Democratizing Legal Tech: The Evolution of AI in Document Review and Beyond

Yeah. Well, nice to be here. So nice to be back for you bring me back again, and those are two kind of words. You have been equally influential in my career, and I really, really appreciate our friendship. And, yes, at the time, a lot has happened for me since then. At the time of that recording, I was not at Relativity and I was at a great law firm and I chose to leave there for a variety of reasons but predominantly because when I saw where AI was taking us, I knew it was gonna change what we do and I needed to lean into this moment and challenge myself to take advantage of what we could do with this.

Yeah. I was gonna step back and say, you're using the word transform and I love that word. But I am not necessarily an AI expert, but I do consider myself a transformation expert. Because of how long I've been doing this, probably like you or many of the listeners, we've had to learn through many transformations.

Now, I'm gonna really date myself when we think back to I was around before the Internet, before cell phones, before active learning in our industry. Right? So these are all transformations that we've lived through before. And the fact that we're here, that we're listening to this podcast probably means that we successfully navigated those moments.

And this is just that next moment. But generative AI is happening faster. It has tremendous more upside and opportunity, I believe, for the dramatic way it will transform or change what we do in the legal industry than any other of those transformations, probably all those other transformations combined.

I think the legal industry has been sideswiped by the word transformation because it's really not about whole scale swapping out ways of working. It's about learning through progressive steps evolutionary. Right? And that's what you guys talked about this morning in the keynote.

Yeah. Sure. Again, I think that we start with Relativity itself. What makes Relativity so good today, and it's important to lay this foundation is, Relativity is one of the best, if not the best, system of record.

We have this amazing engine and storage platform where you can ingest and manage data and uncover insights from that data than any other platform out there. In AI powering that or a layer on top of that to access that knowledge and that information, that is the unlock. Right? This turning what was this amazing system of record where we put evidence and legal data and then being able to make it a system of action where you can mine and find out information and proactively learn things before maybe find out about your case and what's important about it before you go to court and before you have to have a deposition.

And then to your use case, then when we start expanding that and thinking about all the compliance and information governance or this great organization, legal data intelligence, which has these great use cases for expanding what we have been doing for years in discovery and investigations into these other areas that are underserved by both system of records and system of actions to actually do something and have a process around it. And AI is accelerating what we can do because it makes it so much faster to engage in those areas, to expand what we're doing, and put process around it.

Well, she kind of does these three things in these seven systems. So we have ontologies, and we talked to Erin yesterday. We have ontologies and lineage about data usage that we never had before. So, like, I think it's about data intelligence, but legal can lead the way because we've never there's no other teams that actually live off these indexes.

It's the value and the opportunity that we have in legal, I think, that doesn't exist anywhere else. I don't want to get too technical on this, but what you just said I think is very important. And I want to go one level deeper than what you just said because that's what AI brings, right? Today, in order to mine that information or let me back up.

Before AI, before generative AI in particular, in order to mine that information, you had to have labeled sets of data that some human classified in order to even find out what was there or what could be there. But generative AI, through a deep research capability, an agentic capability, doesn't it can actually go out and tell you what's there for you, right? You don't have to build a model or have a truth set or go have a human label documents that you use as examples or to train a machine learning AI, generative AI, and now the capability of AgenTic deep research AI, can go out and mine the data for you and summarize what it found back so that it does unlock new data sources that it was either too vast or we didn't have time to deal with.

So now it makes that accessible and available.

Sure. Well, first of all, think about agentics and agents in particular can be defined many ways. But in my mind, I'll ground it in two things. One is we're giving AI tools to take action to investigate and do things on its own.

Right? So if you think about human intelligence, when we define human intelligence, we usually go back to the stone age when we started using tools. And using tools is what makes humans different than animals. Right?

And so it was one of the things. And so I think we're at that moment with AI, where we're at the stone age of AI, where we're now with agents giving AI tools in order to take action.

And in the example I just gave with deep research, the alternative is I have to go run an index on the data in order to mine what's there and have a truth set or machine learning AI to kind of look at patterns.

I don't need to do any of that pre work in the future. I can just have agents, maybe that are time bound. You have twenty minutes to go look for as much as you can in this data set and then summarize what you find. Like that's what an agent can do with a tool.

And so that's how I see it. It's an assistant that we're empowering to take action on our behalf and at our direction. You give them some protocol prompt, but really what we're telling it is, here's the parameters of what I want you to do.

We safeguard it with security and permissions of what it's allowed to do, but then it takes action for us and brings us back a result that we can then either accept or refine through a human in the loop process.

Not to overhumanize it, but when I'm coaching our customers about using AI in our products, I do humanize it in the sense that you wouldn't trust if I'm a partner at a law firm. I shouldn't trust an associate who doesn't have the time that I do and I'm developing to learn a skill and to learn a practice area, I don't trust them to do all the research and then just file it without looking at it. You shouldn't do those things with AI either. AI needs an assistant, not a replacement.

Right? So And then I'll flip it and say, companies like Relativity, it's incumbent upon us to ensure that we build products that encourage good behavior, that encourage human in the loop and build workflows. You know, we're leveraging LLMs and we're trying to use the best LLMs possible and we engineer and add our secret sauce to all these things. But the most important thing that I think we can do is ensure that our products follow a set of AI principles in our development practice to ensure security and confidence and human in the loop.

And, we build the framework and workflows to encourage good behavior by our users so that they don't get in trouble. I don't want any of our users to find themselves in a bind where they've done something that is inappropriate or uncovers a hallucination because they didn't check. We build into our products. That's our responsibility to build a great product that encourages that our users use it properly.

Yeah. Well, for starters, forgive me if you heard me say this, but in the twenty five plus years that I build my time over my career, there was never a single day in my entire career where I got done with the day and said, I've done everything I can possibly do. In fact, it's the exact opposite. Every day of my life was a compromise.

What can I put off till tomorrow or what can I not do? And so if I can have AI accelerate and do the things that are mundane and repeatable so that I can do the higher value things more or do more of it, that's the win that I want. Perhaps I choose to go home earlier once in a while. But overall, I think we all have a tremendous amount of work to do that we just can't get done.

And it's usually the higher value work that doesn't get done. I remember early in my career, we would sit around and we would have an hour. I'd go into a partner's office and we'd strategize and think about the case and just sit there and relax in the moment and just riff on how can we win this case. I don't ever do that anymore.

The world's so fast. Emails and messaging and all these things. We don't have time to really think like we used to. And I'm hoping that this will be one of the things that help us do that.

And then second to your point, like, don't know if you've noticed, but the amount of data and documents in the world isn't stopping growing. In fact, it's accelerating. There was a point today in the keynote. If you just think about the billions of people who are prompting and chat GPT and copilot right now, that's all the future evidence that's going to be in our cases and investigations tomorrow.

That's all entirely new data sources that we're going to have to deal with. The amount of data in the world is exponentially growing. We have to find ways to deal with that faster and easier and better or else we will fall behind and not be able to perform the work that our clients desperately need us to do.

Yeah. Well, I think we're definitely going to see an influx of I'm hoping for new users, right? We, of course, have a tremendous amount of need for our current user base. But I think with products like aiR Assist and AI, we have to draw in up the food chain in a law firm or in a lawyer world of higher level associates and partners more into our product and get them closer to the data because the time that they have to get answers and the iterative nature of that is going to require things like aiR Assist, a chatbot style natural language query capability where you have a partner or senior associate or the lead investigator into the product that are working in there iterating and learning about it and mining that information at speeds and scale that they that were unimaginable years ago.

Again, because the amount of data is growing so fast, that's option we have. But I will go back to where you started that question in a two part answer is that maybe I'm an idealist. But I do think we need to rethink the billable hour. I'm a skeptic in that I don't know that it will actually happen in that maybe before I retire.

But I do think that we have to start tying what we do in any industry to the value we deliver. And I'm not so sure that a billable hour is the value I'm delivering. The outcome, what I learned, the risk reduction I did for my client, the outcome of the lawsuit. Those are the things that we should be tying our cost to commercial models to and I think our customers would appreciate if we do that.

No. Not at all. First of all, I don't think that AI replaces all of review, just to be clear. I believe that AI augments review.

It can replace certain tasks. We think about first pass review. I do think if I'm the first pass reviewer and my only job is to review for relevance that I should be upskilling and thinking about what's next in my career path because AI can do that as good or better and a lot faster than you can do it today. Today, in our product, we're reviewing our customers have the ability to do the equivalent of what a few thousand first pass reviewers could do in a single day.

You'd have to hire a few thousand people to do what AI could do. And we're getting more consistent and more accurate results than those traditional first pass reviewers. So I do think there are some things that you have to think about what I do differently or what should I personally do differently in my career if I'm in a certain role. However, QC, validation, learning, we still have to do something with that data.

And, the pile of good data or relevant data to my case is getting bigger. So, that data still has to be looked at, thought about, analyzed, digested by humans in order to come up with the strategy and the arguments that win the case or reduce the risk.

Yeah. Mean, would argue that like you're missing out on probably cases or matters that you have to pass up because you don't have the skills or the people or the technology to do it. And this is much cheaper than having to go hire a team or of humans to help sift through in a large class action. Right?

So if you're a small plaintiff's firm, this is actually a leveler of the playing field to allow any size, any firm, any size, I think, to play in a type of matter that they want. So I do think in that way, it's access to justice as a level up for those smaller and mid sized firms. You use the term access to justice, so I'm going to spin it a slightly different way, which of course, we're very proud of our initiatives to help out pro bono cases and using this technology that we make available to customers. I think it has a tremendous opportunity to open up access to justice for pro bono purposes and the backlog in government.

We're actively working. We talked about it in the keynote today. The backlog of FOIA, like Freedom of Information Access Requests, that's huge in the government and it's way way way too time consuming. But this type of technology could basically undo that entire backlog in much more economical way for all of us that helps both, I don't know, I'm a US citizen, I want the government to be more efficient, but I also want access to this information so that we can hold the government accountable.

Right? So, I think it has a lot of benefits on both sides.

Yeah, thanks for asking that question and for bringing that up. We're very excited about this. Our products have been in the market now. Our generative AI products of aiR for Review and aiR for privilege have been in the market for over twenty months.

We've seen an incredible uptick in adoption. But we also sense that our customers were holding back because they're wrestling with how to price this, how do I go to my client, how does it cannibalize my review business. Not to mention, perhaps it is more importantly to us at Relativity, there was this competition between our aiR products and our best in class that we've been using for decades awesome review tools that are already integrated into Relativity. So, we had this internal competition between our aiR products and our traditional products.

So, we wanted to change and disrupt that whole notion by integrating AI into our Relativity platform so that they could work together so that you don't have to approach each project as, Do I use AI or not? Instead, we wanted you to choose the best tool for your situation that you should leverage all the non AI stuff if it works for your case and AI when it's appropriate. So like this really democratizes to use Phil's word and brings in and levels the playing field making it available to all of our users, not as a premium, but as this is something that's in my toolkit that I'm going to deploy anytime I get a chance when it makes sense.

Thank you for having me. Great to be here. Honored to have this been my second time. Thank you.