This week on The Geek in Review, we discuss the future of legal technology with Dr. Megan Ma, a distinguished research fellow and Associate Director of the Stanford Program in Law, Science, and Technology at the Stanford Center for Legal Informatics, also known as Codex. Dr. Ma’s groundbreaking work in integrating generative AI into legal applications takes center stage as she shares her insights on translating legal knowledge into code and the implications of human-machine collaboration in the legal field.
Dr. Ma discusses her unique background in law and linguistics, emphasizing the challenges of translating legal language into deterministic computer code. Her fascination with language and its nuances has inspired her research at Codex, where she explores how legal concepts can be effectively communicated through technology. She recounts her academic journey, explaining how her multilingual skills and interest in natural language processing have shaped her approach to developing legal tech solutions.
At Codex, the mission is to empower the legal system through innovative technology. Dr. Ma describes Codex as a collaborative hub, where experts from various fields work together to address inefficiencies and pain points in the legal system. She highlights the center’s commitment to human-centered design, ensuring that their technological advancements are co-created with relevant stakeholders. This approach ensures that the tools and solutions developed at Codex are practical and beneficial for both lawyers and clients.
One of the standout initiatives at Codex is their mentorship model, designed to mirror the traditional mentorship found in law firms. Dr. Ma explains how they use AI to create legal personas based on the redline practices of experienced partners. This innovative approach allows junior associates to receive focused guidance, helping them improve their skills and knowledge in a more efficient and impactful manner. By integrating AI into the mentorship process, Codex aims to bridge the gap between theoretical legal education and practical experience.
Dr. Ma introduces the concept of agentic workflows, where AI agents make autonomous decisions based on specified goals rather than predefined tasks. This dynamic interaction is particularly useful in legal negotiations, where unforeseen circumstances often arise. The negotiation model developed by Codex includes features like client rooms, expert consultations, and various levels of difficulty to simulate real-world scenarios. This hands-on training tool is designed to help young lawyers navigate complex negotiations and improve their problem-solving skills.
In the Crystal Ball segment, Dr. Ma shares her vision for the future of legal technology. She emphasizes the importance of developing tools that tap into the legal brain, focusing on the process behind legal decisions rather than just the end product. By capturing the experiential knowledge of seasoned lawyers, Codex aims to create more effective and intuitive AI tools that can support the legal profession in new and innovative ways. Dr. Ma’s insights highlight the potential for AI to transform the legal field, making it more efficient, accessible, and responsive to the needs of both practitioners and clients.
Listen on mobile platforms:
Apple Podcasts | Spotify | YouTube
Contact Us:
X: @gebauerm, or @glambert
Email: geekinreviewpodcast@gmail.com
Music: Jerry David DeCicca
Transcript
[00:00:00] Marlene Gebauer: Welcome to the Geek and Review, the podcast focused on innovative and creative ideas in the legal industry. I’m Marlene Gebauer.
[00:00:13] Greg Lambert: And I’m Greg Lambert. And on today’s show, we’re going to bring in one of the biggest names in the application of generative AI within the legal field, Dr. Megan Ma.
[00:00:23] Marlene Gebauer: Yes, Dr. Megan Ma is a distinguished research fellow and Associate Director of the Stanford Program in Law, Science and Technology at the Stanford Center for Legal Informatics, or otherwise known as Codex.
[00:00:35] Megan Ma: Her research is at the forefront of integrating generative AI into legal applications, focusing, translating legal knowledge into code and exploring the implications of human machine collaboration. Megan, welcome to the Geek Review. Hi, Marlene. Hi, Greg. Thanks for having me. And that was such a generous introduction, so I feel like a little blushy from that.
Greg Lambert: well, I will tell you this. I had AI write that for you.
[00:01:03] Marlene Gebauer: AI loves you, Megan.
[00:01:06] Greg Lambert: Megan, could you mind sharing a little bit about your background and about the mission of Codex there at Stanford?
[00:01:13] Megan Ma: So, my background actually is in law and linguistics. I wrote my Ph. D. Okay, what
[00:01:17] Greg Lambert: does that mean?
What does that mean?
[00:01:19] Megan Ma: Yeah, so, I obviously have some legal training, but also from the linguistics perspective, I wrote my Ph. D. on how do you translate law into computer code, and I dug into sort of the science of language, what was important about, Natural language in the role of legal language, and how natural language was its primary medium of communication.
and in the era of pre LLMs, it was all about expert systems. It was use of NLP. And so if we were to take concepts that, have always been encased in the way that we as humans use natural language, how do you actually Make those types of translations in more deterministic languages, such as any type of computer program.
For me, I think it was from this inspiration that I speak three languages and every time I speak a different language, I have to think in that language and sometimes I make a bunch of concepts up just because there aren’t exact translations. And so that kind of inspired me, you know, if you were to use a different medium to communicate the law, what are the sort of losses that exist?
and then the mission of Codex. So why I joined Codex was because, well, I think it’s a place that’s sort of like an island of misfit toys. I think that translates well in legal tech as well. and people that are in that community. and I think The center focal point for us is always how do we enable legal empowerment through information technology?
That’s what you know, Roland and Mike, who sort of co founded the center. That’s still their mission today. we build technology for law. We think about all the different inefficiencies, the pain points that we see in our legal system, the stakeholders across our ecosystem, and how do we kind of think about technology as a possible solution.
So, so is legal like a fourth language, do you think? There’s lots of comments about whether it’s an alien dialect to natural language.
[00:03:12] Greg Lambert: I think we’ve turned it into one, so that’s True.
[00:03:17] Marlene Gebauer: Codex emphasizes human centered design in legal tech. can you explain what this entails and how that benefits both lawyers and clients?
[00:03:27] Megan Ma: So for us, how we see it is that everything is applied research the way that we operate. it means that it’s co created with relevant stakeholders. And so it starts at the user persona who we think the technology is going to have engagements with, and it’s across the entire life cycle.
And so this translates across all of the research that we do. For example, all of our industry partners, All of our academic partners, as well as all of our tech partners. We constantly have iterative feedback loops in, how different people engage with the systems that we build. for example, in a recent project that we put together the negotiation simulator, we co created it across engineers, as well as experience mergers and acquisitions, M and a partners of 40 years of experience.
And the question that we kept returning to is. How would you actually integrate this into your workflow? And that starts from the very insemination of the project. it’s not, this technology, how do we retrofit it into your practice? It is, this is your practice right now. These are the types of analogs in those practices.
[00:04:30] Greg Lambert: Does this technology actually even make sense in light of those factors? Interesting. You know, last week we had, Conrad Everhard on and I think he’s partnering, with Codex as well. So, how can law firms themselves reach out to Codex and leverage some of the innovative technology solutions that, you set up or established?
[00:04:55] Megan Ma: So we, there’s multiple ways to do so. our kind of generic catchall, sometimes we use the term like all you can eat codex buffet in a sense where you get to participate not only in the events, but also in the research is become one of our affiliate members. and the membership program is something that we can share more information about.
But if it’s a more specific project that is of interest, you have particular hard problems that you would like to tackle, But as many industries do, they have stakeholders to report to. And if the value isn’t immediate, it’s a little bit hard to justify and expend resources on.
And so we kind of ask you to toss those problems to us. And as academics, that’s our bread and butter is to do the hardest problems that might lead us to nowhere at times. so that’s another way, if there is a specific use case of interest. One thing that we’re starting to launch soon is our rapid prototyping lab.
This is the idea that it’s sort of an affiliate in a way to Codex and our broader research center. But the goal is that we prototype across a brain trust of our academic researchers. our industry partners as well as our tech partners. basically it’s something very similar to how Flatiron Law got together with us.
I proposed, I want to kind of help improve legal education, help, young lawyers figure out those nuances when they have to negotiate a deal. And so how do you do that? Well, I asked, Lenny and Conrad, and they basically downloaded their brains to us. They spent, hours with us on these user interviews, translating very specific war stories to us.
How do you capture all of this experiential knowledge? And we took it and we basically trained an agent. And now you can, you know, chat with a machine council to kind of, hone in on some of those skills.
[00:06:36] Greg Lambert: there’s always been kind of this, wall between academics and the practice of law, at least in my experience, that there’s a lot of things that, People within the law firm environment tend to think that, well, what they’re doing at Stanford is, you know, it’s much more empirical research on, and with the end result of that they’re going to write a great paper, that’s the end.
And it was like, and, you know, I’ve got to hours to bill, how do you break down that wall and determine. Whether or not there’s, an encouraged interaction between the academics and the practice of law.
[00:07:21] Megan Ma: Yeah, absolutely. So I think you’re right that there is a tendency for some of our academics to live in our theories.
I think a lot of the technology that’s put out there by vendors in the legal tech space, they have the advantage of having the data already, and having those deep seated relationships. But as a result, they sort of treat generative AI, some of this really emerging technology and frontier technology as almost a plugin.
We have our existing tech stack. How do we sort of like just integrate that piece and component I think there is Definitely an opportunity for this technology. People talk about hallucinations, but the other side of the coin is the creativity. And we really, really tap into this creative element of generative AI.
And we want to say how do we use it so that we can. Build out, and play out almost in a totally different way, evaluations of the practice. We don’t have any legal metrics that exist on, what is a good contract and whatnot. And, academics can theorize about it, but why we have industry partners is because You’re war stories.
Your experiential knowledge captures precisely what is quality and generative is capable of kind of figure out ways to mirror and simulate these experiences. And when we simulate them, we can start building out an empirical rapport. So for us, that’s how we see the bridge existing is that we want to preview to young lawyers and law students.
You know, this is what you were taught a theoretical way of understanding the law. Okay. Thank you. But this is actually glimpses of practice that we can showcase to you through these different tools that we’re building, like the negotiation simulator. And then you can kind of start saying, as I graduate out of law school, into my early days of practice, I can continue to refine those set of skills and you have that natural lead in as opposed to feeling a little bit disgruntled and disillusioned when all you’ve been doing is kind of like handling paper day in, day out.
[00:09:16] Marlene Gebauer: Megan, I’m really interested in learning more about the mentorship model that Codex has designed. how does that support development of legal professionals and technology innovators?
[00:09:27] Megan Ma: So, I’d love to say that we kind of developed that, but actually what we’re doing is mirroring the mentorship that already exists internal to law firms.
in all different departments at law firms, you’ll see partners, being assigned to young associates and, young associates in their kind of, Totally naive ways will follow in the tracks of the partners, and sometimes even their stylistic techniques become ones that these associates embody.
And so we sort of understood that as incredibly entrenched in existing practice. So we started to design tools that we saw like generative A. I might be able to generate a contract, but it isn’t capturing actually the specificity and the granularity of the specific partner.
And so if you can imagine that a partner has asked you as a junior associate to just red line a contract for the first time, you don’t even know what you don’t know. In order to avoid, a one on one situation where you go into a room with a partner and start maybe peacocking a little bit and kind of like noodling around, you actually can get a very focused practice by developing what we call legal personas or AI legal personas.
And what we’ve did is we’ve partnered with a Canadian law firm. They actually gave us, redlined contracts, sets of them. And then we have versions one to 10 of those contracts. And we essentially extracted all of the Delta between these different versions and trained a model. Then we’re also adding a layer of, other components like their yellow pad notes, their scratches here and there, maybe they’ll have redline contracts where they draw crutches on them because, that’s what they do.
their signal of saying, redo this whole thing. And then essentially it captures the spirit of this partner. And so how that applies is imagine I’m on, you know, Microsoft word favorite tool in the world. and I’m redlining a contract as a junior. And then I’m like, Greg, Greg’s my partner. I’m going to make sure in advance that the next one on one I don’t want to get yelled at, or I want to sound smart.
And so I will select him as like my AI partner. Your sort of red lines that we’ve trained on, and your kind of understanding of the contract will then overlay my own red lines and then I can compare, you know, are there differences or are there flags that A. I. Greg has identified that, you know, I had no idea about.
And so when I go into my one on one, I can say, you know, I’ve gone through the exercise, and I have a hierarchy of issues I want to bring up, and then my meeting becomes much more focused and much more valuable. Or there’s a circumstance in which, you know, Marlene, she’s like THE expert on all things, transactions, and so I want to get Marlene’s eyes into it, as well as Greg’s, and so I have them.
All of your red lines overlay mine. And so then I can have this interesting comparative perspective. So it’s supposed to mirror an analog of the mentorship that already exists in the law firm space, but replicated in some of our tools as well.
[00:12:21] Greg Lambert: I have a couple of questions on that. one, do you think, using the Canadian, law firm is going to skew, to being too nice?
[00:12:32] Marlene Gebauer: or will they approach things differently? I mean, honestly, and, do regions make a difference?
[00:12:39] Megan Ma: I think it’s certainly. Does, for us, we actually asked them, do you have a particularly thorny, law firm partner that we can work with because their spirited nature could, be most revealing in some of our empirical tests.
I think, but I think like Marlene, you raise an important point around sort of jurisdictional differences, especially in the specific language that’s used in the contracts. We definitely see that. The way we are training our models is not necessarily meant to be scalable across jurisdictions. The idea is, is that these models are playing on the specificity of the person, and it’s meant to be internal to the law firm.
Again, it comes from a, heart for training, but also potentially useful internal to the law firm if you have, say, an M& A deal and your target company has in the past run into some employment issues, for example. So you might. In an analog way, ask your employment partner, but that employment partner most likely will be quite busy.
And so in the same way, this tool allows you to preview or flag some issues in advance that you can then very quickly send an email to the employment partner on, but again, It’s not meant to be tuned at a generalizable method. It’s meant to be very, very specific.
[00:13:52] Greg Lambert: How much data do you need to create something that mirrors the personality of a partner?
[00:14:01] Megan Ma: So actually less data than you expect, but what’s interesting is the data has to be quite particular. and what I mean by that is we, in the beginning, we just focused on red lines and we thought we just need an enormous number of red lined contracts and especially the differences in the red lines.
But I think what you sort of all implicitly know is that the contract can only say so much even when you redline. The act of redlining is sort of an end stage process, meaning you’ve made the decision in some way to do that edit. Why you did that edit is possibly tethered to another document, maybe it’s a direct email from your client, maybe it’s associated with a number of external events, or maybe it’s something that is.
a gut reaction based on other cases that you’ve had. And so in order to capture those experiences, we needed to train on other kernels of your, I guess, your brain. As the partner, your kind of like playbook. And what we mean by that is some partners have their own sense of issues lists. Sometimes they have their own kind of document.
We had to add those pieces of information as relevant context in addition to those delta red lines, in order to be able to better capture the spirit.
[00:15:18] Greg Lambert: you mentioned this before that you developed a negotiation model for inputs, from attorneys.
what are some of the features of this model and how does it improve negotiation outcomes?
[00:15:34] Megan Ma: Yes. So, Some of our core features is that we have a client room. So the first skill that we’re trying to teach them is how do you even approach a client? and you know, especially in some deals, and especially if the company that is selling is particularly small, then you tend to have very, Passionate owners, I think, and clients, they might say like, you know, I came here, with nothing and started this company with my own two hands and we’ll go into lots of narratives.
How do you kind of tease out the relevant facts that you need in order to represent them? One of the features is that some of the deals that you will do involve a whole slew of other specialized lawyers that you might need to interact with. And so say you’re negotiating the new corporate structure post merger, and you want to maximize the tax benefit of the buyer, right?
You obviously will need to consult a tax lawyer in that regard. And so we have a functionality where you can actually click in and consult the expert and I’ll redirect you to a separate agent that knows everything there is to know about tax. and then you can go back into the negotiation room.
There’s also, naturally in a sort of pedagogical way, we have junior associates, senior associates, and partners. So three levels that are built in. and the difficulty is actually reflective of the reaction and sensitivity to the industry. And so what we mean by that is, for example, senior associates, say you’re negotiating the reps and warranties, there might be a sudden cybersecurity incident that you found out Related to the target company.
The target company, it turns out has suddenly, unfortunately had an incident where they had their databases breached and customer data was leaked. So as a senior associate, this is an unforeseen circumstance. Your partner might put the challenge on you to figure out how to navigate through this.
How do you then pivot some of your original sort of, representations and warranties, how you like rejig liability there and things like that. and then with a partner level, You’re really called in if the deal is something is really nuclear that’s happening with your deal. For example, it wasn’t just one cyber security incident.
You realize the entire security infrastructure of the target company is called into question. And so, you know, at this point, as someone representing the buyer, I should probably think about just like drawing out the assets and buying it, doing an asset sale instead of the original sort of maybe equity sale or some sort of other, form.
And so, In this case, then you have to navigate the difficult waters of pivoting out of the equity sale. And so this is kind of what we did in terms of figuring out the difficulty. And then we also have different modules that play into different types of skills.
[00:18:09] Greg Lambert: I like the, teasing a story out. it reminds me, I had a history professor that, would say, you know, beware of these stories because, you know, someone will say, well, I came into town with nothing but the clothes on my back and a knapsack on the end of my walking stick. And when you find out that he actually had three million dollars in the knapsack, it kind of changes the narrative.
[00:18:32] Megan Ma: I mean, so that’s something that we’ll ask, our young lawyers or students to tease out in the fact finding.
[00:18:39] Greg Lambert: Yeah. in the library world, we call that a reference interview. That’s
[00:18:42] Marlene Gebauer: right. That’s right. get down to the real nuts and bolts. Megan, you have referenced that the negotiation model, is comprised of agents.
So, It sounds like there is an Agenic Workflow in this model, and I don’t know that everybody knows about Agents and Agenic Workflow. It’s, relatively interesting, I guess it’s become relatively popular recently. so could you describe the concept of Agenic Workflow and how Codex integrates that into, The negotiation model and possibly other projects.
[00:19:18] Megan Ma: Yeah. So agents are sort of all the rage now, of course, rightfully pointed out. they’re effectively like to put in the simplest way, it’s just software programs that know how to take autonomous decisions based on specified goals. And so the key word here is it’s goals and not tasks. And so if you communicate to these agents that, I want to make sure that this client gets the best deal.
That’s relatively abstract. Actually, a lot of agents aren’t able to get at that level of abstraction. But that’s kind of giving you the idea that you’re not saying like, Right. X thing like it’s not down to that level of granularity, which is what our first interactions or first engagements with large language models was like, and then for us, a lot of what we do is in the agent space.
We have even multi agent simulations. for our negotiation, the whole goal is that we want, this, machine counsel, so to speak, to be highly dynamic. They know the goal that they have to get this deal over the finish line, but we didn’t want them to be specific about what tasks we want them to achieve.
And the reason is because you’re trying to help the user or the young lawyer navigate through sort of unforeseen responses. and even though our model, we actually have an agentic framework that breaks down communication. So what, the machine decides is the belief of the user. what it decides is the intention of the own counsel and what their goal is.
And then finally, what actions they would take. Basically allows any young lawyer to be able to react accordingly. And so to your question, we integrate it because to us, that’s actually a much better teaching mechanism. If we just, have tools or models that are still at the task level, then it doesn’t give that flexibility that we want our young lawyers to actually have to make decisions for themselves and to react naturally.
[00:21:23] Greg Lambert: I’m curious, does the agents themselves, do they just talk to each other or does it get to a point where the agent has to get more information from the human and then it asks the human to input more information so that they have a better pathway to their goal?
[00:21:42] Megan Ma: So it’s both actually. So just as we have the option to solicit experts.
So say in that example of the renegotiating the corporate structure, I can make the decision that, okay, at this point, I need to consult with my tax lawyer to get more information on the machine council side. They can automatically or autonomously make the decision that they need to consult the tax agent too.
and so that is, Automatic because you’re sort of mirroring two teams of lawyers effectively. But the human part is that, you’re trying to get from opposing council signals or trying to read the signals that more information or, more expertise is required. So, that’s why it’s again, very, very dynamic and it’s all reactionary.
[00:22:34] Marlene Gebauer: This is interesting because, I haven’t heard it described like this before. I mean, I’ve read that agents can kind of do specific tasks as part of a overall, task. I think the example I’ve seen is like writing an essay and they kind of break it down into, you know, it can proofread, it can do this, it can do that in order to write the essay.
So, is that a misunderstanding on my part or is that. Sort of different than what you’re doing.
[00:22:59] Megan Ma: No, certainly agents can do tasks. the whole idea is that, when you describe an agentic workflow, there’s someone that is by someone, I mean, one agent that could be the planning agent.
So they decide what are the next logical steps to take. just like Microsoft has this, framework called micro agents where you have one agent, that’s the planning agent, and then it. Signals to other models that are going to do the writing that are going to do the research that are going to do different components.
And it is a task for us. The agents that we’re building here with the negotiation model is a communicative agent. And so the goal is actually more to touch upon. How do you engage dynamically based on a given problem that you want to solve? the goal is that you just want to get this deal over the finish line.
[00:23:47] Greg Lambert: well, let me, let me actually ask a little bit more on the agent’s, process. Cause I’m interested in how do the attorneys interact with it? Is it text based? Is it voice? Yes. What, I could actually see people becoming very, addicted to this sort of thing, and actually, you know, turning it into a human.
[00:24:12] Megan Ma: So it’s both actually, we have a text dialogue, typical of like what you see of a chat bot, you just text and it will respond. we also have a. Text to speech function, where actually it can communicate. in a dialogic way. And so it can respond to you. but actually because some of our young lawyers or law students are in the Tik Tok era, it can come off quite slow for them.
So they actually prefer the text engagement, we have now started to develop our version of the negotiation simulator on smartphone. so there might be an opportunity to do much more of that Speech to text or, text to speech engagement.
[00:24:54] Greg Lambert: so Megan, let’s get back to some of the work that Codex is doing, and how do you, how is it that Codex and the work that’s being conducted there influences, the legal education at Stanford, but also beyond that? Are there any specific programs or initiatives that you’re pretty excited about that you think are really kind of affecting the market?
[00:25:20] Megan Ma: Yes. So as I mentioned, I’m super excited that will be launching soon. Our rapid prototyping lab. It’s giving an opportunity for our law students as well as our CS students or engineering students to get exposed to this kind of nerdy niche of legal and AI. and the whole idea is that you are building here.
so that’s one area that’s particularly exciting for us. The other is that we do run bootcamps as well as typical classes. And in the past we’ve worked with open AI, for example, where they came in and did a tutorial and kind of helped both CS law and business students, learn to use generative AI tools and build things completely from scratch.
we also run, of course, our classic large language model and law hackathons. We do have one coming up on September 8th, and this brings in not only students, but, participants from all over the world, actually, at times, but all across particularly the Bay Area. And they get exposed to lawyers that come and mentor, as well as, you have some of the folks that are on the ground.
You know, at the forefront of building, the next generation of these models as well. and so it creates this new interdisciplinary community.
[00:26:32] Greg Lambert: Has there been any interesting outputs from the hackathons in the past?
[00:26:37] Megan Ma: Yes, so actually one of our student teams that participated in our last hackathon, they came up with an end to end tool to help veterans.
this is traditionally a process that was very arduous. It was very very Difficult and actually even demoralizing.
[00:26:51] Greg Lambert: I can verify that
[00:26:54] Megan Ma: and so one of our students is actually a veteran and you know He came to us with this use case and his team very talented Cs students put it together and they won our hackathon And actually they were recently featured on NPR.
[00:27:09] Marlene Gebauer: So yeah, they’re doing great stuff. Very cool. So I know you were saying, you know, this is kind of a nerdy little community. so, We have our nerdy technology developers and you know, sometimes they see things one way and then the legal professionals that You know, they might be working with see it another way how do you ensure that the development work aligns with the needs of legal legal professionals when They in fact may be have different goals or speak very different languages.
[00:27:42] Megan Ma: Yeah. So I always say that I think I’m like a glorified translator of sorts in my role. And it’s because to your point, I think engineers and lawyers speak very differently, not only in the specific language, I think also in their cadence of speech. And what I mean by that is lawyers love to speak an anecdote.
Because anecdotes are really where they’re kernels of wisdom and that’s where the knowledge is encased for them.
[00:28:10] Marlene Gebauer: They’re telling stories. For engineers,
[00:28:12] Megan Ma: they like see it as like bullet points and everything is like from an implementation perspective. and so when you kind of have the two together, you sometimes have this bit of a gap.
The anecdote is meant to share with you what the core of the problem is and what they’ve learned over time for lawyers, but it doesn’t necessarily translate to a two or three step process that this is the solution. And so I think a lot of the times my role is, breaking down the anecdote into something like these are next steps we should be approaching.
and it’s not going to necessarily arrive at a solution because that’s maybe not our current goal. But this is kind of our next step forward. And that type of kind of middle sliver is something that I often find like a necessary part that our nerdy community has to do. because otherwise user needs are just not clear.
not really because entirely of the language. It’s a lot of the language because we do have our own legalese speak, but I think some of it is merely just. Just the way we present ourselves is fundamentally different.
[00:29:18] Greg Lambert: Does the law school and the engineering school at Stanford have a battle every year?
Yes. Fight it out.
[00:29:31] Marlene Gebauer: And you cocktail hours is what they need together.
[00:29:34] Greg Lambert: Yes. that helps everything. So, well, Megan, we’re at the point of the show where we ask our crystal ball question. So I want you to peer into your crystal ball and look into the future here. what trends do you foresee happening within the intersection of law and technology and even Codex’s mission, say, over the next two to five years?
[00:30:04] Megan Ma: I’m super terrible with predictions, but what I’ll say is like, maybe what my hopes are for, going forward is, my goal or anticipation is that we’re finally able to build tools that tap into what I call the legal brain. I think a lot of the tools that are available right now. Are trained on the product and the legal work products of contracts, briefs, case law regulation, but those are end stages as in it’s already been done.
And as I said, with the red lines, like all the thinking and the hard work. It’s already, on the page, but it doesn’t explain, the process that got there. And what we really need in order to have genuine agentic workflows that make sense for the community is being able to tap into that process.
And what is legal process, being able to define, parameters around legal work, being able to draw better metrics and all of that. And that comes with evaluation as well. And so my hope is that within these next few years, That becomes the pivot and that becomes the focus and why we care about experiential knowledge is precisely because of this.
[00:31:16] Greg Lambert: I want to pull on a thread here for a second and see where it goes. because a lot of times, especially in the law, something has already happened. And that’s why law tends to be a lot more reactionary than proactive. Even in the counseling phase of things, you find that a lot of times our clients wait too long to reach out to us to ask for counseling, and instead they reach out to us and ask to fix a business problem that they’ve run into.
Do you think that I know you’re looking at what’s, what the finished product is and then going back and looking at the process. Is there any expectation that it could even go back further than that and then identify what were some of the behaviors that created this need in the first place?
[00:32:14] Megan Ma: Absolutely. So I think you hit the nail on the head there. Like we are very much focused on what legal behaviors have enabled us to. act on certain things. I always kind of, run on the joke that the law doesn’t ever live in the present. It’s always either from the past or it’s trying to create the future.
So with contracts, we’re defining relationships going forward. One of the reasons why we’re running multi agent simulations is because in the use case of patents, for example, a lot of it is driven by in house counsel behavior, and we wanted to understand why certain in house counsel are much more hands on when they solicit help from outside counsel and why some in house counsel are so hands off.
Revealing those behaviors and being able to play that out in a law firm simulation, so to speak, is going to help kind of identify what are these, what is the past telling us about the way that we are as lawyers.
[00:33:13] Greg Lambert: Cool. that’s where I want to go. All right. Well, Dr. Megan Ma, I want to thank you very much for coming on and talking with us today.
[00:33:23] Marlene Gebauer: Thanks so much for having me. And of course, thanks to all of you, our listeners, for taking the time to listen to the Geek Review podcast. If you enjoy the show, share it with a colleague. We’d love to hear from you, so reach out to us on LinkedIn.
[00:33:37] Greg Lambert: And Megan, we’ll put, the links on the show notes for Codex and yourself.
what’s the best way for people to reach out and find more about Codex, or your other work and, or to reach out to you if they have more questions?
[00:33:52] Megan Ma: Yeah, you can certainly reach out to me on LinkedIn, please follow Codex’s Twitter. that’s where we kind of post our latest work.
[00:33:59] Greg Lambert: Awesome.
[00:34:01] Marlene Gebauer: And as always, the music you hear is from Jerry David DeCicca. Thank you, Jerry.
[00:34:05] Greg Lambert: Thanks, Jerry. All right, Marlene, I’ll talk to you later.
[00:34:07] Marlene Gebauer: Okay, bye.