a piece of work product, whether it's back to Evernote or something else.

For some lawyers, I think having the tactile piece of paper still has value. To have a binder with the twenty key documents that you come back to in a case still has value because that's physical, that's real, and you're engaging with it through a different means than a screen. And so I think we're gonna be figuring out over the next year or two where is the value for when you reduce a conversation with the data or an output out of a legal tool that's drafting you something, and then you turn it into a Word document that then you're working with within a different structure. And I think you're gonna see these information systems serve different purposes, whether it's discovery, whether it's finalization, or whether it's actually reducing something down to an exhibit set that you then are working within paper.

It's just a new means of query. It's a new means of looking at and engaging with data that's net new. It just didn't exist before. I I love a prompt.

Right? Because as a discovery lawyer, I go, if there's an important document in a case, what would you rather have? The prompt or the output of the chat TTP or whatever the Jet AI tool was? You probably rather have the prompt.

Right? Because the prompt says, here's what I want you to do. Here's the tone that I'm looking for. Here's the persona that I want you to undertake, and here's the background that I'm giving you to draft that.

It's gonna draft it different every time depending on the model that you're using, depending on how you're engaging with it, but that prompt, that is gold. It's money, and it's a new way of engaging. And And so I think what we're talking about here is just a new spectrum where you start with the conversation with data, and then you're struggling with when do you reduce that to the word document or the spreadsheet that we traditionally work with, and then how do you reduce that to paper. There's always gonna be a place for those different things, but I think the discovery happening in paper is absolutely dead.

That the discovery with paper is dead. You have to do that within these engines because there's just too much data, and there's just too much noise that's hiding your signal.

I think that ninety five percent of people who are utilizing Gen AI tools today probably don't really appreciate that and it affects whatever answer. Very similar to how a good litigator in a deposition or at a trial, if they don't ask the right question and fill it with context and instructions, you're not gonna get a good answer. So how have you been able to figure out, Clinton, that a good prompt is not the same as doing a Internet search engine search.

The short answer is I started using it, and I used it for virtually everything.

I use it for personal things to everything from asking you questions about an art museum that I intend to go to and engaging with art to parenting questions to using it in my legal practice with our approved tools within the firm that are, you know, approved by security and for client data and client consent. So it is actually just using it. And what I have found to be the coolest thing about using any GenAI tool is it really will match your energy. Right?

If you wanna put a case into a GenAI engine and say, summarize this case, it will absolutely meet your energy, and it will give you a high level summary of the case. It will also draft your LinkedIn posting that has a hundred and fifty words, that has a cultural comparison that's appropriate for LinkedIn, that has the tagline at the beginning of it where you cause the hard return. It will do what you tell it to if you take the time to engage with it. But in my experience, it's really just using it.

And I will still have moments as I'm sure the both of you do, really, probably every other week, but more likely weekly, where your hair actually stands up on the back of your neck and you say to yourself, I can't believe it just gave me that result. It was so fantastic. And you run up and tell your wife. And it's just that just happens all the time when you're actually engaging with these tools, and it's really special.

It's a it's a net new experience using technology. It just really doesn't happen before.

Well, there's so many Gen AI tools that either are, you know, for general purposes or then are for skill specific purposes. Right? So the skill specific tools, I think, are going to be very important to have a structure and a system by which you're using them. The general tools, I would say that we definitely will have prompt libraries that can be then reduced to the instant issue you're trying to address.

But I think that the general purpose tools are going to still be a lot of in a relationship and engaging with it. And I think that one of the things that when I hear your question, I go, what is it that I'm trying to do? Because when I look at a certain GenAI tool, if I'm just trying to ideate, I'm gonna talk to it another, a particular way. If I'm trying to build a very specific type of work product that's going to have a specific purpose, I'm gonna get much more granular.

And I think as we come up with use cases for, say, for instance, deposition digesting in the context of road accidents for a trucking company, I definitely think that you will build a very specific prompt that you will then reuse in that repetitive litigation setting. And so it's a lot of it's just purpose driven. Right? Because when I think of a prompt or a use case, I go, it's endless.

It can do virtually everything up to and including talking. Right? And so how can I leverage that? It's just there's so many use cases right now that that the prompt libraries, I think, are are very kind of in their infant stages.

But I think we're definitely gonna get there where we have a lot of structure around prompting when it's for that very specific purpose and it's repetitive, particularly when it's repetitive.

I think that comes from just experimentation, Eric. And I will say that one of the things that's really helped me is I've narrowed down the tools that I use on a daily basis to just a couple that fulfill the eighty percent of my needs so that I can learn how those particular engines work. And so when you're working with more of a tool that's specifically driven for a legal consumer and is built for a legal consumer, one of the things that that is really important is finding the right set of data for that engine to react to. So what pleadings you put into it and tell it to engage with that relate back to your prompt, That's one way in the legal setting that you can be very careful is restricting what data that it's reacting to, and that will drive better work product.

I have had the exact same experience as you where you over engineer a prompt, and what you end up getting is almost your prompt back to you in some ways with very little, you know, exercise of what we like to think of the freedom of AI to give us new ideas. And so that balance is just going to come with a lot of practice and engagement. We are adding a new team member to our team, and this team member is like no other team member we have ever had that will take our feedback, that doesn't require a lot of rest, that can absorb and reiterate and absorb and reiterate in a way that human beings cannot, but we have to learn to talk to it.

And that's just going to come almost relationally to learn what it reacts best to and how it works best, and then sharing that within your circle so that the rest of the practitioners around you can benefit from that.

Well, the first thing I would do, Daniel, was object because I'm the lawyer in the case and that's completely inappropriate for you to try to get my work product, so we would start there. But in terms of the underlying client prompts, I think that that's going to have to be dealt with in the eDiscovery space because it is data. If it still exists, it likely is discoverable to the extent that it relates to the claims and defenses in the case, and so will need to be discovered. But I think the discovery will happen, and we deal with new sources of data all the time just like we had to absorb social media when it first came out, and standards develop over time, and that will happen for our prompts.

But I think the more important point here is understanding that when you are hearing about output out of any AI tool, it's very difficult for you to engage the output of that tool without understanding both the prompt as well as the environment that the prompt is exercised in. So if it's a for instance, these a lot of these publicly facing GenAI tools, you would almost need the settings that were applicable to that user. Like, what did the user tell ChatGTP how to engage with the user or whatever that tool is? And then you have to understand is it session specific?

And so when you're hearing news stories about something that an AI engine did, you can't really absorb what that AI engine did without knowing what the prompt was and the environment in which that prompt was operating. And those are going to be very, very difficult things for us to deal with both culturally and then within the specific confines of a lawsuit as we look forward.

Well, the example that you gave has an ambiguity in which caused me to make that kind of jest about, like, me being counsel of record and not, necessarily a custodian. But you're honing in on eDiscovery question that is something that the bar is going to have to deal with. Right? And so it's that question of the discovery of a prompt that was used to generate a set of responsive documents. Right? And so I think you'll start to see that being dealt with in ESI protocols and through negotiation.

Because the carrot for the defense bar is the certainty of once I do this exercise, I have a compliant production that allows us to get to the merits of the case and move past the compliance exercise of a document review. And so that's why the keywords are given over a lot of times in that context because it allows the parties to get to the certainty of agreement. And so discovery of a prompt that is used to generate a set of responsive documents, that is something that the discovery bar is going to have to deal with. That is fundamentally different than a lawyer engaging with data and through their very prompt asking a specific question about a pattern or an event or a person or stress testing something, that's fundamental core mental work product that would never be discoverable.

It would be similar to you although I don't think of any of these Gen AI tools as a human being, but you would never discover the instructions the shareholder gave the associate in order to write the chronology that led to the summary judgment brief. Right? That would never happen. And so it's very similar to that when you're deeper into the confines of the actual creation of work product in the context of velocity.

No matter what tool we're using as lawyers, to the extent that we're using information to advise or give our professional judgment or recommend to a client that a particular course of conduct happen, it is up to us to determine where we need to be in terms of our certainty on the underlying facts of any given thing. And so we would have to and I think however much AI is going to unlock litigation and new potentialities, ultimately, the human judgment and the legal judgment and the professional judgment has to lie with the attorney that's making those recommendations. And so drilling down to what it's basing it on and using other sources to stress test the AI and the thinking absolutely is going to have to happen.

And that's one of the core principles when you look at a lot of the ethical advice coming out for lawyers. It's drilling down, it's checking, it's making sure that you're understanding what the AI is relying upon for a particular factual, you know, representation, and is that, in fact, reliable. It's where the hallucinated cases come in. That's the perfect example.

Right? It's a lawyer puts something into an a GenAI engine and it spits out a statement of law, and then it has these fake cases. And I look at that, and I go, well, the feature in that is the law that it's predicting. Right?

Because that's something that you probably should look at as a lawyer. You should calibrate that in a in a recognizable, you know, legal database and test that and research it. But the fake cases just show reliant judgment on an engine that hallucinated something. It just shows out the lack of professional care, and that's what people are reacting to.

It's that not testing, not ultimately retaining your professional obligations to drill down into what that engine is relying upon, and that's gonna look different in every circumstance. There's no way to make a bright line rule other than to say, you have to test those important things, anything that's material to the judgment that you're exercising on behalf of your client to make that recommendation to them.

That is not a lawyer. One can actually say that rule five point three b of the rules of professional responsibility say that all these JNA tools are not lawyers and therefore there's a responsibility for lawyers to work. We had a lawyer tell Eric and I once that, Daniel, JNI is no different than any other technology in any other, shortcut process by which if a CEO lawyer receives output from an associate and just signs their name to it and submits to the court without looking at it, it's the same effect of going through a JNI tool, getting the output, signing that into it, submitting it to the court.

Either way, there's an irresponsibility.

So the lawyer that's gonna do it before is the lawyer that's gonna do it now is basically the argument. I mean, do you buy into that argument? Is there still a lack of competency even if a lawyer is using this tool?

So I definitely think that the signing lawyer has a rule of an obligation to look at the facts and the law and make sure that there's a reasonable basis for the facts and that the law exists or you're making an argument for a reasonable extension of it. The way that you satisfy that in day to day practice will vary by circumstance and by the team that you have around you, of course. But when you're looking at AI, to accept an AI output without any stress testing of it, I would say that that would be something that would be very, very dangerous from an ethical standpoint and could lead you down a really dangerous path.

Where I thought you were going with the question where I think is a really important place to go too, though, is just the competence in the ability to prompt what would be a reliable output or even to engage with the tool and the practice that that takes. And I think that where we're gonna be talking about competency a lot, once we get through right now, the big talk of competency is the signatory to a hallucinated case brief. Right? But I think we're gonna be talking soon about the competency of understanding how these tools are working, understanding what GenAI is actually doing in terms of predicting a textual response to what you're saying, understanding how that works and practicing with it.

And I think you're gonna see some some things come down the pike in terms of ethics about those lawyers that really didn't understand how the tool worked and relied upon it in a way that factually situationally, you know, has some errors around it. And they're gonna be looking at those circumstances as well in terms of how they use the tools to generate those results.

That combines a couple of different concepts in there, Eric, and it's an interesting framework to think through. When it comes to the automatable tasks, right, the binary decisions that happen within a document review, those have been subject of discussion for as long as I can remember. It's the reason why we had continuous active learning and predictive coding before that developed. It was always around those automatable things that technology seems to be able to replace and replicate what the humans will do.

But what the machines can never replicate is trust, judgment, experience, and applying the law to an individual client's dilemma and advising them on how to navigate treacherous waters that could lead to litigation or lead to a great output. It's just changing the way that you might find the information on which you base that judgment. So one is this reducing barriers for the lawyers to engage directly with the data, which I think is quite lovely and is just net new and has never existed before. And so, like, those lawyers that have that judgment that the machines are just never gonna replace relationship, that is one it's going to enable them to get into the data so much faster.

And then in terms of the way that we look at a lawsuit will be completely unlocked and different than it ever has been before. So when you get a complaint, you now all of a sudden can look at that complaint through a different lens. You can put it through a series of prompts to create timelines, to create interview outlines, to create the first phase project plan, things that you could have never done before. You can run it through.

I think we're we're just not that far away from saying, draft me a motion to dismiss. Okay. Draft me the response to the motion to dismiss. Take that draft response to the motion to dismiss.

Revise the motion to dismiss. Do that over and over again until you've reduced it to the best version of a motion to dismiss, and you know exactly what facts are gonna be the key things. And then go and look at the way that court has decided things and tell me how I should posit those facts in this brief. We're just gonna be able to iterate more quickly and do things that just were not possible.

But at the end of the day, there is going to be a lawyer with judgment, experience in that matter, who's gonna advise the client. We put it through this process, but you know what more importantly? I think this is excellent. I've looked at the case law.

I've looked at the facts we're relying upon, and this is the best chance that we have to position us for success. And I think that's going to happen just naturally as we look forward. The final thing I would say in terms of the billable hour, I find very few lawyers, and maybe it's just my firm, but I don't think lawyers sit down and say, how many hours can I bill for this? And then work to the max hours.

I think most of us sit down and say, how can I create the best work product for my client that I possibly can? What's proportional to spend on this based of the needs of the case, the client's budget, the client's expectations, what I know they want from me? And then you work within that, And it never feels like we have enough time. And so I think these tools are just gonna enable us to just practice at a much higher level because it's gonna replace those things that were baseline things that we had to do to get to the value added work.

We're just gonna do them faster, and we're gonna spend, you know, money, but it's going to be on the value added stuff that the clients want. It's gonna position us for more success.

So it's hard for me to address a hypothetical that I don't live on the regular. Right? But what I would say would be is number one, that's just missing the fact that data volumes are just growing at an exponential rate like we have never seen. I don't know exactly what year you graduated law school, Daniel, but for me, it was two thousand and one.

And when I started law school, we were still using paper. So my first year of law school, I used all of the books. Right? And then I started practicing, and you use Westlaw and Lexis.

Those were the major tools when I graduated. And that was, like, a totally different experience. But now, can you imagine with the amount of regulation, with the amount of case law, trying to do anything with a paper library. It just wouldn't even be possible.

And then you flash forward, and you say, my first piece of litigation, I remember crawling around finding boxes of documents, and it was a big deal when we sat in a conference room with twenty or thirty boxes. You literally would travel to a different city and you would do that and you would tag them with post it notes and clips as the owner of a paralegal yelling at me for not being careful with how I clipped the documents and it made it confusing. And so there is that. And now we use eDiscovery tools, so we would never think of printing everything to paper and having people sit in a conference room.

This is very similar. It's we have exponential data growth. We have people creating data footprints that are, I would say, ten times as big as they were even four or five years ago. We have cell phones that are astronomically large in our collections now.

And so these tools are just gonna enable us to engage with that data in a different way. I talked to my kids and I said, well, if you don't have AI to do a broad based lingual search, what you're gonna do is you're gonna find a couple of sources that are gonna be very thin and narrow, and you don't know even what you're missing. AI is gonna allow us to start to see what we're missing and engage with datasets differently, and I just think it's going to become a reality no matter what the pushback is around what is a perception around the economic models of a law firm. The final thing that I would say is it's just going to unlock work product for our clients, value work product for our clients.

It just was not possible before the advent of AI. You couldn't draft a complaint, and then before you filed it, put it through an AI tool and say, tell me what are going to be the top five things that they're gonna argue are factually missing from this complaint that could matter to a motion to dismiss, and you're gonna reiterate the complaint around net new work product that probably would have never been created if it was going to be based on the billable hour. So it's just going to unlock these potentialities for clients to get more value from us regardless if it's based on value based billing or an AFA or if it's based on an hourly rate.

It really doesn't matter because both of those are just surrogates for how we charge for legal services for solving these problems. But AI is going to unlock all these new ways of solving these historical problems, but also just giving us new tools and new ways of looking at things.

I think with any technology revolution, you have those that will be adoptive and those that will be reluctant. Right? I mean, with social media, there was a whole string of lawyers that said I will never use this. And now you look at LinkedIn and the usage of LinkedIn, it's the exception where somebody's not really on there or using it.

It just became adopted over time. I think that there's a lot of advantages as a professional to being an earlier adopter in this particular trend. One of the things that I have been having just a ton of fun doing is creating a persona and a Gen AI agent that actually disagrees with me. And so I'll set it up to be a very well educated person who completely disagrees with something that I hold dear and fundamental, and then I'll argue with it.

And it's making me a better articulator of arguments. It's making me hold my feelings in check more because I'm learning to do that in a safe environment. And so, you know, I can imagine a world where you're gonna have a client meeting and you create the personas in an engine and you have the ability to see and to engage with them prior to that meeting. As you all know, because we talked before the podcast, I put your podcast page into a GenAI tool, told it to train like it was the two of you, and then to ask me questions, which I answered verbally, and which it told me to talk in shorter sentences, not to be so convoluted.

It gave me very candid feedback about my verbal speaking style to make it hopefully better for a podcast, which I'm sure all your listeners are now going, boy, he really didn't listen to it very well or didn't give great advice. But, nonetheless, it is possible now, so it can make us better professionals. And so I would just encourage everybody, start using it in your free time. If you're not comfortable using it for the practice of law, that's fine.

I get that. That is actually there's a valid conversation around that, particularly if you don't have access to closed circuit tools, and we all know we should not be putting anything client confidential in a public gen AI engine. But, like, just start using them and see the power. Have it articulate some ideas back to you, and you will start to really see the value in making you a better professional.

For the next generation of lawyers entering the field, I personally think that it's very exciting.

Especially for those coming up with a world that are shaped by AI. They're shaped by data. Where do you believe this new class of lawyers needs to understand about the future of practice if they want to lead and not follow?

So what I would say again is to engage with it. So one one way that a newer professional could engage with AI would be to train Intelligen AI tool to that it is a shareholder at a law firm and model up the type of shareholder that you want to be. And then start asking you questions about what it would be doing right now if it was you and tell it your circumstances, obviously, nothing client confidential. But that'd be one way is to think about it like a career coach and start to see what it would advise you to do.

Another thing that I would advise our young professionals to do would be to do your best to create a piece of work product in an AI engine. Obviously, one that's approved by the firm. It has anything to do with client confidential information, but you could create any kind of work product and then compare it to a lawyer's that you really respect and see how well it did. And then think about how you can iterate with that GenAI to get closer to what the lawyer you respect actually approved and filed in the case, and see how you can replicate that.

I will say for our for Baker Donelson Summer Associates, we actually got them all access to one of the GenAI tools that we use internally within the firm, specifically because we think that is going to shape and impact their entire legal career. So the more they lean into it early, the more they think about creative ways of having a coach. Think about Copilot. You could put the email you're about to send to a partner and say, does this strike this type of tone that I'm trying to strike here?

Do you have any suggestions for how to write more professionally? And it will help you do that and to learn to be a better professional much more quickly than Daniel and I could have possibly learned as we were coming up through where we still had a lot of dictation happening, where we still had very little use of email in the early two thousands. So it just think of it as an opportunity to engage with a professional that has limitless time, limitless energy to mentor you, and it does a really good job. I can tell you that because I've used it for those types of things a lot.

Eric, Daniel, I really enjoyed it and thank you all for having me on.