I had the honor of speaking with Joshua Walker. Joshua Walker is the author of “On Legal AI”, perhaps the first systematic and practical treatment of this topic. With over 20 years of experience in this and allied fields, as well as over 17 as an IP attorney, he is seeking to deploy the next next generation of advanced legal solutions. Walker is a co-founder of CodeX, as well as Lex Machina. For the latter, he led the successful spin off from Stanford Law School, serving as CEO and Chief Legal Architect. Currently, he is the Chief Product Officer for a large risk management company. All thoughts herein are his own and not those of any particular party.
[THE BELOW HAS BEEN EDITED FROM THE AUDIO INTERVIEW FOR GRAMMAR AND TEXTUAL COHESION]
What got you first interested in all of this kind of legal artificial intelligence and in technology?
First, just a caveat: The book, On Legal AI, gives a somewhat more coherent narrative but I’ll give it a shot. You ask some great questions!
For me, it started when I was working in Africa and just trying to build a database without resources. That was the beginning in terms of “work” focus. Really though, it was my Dad. My Dad, when he was I believe about 50 decided to create a business. He’d never done anything with computers before but he created a database from scratch, and using that database he collected, corrected, and normalized a vast amount of hotel data that could, with a substantial degree of confidence, hotel revenues based upon certain vectors. He predicted what a hotel was going to do depending on its brand, where it was built, how much was spent per room, and other factors. It’s a lot more complicated than that, what he had to do to make it work, but it was a predictive modeling engine for a hotel’s revenue.
Many will have had different experience of the creation of Lex Machina. But for me Lex Machina was essentially the same thing. We were looking at a (semi-)discrete set of data elements (which we had to resolve and extract through a massive amount of work, technology, and the combination of the two) so that a user (a) get a handle on what was happening in this case and (b) what was likely to happen based on past experience (with all appropriate caveats; law is and remains unpredictable). It’s the same essential story about turning raw data into a usable dashboard and model that, in turn, makes the entire system better.
The ultimate goes is to help people.
In theory that is the goal of every lawyer but we have a problem. We’re pretty impractical.
I loved law school I felt like it made me smarter; and I’m so grateful for the experiences and the teachers I had. But I also think the structure of it is kind of backwards. If you go to a teaching hospital, every single one of those doctors that teaches also practices medicine or they’re a bio person that’s doing complex scientific studies, right? No one is doing something that’s purely academic. Even the deep science has a practical underlying goal, generally driven by helping people and conquering disease . . . improving human health. There is a role for deep research, but it generally has to be applied. And the doctors that teach you practice.
If the legal academy is not helping people, if it’s just about, “I think the rule should be this as opposed to that,” or, “This is illogical,” I have an issue with that because I think the goal of the law is to help people. The law is not an end in and of itself.
Even Gottfried Leibniz—the originator of modern binary notation and co-inventor of calculus—started out as a lawyer trying to solve practical problems. He was trying to solve legal and diplomatic disputes. Leveraging and/or creating binary code was a means to an end. He was trying to invent a “concepts” language in order to enable “mechanistic” or semi-automated resolution of disputes. A bit too ambitious? Naturally. But he had a very defined goal and that drove his success—even if his “success” was not at his original goal but in becoming one of the incidental parents of computer science.
As an aside: Leibniz was working on this stuff and for some reason the law has completely lost track of that; and to some extent so has computer science. There is a natural affinity between computer science and law that we haven’t been exploring for 300 years too long.
I think there is a massive amount to be gained from merging the liberal arts with the sciences and math. I think they’re different, I think they both have their own values and that can get lost sometimes these days in the sprint for STEM. I think combining those two things and using both domains is incredibly powerful. As a lawyer you can really do something to help someone solve a dispute or an issue. In the computer science world, though, whatever you do, if you can automate it you can do it times a million, right? Because of the power of software: Write once. Execute a million plus times, for limited or no marginal cost per operation. Adding software or software like economies of scale to the core of what we do, while maintaining and even enhancing the core powers and the core values of the legal profession, I think that’s our JOB as attorneys and I think we’ve generally lost track of that.
I think we’ve lost track of it in the academy, in law firms, and in general. How do you help people at scale?
And so there has been a little bit of a sea change, I think, in how people think about this but it’s only at the very beginning, right? We haven’t convinced most lawyers to use basic technology or things in the way that they could—not to replace law, but to scale it and to improve access to what we’re already doing.
How do you respond to those who say “Well, we’ve just been doing this this way for a while and it’s financially lucrative for us, so why should we bother changing how we do this?”
So with regards to the matter of regulation, I wouldn’t burn everything down. I don’t think we need a revolution. There’s a lot of great things about the profession; but I think its conservative (with a small “c”) operational structure really hurts a lot of people too. The people that suffer are mostly the under-resourced, but it’s really ALL consumers of legal services (or all people that NEED legal services, which is a much larger set). We are as a group failing all of them—compared to what we could be doing with smart application of modern technology and operational methods. It actually causes immense harm when the legal profession isn’t working efficiently or reasonably, given the size and scale that we are working at.
US Law firms are generally LLPs (or similar corporate forms), so every dollar each one spends pro rata comes out of the other partners’ pockets, right? So one partner has to give up X number of cents to pay for that bit of technology investment. That’s why law firms haven’t historically been great R&D investors. They are doing it, right? It does happen and people can find ways to innovate regardless of structural constraints. They can have different organizations, contracts, and risk allocation tools; but it’s much harder for Wachtell to innovate or Skadden, or anyone, even these big firms, than, say, Google or Amazon. It’s very easy for Google to innovate because they build one piece of software and it’s used by X hundred million people pretty much overnight. We don’t write briefs (or brief development tools) with the same level of scalability.
That is the reality. The reasons for these organizational constraints include making sure that we’re doing the right thing for our client and that we don’t do crazy stuff, right? Well, the problem is it has actually crippled us in terms of accessing capital for scalable legal solutions; and that is a fundamental problem. Eliminating or watering down ethics rules is not an answer. Rather, I recommend adapting them to deal with research, access to capital, and the greater ethical need to develop scale; because if you cannot scale legal services you are (a) going to shrink the profession (relative to the overall economy), ceding adjacent services like contract review and extraction to non-lawyers and, far worse, (b) crippling the profession’s ability to meet the vast majority of legal need.
Here are two examples of why use of modern data tools is an enhancement—and not a threat—to traditional legal practice. When I was with Lex Machina we were selling software as a service. My first sale was in 2009, at the bottom of the NASDAQ rout, an economic nadir. I remember pitching the service right after that to a very senior partner at a big law firm. He was an old school litigator. He’d been around for a while, done some of the biggest lawsuits and arbitrations in history, was famous guy. And he clearly had less than zero interest in or need for technology. So I just quote Churchill (who was [paraphrasing] Santayana): “Those who do not know their history are doomed to repeat it.”
To wit: If you don’t account for all the past actions taken by this particular judge over this particular issue, that’s crazy.
It’s not about technology at all. It’s about knowing the field you are practicing in. It’s basic competence—especially now that more and more data are readily available. Not every attorney wants to learn about “technology” or computer science, per se, but they all want to perform well relative to their peers.
I can’t imagine being a litigator right now and not using empirical data, and it has nothing to do with technology. You’ve got to learn your history. You have to know what works and what doesn’t, and if you introduce something new, do it but know what you’re getting into, and make your client aware of the fact that this one judge only grants this type of motion with this type of argument, 1 out of 100 times in the past, this has been the result. We can try to get a different result but it’s risky.
And judges can change their minds, so you don’t sell the data as a prediction, but as history.
I forget which philosopher said this, but: You have to teach the new through the lens of the old—through the lens of what people already understand.
For me, a classic liberal arts nerd, that meant learning math through reading history. For example, I learned a lot about risk through the book “Against the Gods: The Remarkable Story of Risk” by Peter L. Bernstein. I learned a lot of math through the book and its stories of how folks he writes about came about and did the things that they did.
Conversely, if I’m talking to a CFO, or a head of engineering, or a very engineering-focused executive, they don’t care about what the law says. They care about the data. So you have to give them a spreadsheet which says that: “If you don’t do these things you could go to jail”; or, “If you don’t spend this money on this type of license you’ll probably get two or three lawsuits and here’s the probability of that.” You have to convert your qualitative stuff into quantitative language, and it’s the same for any tech lawyer. You don’t tell them to do something new. Instead you say, “To keep doing what you’re already doing (and improve on that) you have to X, Y, Z (clear actions summarized clearly in their language); or, “Here is something more of what you already love doing.”
The other thing that helped us sell and survive in the past recession was that law firms said, “If my competitor has this, then I must have it too.” And in the case of Lex Machina, not only did other law firms, e.g. your competitors use it, but so did some judges and so did counterparties. If the judges use it you have to use it or something else like it that is an equivalent. At the time, no one else had done as much work as we had done, nor had any been as open to criticism and brutally focused on improving our data until it was really as accurate as it could reasonably be, with our resources under the contemporaneous circumstances.
We did well because we worked hard at staying humble and improving our product, listening to our customers of all kinds. But the other reason that it took off was competition. If a law firm’s competitors were using a competitive advantage, they had to as well. It’s a mixed attribute, but attorneys tend to care about status. So, if another law firm or another lawyer is using something and getting a better result than you, your firm is going to want you to dig into the same pool of resources.
You can’t just say it’s some new technology and it’s going to help you. They’ve heard it before. Most people that say that are lying. They’re not actually going to help them; so you have to be really crystal clear about how this is going to help them and do it in their language. And, second, it’s competition that will drive adoption. It’s how these other law firms and other lawyers are going to get ahead of you unless you evolve and adapt, and that’s not just marketing. I think it’s true. Litigation data is a sales tool, along with everything else. Law firms use it to define strategy, quickly, for pitches—regardless of how unfamiliar or unfamiliar the jurisdiction is.
How would you define artificial intelligence?
Artificial intelligence has become a marketing meme, but it is a term that at its core relates to models which then give rise to algorithms, which algorithms are addressed to aspects of thinking, perception, and action. (Thanks to Professor Patrick Winston of MIT for this definition, aside from the marketing bit. Above is an imperfect paraphrasing.)
I think people are massively overusing this term. However, the hype ratio for AI is not as high as Bitcoin was at its height. The reality of AI is it is actually addressing myriad successful projects, but that reality is much murkier and much harder to realize than people give it credit for. It has so much to do with the data. We say in a new book that, “Data is the mother of AI and math is the father.” So, it is really about applied mathematics operating on datasets. And most raw datasets are a total entropic mess. So, before you do any functional algorithm or model building, you have to clean the data.
And if you want to turn that AI potential into law: We are talking applied math working across historical events that are essentially legal /highly complex qualitative events. That definition in my view is probably the number one thing to keep in mind. But, in general, AI is a marketing meme and you should ignore it and really think instead about data and the mathematics. If you can do that, all of a sudden it becomes traceable.
The other thing that happens when you define AI applications as data and mathematical models on top of those data, is that the data really matter in terms of the ultimate identity of the finished product. Every legal AI “thing” is like a fingerprint. They’re not the same at all even when they perform the same function, and even when they perform the same function with similar results.
What has what has been the biggest challenge in Legal AI that you have had to overcome and how did you overcome it?
Lex Machina took a lot of miracles. There were many, many miracles required to get that project going in the right direction and to help it scale. At the time engineers wouldn’t touch law — and ESPECIALLY not patent litigation — with a ten-foot pole, so there were a lot of things that had to happen (necessary conditions).
Then, institutionally, you can imagine that most law firms and law schools at that time would have killed a project like that. It took a pretty forward-looking law school to do this work. This includes individuals like Professors Mark Lemley (whose brainchild, the IP Litigation Clearinghouse gave rise to the project, and who was the primary reason so many donated to the project), Josh Becker, and Dean Larry Kramer.
What we did with the spin-off is we made it very, very altruistic. The charter of the company when it spun out required it to give—it was mandated—free access to the most advanced data tools to judges, academics, the press, other government officials, nonprofit attorneys, and others. That was a big deal. It took a lot of miracles to get that thing out the door to the market, but it was such a joy to do, even in the most difficult times. I just feel very, very fortunate; not least for the fact that it is still running, and scaling across new data sets—through the blood, sweat, and art of a lot of people who are not me—particularly my successor CEO Josh Becker, who is a titan. People are still using it every day. And it is growing, precisely as it was architected to do. In fact, my current company uses it for purposes (IP risk management) architected in well over a decade ago.
I currently work for an insurance company that uses it. It’s the core thing that helps us create products around legal risk. It is helping a lot of people and we’re only at the beginning of that work. There are a dozen or more moments where something could have killed this work, but we just found a way through it and persisted. It would take a few days to tell the full story!
The hardest parts, I think, of legal innovation right now are innovation versus greed versus public spiritedness versus lack of resources. Managing how you deal with all those different motivations, motivation for competition, and still do great things, I think that’s still the biggest challenge. And, unfortunately some of the most important projects are the least resourced. I think that’s going to be an ongoing challenge.
Now, in the time of COVID19, where we are at the conflux of both professional and popular fears, is the time when we need that rarest of resources: courage.