Democratizing Legal Knowledge in the Age of AI

Thank you so much. I'm really thrilled to be here. And a follow-up to that is my buddy said that Elon Musk has horrible, horrible judgment, including this one. Doesn't have good judgment at all. He...

Yeah. A 100%. Yeah.

Sure. So I am a mediocre coder. Anyone who works with me will attest to that. But really these days as I...

With vibe coding, I really don't need to be a good coder. You need to know exactly...

You know, you need to know generally how the code works and then you just tell the AI to do the things because even the best Silicon Valley coders are not reviewing code these days. They're just saying requirements and then looking at the outputs. So I would say that, yes, I am indeed a mediocre coder, and I have been since 1985. But I've always had a technical mindset that is, you know, instead of going out and playing with my friends, I'd be on my Commodore 128, and I would, you know, spend three hours to try to get pixels to the right to make a circle, and then I would rejoice that I'd made a circle in my Commodore 128.

So...

And then fast forward to, you know, the nineteen nineties where I was building websites and building out databases for my university. And then when I went to law school, I was building out CD ROMs for my professors' books to be able to have interactive CD ROMs. So throughout my entire career, I've always had the tech plus law. And I will say that I am maybe one of the worst coders on the planet.

But even, like, before I was in law school, I worked for a judge. I worked for, the United States District Court for the District of North Dakota. Part of that was building out their database of all the judicial opinions that had been scanned. I had to be scanned.

So I scanned the judicial opinions, created a database that could be searched, and go back to the beginning of time. And so while I was doing that, was thinking about going to law school, and I talked to one of the judges that I was working for, and I said, I think I wanna work for Thomson Reuters. Guess at that point, West Publishing, which was...

You know, I was in North Dakota. That was in Minnesota just down the street. And that judge said, you know, if you do legal tech, that's gonna be the worst decision you've ever made in your career because nobody respects legal tech. You know, you...

Be a lawyer and be a good lawyer and do not go down the legal tech route. So that was in 1998. So I did. You know, I graduated from law school.

I worked for a state appellate judge and a federal district court judge. I worked for a great law firm, Robins Kaplan. But always, I'd been gnawing at me saying that this legal technology, law could be better. That is all of the drudgery that I was doing as a lawyer.

I thought this could be made better. So in 2015, that's when I pitched Thomson Reuters. I said, here's legal tech that I think can change the world. You should build it and hire me, they were dumb enough to do that.

So that in 2015 is how I started my legal tech journey.

Sure. So, yeah. The democratization is something that, you know, my friend Daza Greenwood out of MIT, I guess now Stanford, he talks about when the courts decided to outsource their publishing of their judicial opinions to Thomson Reuters and LexisNexis, he said that that's the original sin of the legal profession. At that point, that law was not democratized, but instead put behind a paywall.

So anyway, so that's kind of the backdrop. So when I went to go into legal tech, I thought, well, I wanna go to a company that has the corpus of the cases, the statutes, and the regulations. At that point, I thought I have two choices. I could work for Thomson Reuters or I could work for LexisNexis.

And, of course, Thomson Reuters was down the street. So that's when I, you know, decided with that. But then, of course, there was a third option that I didn't know about at the time, but that's Fastcase. That is in 1999, Ed Walters and Phil Rosenthal were associates at Covington, big law firm.

And they said, why are there only two options for this? We should democratize the law. We should have a third option where we could have, you know, solo small lawyers don't have to pay Westlaw bills or Lexis bills to do this. So they, in 1999, started democratizing the law.

And then fast forward to, you know, CaseText before they got acquired from Thomson Reuters. They got all their data from Fastcase. So Fastcase was a distributor of this law democratizing into CaseText, and, of course, CaseText then sold to Topps and Reuters for $615,000,000. So this whole idea of, you know, the law should be free.

And because we as lawyers know that there's a saying that the ignorance of the law is no excuse. You're bound by the law whether you know it or not. But then what's the excuse of locking that same wall behind a paywall?

We should truly democratize the law because it binds every one of us. So anyways, so I've always had this kind of democratization of the law. And then there's also the democratization of the code. That is, you know, the West Publishing had over a 100, they had the West Key Number system and still have the West Key Number system.

And I, as the lawyer, loved the West Key Number system because it's been classified. All the cases were saying, you know, if I have a negligence case, it's tagged with negligence. And if it's got this element of negligence, it's tagged with that. So I thought this is beautiful.

This is amazing. So while I was at Thomson Reuters, I thought this is one of our key jewels. And I said, you know, there should be something like this, but in the public domain. So it's not locked behind a paywall.

So I, from scratch, I did not copy anything. I, from scratch, built from first principles, essentially, what are all the causes of action? And the way that I did that is going through the jury instructions for California, New York, Texas, Florida, Illinois, and a whole bunch of others. And I collected all the causes of action and put that into SALI.

And I built SALI up from 1,000 items to now 18,000 things, everything that matters to the business of law and to the substance of law. And we made it all free and open source under the MIT license. So all you have to do is to be able to give attribution to SALI, and you can fork it and do whatever you'd like with it. So because it was free and open source, that was the only way that we got a whole bunch of people to adopt it.

Thomson Reuters is using SALI. LexisNexis is using SALI. iManage, NetDocuments, Microsoft in house, Intel in house, and all the biggest law firms in the world are all coalescing around the SALI standard only because it's free and open source. If it were a restrictive license, nobody would adopt it because they would say, well, I don't wanna have a liability of SALI suing me if I do something wrong.

So it's only because it's MIT licensed, only because it's free can it be adopted. And so really that's that democratization of the taxonomy, SALI, was thing number one. And then when I started doing FOLIO, FOLIO is a fork of SALI. And so we essentially took the MIT license.

We now give attribution to SALI, but we're...

Now we're building on the FOLIO side. And what we're doing on...

Is not only building out more of the taxonomy, we're also building the tooling that goes along with the taxonomy. So since January, I've been vibe coding like crazy that...

Where you can upload a document into a tool I built called FOLIO Enrich, and it will tag up that this is a motion to dismiss for breach of contract in the Southern District Of New York. And it will tag up this is a merger agreement with a force majeure clause in China. That all of those are now tagged up programmatically using a thing called FOLIO Enrich, and that's all free and open source, MIT license. I also built another tool that's free and open source called FolioMapper, where so often I worked with law firms that said, hey.

Can you map our internal crappy taxonomy to SALI's taxonomy? It would take me a weekend to do that. So I built a tool where you just upload your Excel spreadsheet of your crappy taxonomy, and it will map it to SALI FOLIO's taxonomy. I'm building this out because I want it to be distributed.

Because I think in the age of vibe coding, I think there's really not gonna be much value to software as a service or to software generally. If I can vibe...

And I literally vibe coded each of those things that I just told you in a weekend, then it would have taken me, you know, a dozen coders and a year at Thomson Reuters because it would have cost me a lot of money at Thomson Reuters. You would have to recoup that investment in the old...

But in the new world, it doesn't cost millions of dollars. I did it on a weekend. Therefore, I can give it away free and open source. So I think that's gonna be the way the world...

Free and open source is going to be the way of the world. And I think what we as humans are gonna be able to charge for is services on top of that free and open source software. So if you want me to implement FOLIO for you, my time is scarce, therefore valuable. You can pay for my time, but don't pay me for the software or the taxonomy.

Yeah. So, SALI is the standards advancement for the legal industry. So this is something that was, founded in 2017. The whole idea back in the day is that we need interoperability between company number one, company number two, and company number 20. For that interoperability, you need to put a unique identifier on every single concept, Like motion to dismiss should have a unique identifier. The goal in 2017 is that if I run an API request to Thomson Reuters and send that unique identifier, Thomson Reuters will send me all of my motions to dismiss and then change that to Clio, that same identifier, that same API call to Clio, and I'll send you all of my motions to dismiss. So the idea is that the concept of motion to dismiss should have a unique identifier even though in California, a motion to dismiss is called a demur.

So what SALI does and the way we built it is on something called RDFS OWL. And through that, you can say, yes, it's a motion to dismiss, but a synonym in California is called demur. And some jurisdictions call it motion to terminate. But all of them have the same unique identifier that is this unique identifier for the SALI identifier.

So, anyway, so SALI is an interoperability standard where you're taking all of the concepts that matter to the substance of law, motion to dismiss and a merger agreement, and to the business of law, flat fee billing, hourly billing, etcetera. So we have 18,000 of those with...

We did with SALI, and now we've worked it to create FOLIO. FOLIO is the openlegalstandard.org, and it's open legal information ontology. Right.

It's a it's a acronym. So FOLIO...

And the joke on that is that, you know, you have a portfolio of different taxonomies. That is all of the law firms and legal tech companies and in house counsel all have their own taxonomies. So FOLIO is essentially a portfolio that can be able to make those all interoperable.

And that's exactly right. Yeah. And also that's number one, the law should be readily available. And number two, the tagging up of the law should also be readily available. Because people are making a lot of money going into law firms and being able to say, I'll tag up all of your contracts or I'll tag up all of your litigation documents with motion to dismiss for breach of contract and I'll charge you a lot of money for that. What essentially Fully Enriched does is makes that free and open source.

But you worked at vLex where one of the products of vLex was Docket Alarm and Docket Alarm is a paid subscription service that you have to paid in order to gain access to publicly accessible docs and documents through local county as well as federal courts. So the irony, though, is that you were making money off of the ability to charge people and organizes the public information, but now you're giving away all the public information.

Well, so the public information, there's two types of information. One is the structure of the data that a SALI FOLIO is, I'm giving away for free. But then there's the data itself. That is the motion to dismiss from the Southern District of New York for breach of contract.

Those are two distinct things. Right? Number one, the taxonomy is free and open source because we made it open source. And then number two, the tooling to implement that taxonomy is free and open source.

But the data itself, that is the motion, the PDF that represents that motion dismiss, that data is scarce, therefore valuable. And that's what Clio Dockets now, formerly Docket Alarm, has.

I think the other thing, Daniel, is that the intelligence layer is the technology. And the way in which Doc Alarm and Vincent work is that it... It's almost like the data is just the data. Right? It's the how the technology interrogates the data through algorithms and AI that really make that product, I think, special.

But also that's the commercial aspect of the product as opposed to saying like, oh, we'll just...

We're gonna buy for a subscription. It's not just about getting access to the data. It's more about what's the intelligence layer, what's the AI do, what are the algorithms that I can use to build and interrogate and build a better position for my client? That to me is like that fourth variable that...

So an algorithm is essentially a sorting function. You wanna be able to, you know, sort things. And so people are familiar with social media, and the algorithm for social media says you probably wanna see this next post from this person because that's something that you'll probably like. So to be able to...

You know, Facebook or Twitter or whatever social media is LinkedIn, to be able to figure that out, they have an algorithm that is saying, okay, we're gonna sort all of the posts and we're gonna filter it to just your friends, and then we're gonna filter just the posts from your friends that you think it was most likely. So think of algorithm largely as just a sorting function that I wanna be able to say, if this person, then this is the next thing. So that's an algorithm. But, of course, also, large language models, in a sense, are algorithms, where they're predicting statistically likely next token, whether that token be a word or a pixel or something else.

And so, really, you know, algorithms rule the world. And, you know, one of the algorithms that we're using for Docket Alarm is something that your...

Since it's audio, you're not able to see, but I'm sharing on my screen right now, is the algorithm is saying, well, what are my odds of winning summary judgment in all the courts? And it says, well, the odds of winning in all the courts are 54% based on 87,000 known So that's the data. But then if you jump to this court, say I'm in Minnesota, the odds drop to 46%. But for this judge, the odds do go up to 57.

But for trademark cases, the odds are about one in three. Now if you were to ask that question, what are the odds of winning summary judgment of a human? That human would say, pretty good or not bad. And if you ask the AI, the AI would hallucinate all over the place.

But because we have in the federal courts you know, 87,000 known outcomes from summary judgment, we can be able to put statistical likelihood on something that, you know, lawyers have as annex data. And the way that we were able to tag up this motion for summary judgment is that we tagged up the fact that that motion for summary judgment has been granted in part and denied in part. Then once you tag that up as a partial motion to dismiss or granted and denied, the algorithm could be able to then say, okay, I want Susan Richard Nelson to grant my motion for summary judgment. I So can filter to be able to say, here are the 167 cases where this judge, my judge, granted summary judgment.

And then you can filter, again, an algorithm, say, I just care about contract cases. I don't care about all the other cases. And then you can say, okay, of those a 167, here are the 55 where she granted summary judgment for contract cases. And now I could take the results of these 55, put the text on there, and say, here are my client's facts.

Draft me a new motion for summary judgment that is statistically likely to win for this cause of action, for this judge. I can essentially repeat my judge's language back to her. And there's an...

In psychology, there's a concept called mirroring, that you trust me more if I mirror your body language.

So essentially, we are now mirroring this judge's language and tickling this judge's brain in the way that this judge's brain loves to be tickled. That judge will say, I really like the cut of that brief's jib. I think I'm gonna rule in favor of this person, and she will have no idea why that's true. So all that is algorithmically driven. But, again, all the things I just mentioned, summary judgment granted this particular court, each of those is a SALI FOLIO tag.

That... We only got that through the tagging that then fueled the algorithm.

A 100%. And so you can imagine if I...

Option one, I say, are the odds of winning this summary judgment? The client says pretty good, and then I spend $50,000 and I lose, In this case, 62% of trademark cases lose on summary judgment? Wouldn't you rather know that in the front end to say, you know, if the odds are only about one in three, I'm not gonna spend the $50 or a $100. Right?

So point number one is that with knowledge comes power, and with data comes power. You're not relying on the lawyer's anecdote, their rifle shot of experience, but instead the whole corpus of data in the courts. And so that's why the law should be free. Because to be able to do this, to download every motion brief pleading in PACER, I would ask you what the number is, but you'll probably get it wrong.

So I'm gonna tell you it'd be about $2,000,000,000 with a b.

2,000,000,000 you'd have to pay to the courts to be able to get every motion brief pleading to be able to get data for this. And so I'm, the administrative office of US courts has appointed me part of a 12 person group to be able to say, how can we make PACER suck less? Then what they do is they go on the website and they say, should we do this drop down menu here? Should we do this?

I say, no. Don't do any of that. Make the data free. Because if you make the data free, I can be able to figure out where my buttons go and how to analyze the data.

But the fact that it's $2,000,000,000 is essentially a...

It's a travesty because we have spent a lot of that $2,000,000,000. Therefore, we have to recoup that investment to the biggest law firms in the world who pay a lot of money for it. But you know who can't afford it? It's poor people.

I say, judge, this is an access to justice issue that you created. You can't point the finger at anyone else. You are charging the $2,000,000,000 as the courts, and so you're causing different justice for rich people than poor people. Their response is, Damien, that is out of scope because that goes to court funding.

The courts are using PACER funds to fund the courts. That is...

To fund the retreats that they go on. It's paid for with PACER dollars. So this is a travesty. The judiciary should be funded by the congress so we don't have to spend $2,000,000,000.

We can do that. So that's point number one, is that only through the data can you have this going from anic data from the lawyer to the true data. Point number two is we talk a lot about AI and how in the age of AI, the most important thing that the lawyers will give is their judgment.

That's everyone's judgment. But really, if you're gonna win this motion for summary judgment, we think of that as judgment, but we can actually quantify it. This is quantifiable. And my judgment is just my anecdotal judgment.

What's better, my personal anecdotal judgment or this, for federal courts, 87,000 outcomes that I've never even looked at?

So really, I think that we have a reckoning that how much of what we call judgment as lawyers is truly quantifiable, that maybe we could put your and my judgment into an artifact. And then all of a sudden, our judgment just becomes another data point.

I'm really grateful for the questions. I've been on hundreds of podcasts. I can count on less than one hand the number of people ask me that, so thank you for the question, number one.

Casey Flaherty, who you might know, is a friend, and he said, Damien, why do you like not making money? Why do you give away so many things that have value? And so that's a variation essentially on the question you just asked because I do give away a lot of things. And it stems back that I grew up Catholic.

I went to a Catholic high school, I went to a Catholic college. There's an idea with Catholic social teaching is that you are bound as a good human to be able to give away your time, talent, treasure. That this is something that you're bound to as a good human to be able to say, you know, give to the least of these. And a lot of that reflects an access to justice.

We need the least of these to be able to...

You know, poor people should have the same justice as rich people. And if we charge a lot of money for things, they can't do that. The answer to your question is my Catholic upbringing. That kind of has come to a fruition.

In November, my buddy who's the president of the board of directors for Saint Mary's University, she said, hey. Do you wanna talk AI at the Vatican? I said, yeah. I wanna talk AI at the Vatican.

So in November, I was at the Vatican talking with a whole bunch of people, and the whole pitch to the Vatican is that, you know, Sam and Dario say that we need to align AI with human values. And the pitch to them is that you've got a lot of human values locked up in the archive. So how about you open up the archive and let all those human values be ingested by the AIs? And it turns out they're doing it.

There's a company that's scanning all the archives books and then making it public domain. And then I'm working with my favorite vibe coding priest. His name is father John Taurazio out of Rome. And what father John Taurazio and I have done is we forked FOLIO to create...

We're calling it the Catholic semantic canon, where it turns out that FOLIO has murder and theft. And it turns out that murder and theft is also biblical. And so we're essentially forking it, and the same unique identifier that FOLIO has for murder is the same unique identifier that the Catholic church is now tagging up the Vatican archives and the items from Thomas Aquinas and Thomas More. And every time they mention murder, it's gonna have the same unique identifier.

And then Pablo Arredondo, who is the cofounder of CaseText, is doing that. He's doing...

Going through all the Jewish texts and all the Sharia law texts and all the Roman Catholic texts. And he's similarly going on and being able to do a comparative analysis of all of those. So you can imagine that maybe as a result of this, you could be able to say, you know, with Vincent and...

That we're building with Clio, you could do a 50 state survey. And so you could say, you know, I'm block and I wanna lay off 40% of my workforce. Does this violate any of the 50 states? And you could also say, does it violate the EU law or The UK law?

And then what if it had the Vatican archives to say, does this violate the ethics and morals of my faith? And there's Catholic venture capitalist, he said that companies that do faith based work, love your customer, love your employees, make more money. They outperform the S and P 500. And so what if we could be able to build as the agents going out and doing things in the world, you know, as open...

OpenAI is doing things in the world, as as Anthropic is doing...

What if the agents could consult the legal oracle, consult The US law and the EU law and faith based traditions to see if what they're gonna do is ethical or immoral? If it breaks the governmental law, EU or US, or ethical laws above a certain threshold, let a human know. That is put a human in the loop for them to give the thumbs up or thumbs down. That is, I think, the end goal where we can maybe nudge toward...

You know, we have privacy by design. We have security by design. What if we have lawful by design? Where the bots are...

Have the law and therefore they can follow the law. That, think, is a good endgame that we should shoot toward.

That's a 100% right. And you're right that all the things you just mentioned are, you know, give away to others. Is Catholic and a Jewish and Sharia law and Buddhist. And so if you think about putting the idea of, you know, generosity, we'll call that a tag within FOLIO, and it turns out it is a tag within FOLIO.

What if we tag every Buddhist scripture that has that and every Jewish scripture and Sri Lanka with that same tag? Then you start and do that for 18,000 plus things. Then you start thinking about a Venn diagram where there's a convergence of all the...

All of the religions agree that murder is wrong, that theft is wrong, that you should be generous. And then, of course, there's gonna be outliers from that Venn diagram. And then you have to say, okay. Let's not let the tyranny of the majority outweigh the rights of the outliers.

So, really, this is a way that we can maybe have a more just society that aligns with human values. Because when Sam Altman says we need to align with human values, I say we've kinda had that for five hundred years. It's called the law. Because Eric's human values are different than Daniel's human values, which are different than Damien's human values.

But the way that we all align our human values is through legislation, which turns to law. And so that's...

We've had alignment with human values called the law, and we've also had alignment with human values and it's called faith. And so what if we were to tag up all the law and tag up all the faith with the same tags and then put that into the engine that says, is it violating human law or ethical faith based law? And is it moral or is it ethical? I think that's something we should be doing.

Eric, I love, love, love that question so much, and I can say that in the hundreds of podcasts I've had, I've never had that question, but I've been wanting to answer it. So thank you for asking the question. Yes. So for those who haven't seen the TED Talk, what I've done is I brute forced all the melodies.

So much like you can brute force a password by going AAABAC, I did that with music with my friend Noah Rubin. He goes, we went until we mathematically exhausted every melody that's ever been and every melody that ever can be to the tune of 471,000,000,000 melodies written to disc. And under the Berne Convention, once it's written to disc, it's copyrighted. So we've copyrighted 471,000,000,000 melodies, putting everything in the public domain to protect the You Stole My Melody lawsuits.

So to your point, Eric, what we did is we took...

Our initial dataset was the diatonic scale. So that's eight notes. And then the combinatorial permutations, the mathematics, the way it works is that you go eight, that is, and then the number of notes, you go in a string. So 10.

If you put 10 notes in a string, the mathematics is eight to the tenth power. So that is, turns out, 56,000,000,000 melodies where you have the diatonic scale with a 10 note string. And so really that was a dataset number one that we did for the TED talk. And then people are saying, actually, you didn't do jazz and classical music?

And so I said, actually, cool. So then I went to the chromatic scale and I did 13. So I did the black notes and the white notes. Then I did that.

So then other people said, well, actually, you didn't do rhythm? And my response was, actually, copyright law doesn't require rhythm. It only requires substantial similarity. But even then, we put a fourteenth note and it was called silence.

So with silence, you can then get every permutation of rhythm. So really the whole point of all the music was that you can do every permutation of all the music, every combinatorial combination of those, every combinatorial expression, because there are such a finite number of things, eight notes up. Contrast that with words. So if you have the English language, there are about a 177,000 words.

So that...

It's not eight, it's a 177,000. So then if you have a 10 word string, that's a 177,000 to the tenth power. And it turns out that there are fewer sands on earth, grains of sand on earth, than there are combinations of 10 word 177,000 combinatorial permutations. So really, music is easy because it's only eight or 13, depending on your...

How you count. But 177,000, that's a lot harder. So then you say, okay. Now let's do legal outcomes.

Then you get fuzzy. Right? So, like, for breach of contract where they said that the contract was not formed. Right?

Then you can tag up breach of contract, You can tag up formation as an element of breach of contract. But then you have to tag up, okay, what are the facts that caused this contract not to be formed? It turns out the facts in the world, there are many, many of those. So the combinatorial kind of brute forcing of that is just mathematically not sound.

But you can get there, that is get to the result that you want, by tagging up breach of contract and tagging up lack of formation and tagging up, you know, for watercraft, for example. And once you tag those things up, then you could be able to say, okay, let's just run analytics to say, you know, for this fact, how many go this way? But for that fact, how many go that way? So this is kind of a poor person's version of brute forcing.

2,000,000, actually. I think there are a bunch of versions out there.

A 100%. And not just possible, it's using good old fashioned AI, where this is a Seventh Circuit judicial opinion. And we've tagged up the Seventh Circuit, which is a court. And we've tagged up the fact that Maria Gonzalez is an appellant who is the plaintiff.

And we've tagged up the fact that excessive force claim is a claim in here, and it's sometimes known as police brutality and unreasonable force. But they have the same unique identifier regardless of the words that you call it. You can imagine taking up this excessive force claim and then taking the text of this and then saying that we reverse and we remand, also SALI FOLIO tags, then now you could be able to say, yes. Under these facts, this excessive force claim is denied.

Therefore, they reverse and remand. Whereas the under different facts, it's granted. And then what SALI FOLIO also do, that is FOLIO Enrich does, is then puts those into a knowledge graph. So we have officer Taylor used a taser.

So noun verb noun is a triple. Officer Taylor is a noun, use is a verb, taser is a noun. And so noun, verb, noun is the foundation of knowledge graphs. And so you can actually build a knowledge graph that is symbolic AI to be able to say that if you use a taser on someone, then excessive force under this judicial opinion.

But under another judicial opinion, it is not excessive force in a different circuit. So this is maybe on our way to computable law. And when you think about, well, can't AI that is large language models do that? And the answer is not efficiently.

So for example, maritime negligence is a claim. That is a type of negligence claim. It is also a type of admiralty maritime claim that is related to admiralty court and related to the area of law that is admiralty maritime law and related to the industry, that is water transportation industry, and related to the asset, that is watercraft. So you can imagine if the AI were to say, give me whether it's this law is broken regarding watercraft, then the AI could follow the graph that I just showed you and be able to symbolically, therefore deterministically, therefore reliably go from watercraft to maritime negligence and say, maybe you wanna make a maritime negligence claim. This is the answer is to be able to say, let's not rely on the probabilistic, therefore unreliable, large language models. Let's make this all deterministic, tag it, and then make a graph so that it will be a 100% reliable.

So that's a ontological idea. Right? If you have this ontology basically charting a path through an LLM to say...

And it's an ontology based on noun verb noun or it's an...

Or type. Right? It's...

I love that question. And again, this is the first time I've been asked. So as a musician, I think a lot about creativity, especially somebody who is arguably brute forced 471,000,000,000 melodies, all the melodies that have been. And Sting, in an interview, said that if he's if he's not surprised in the first ten seconds, he skips to the next song.

And so there's an argument that creativity is statistical unlikely. Because if it's statistically likely, Sting wouldn't be surprised. Right? So really, if you think about the distribution curve of likely or not likely, the more unlikely something is, that's more likely to catch someone's ear to say, wow, that's really creative.

There's probably a mathematical representation of, like, straight on the nose, that's too predictable. But if it's too far out there, like there's a lot of melodies in my data set that go, that's just too weird, right? So nobody would call that creative. They just call that noise.

So there's probably a statistical mean in between the non the nose, statistically likely, and totally so out there that nobody can really even listen to it. There's probably somewhere in the middle that is quote unquote creative. So that's true for melodies. It's also true for legal arguments.

That is the statistically likely argument that always wins, I guess, is gonna win, but it's not creative. Everybody argues it. Right? But it's truly the Brown versus Board of Education, something that is a statistically unlikely argument that actually won the day.

That is where the great lawyering happens. So if...

I used to work at Robins Kaplan. And when Robins Kaplan had a tobacco case back in the nineteen nineties, for twenty five years, the tobacco companies have successfully argued that the doctors' memos saying that, hey. We're causing cancer. The lawyers had argued that, no.

That is attorney client privilege and work product. Therefore, it cannot be disclosed. They had argued that for twenty five years. So for twenty five years, that was statistically likely.

But then Mike Ciresi from Robins Kaplan successfully argued before a Ramsey County judge.

Those are not a statistically...

That is not privilege. That should be...

Those doctors' memos should be public, and he won with that statistically unlikely, otherwise known as creative argument. And so I think you're right, Eric, that that a lot of what we as lawyers do is not a creative. That is we're just making the workday arguments that are gonna win the day. But the best of the lawyers, the Mike Ciresis of the world, are able to make that statistically unlikely, therefore, creative argument that wins the day.

And then the real question is, in the distribution curve of legal work, that is all the work that we bill our clients for every day, along that distribution curve, how much of the work that we have to do is truly creative versus statistically likely? And with that, I've got one more story. My next door neighbor wants to build a fence six inches from my house. And so they said, hey.

We're gonna break ground. And I said, no. We're not gonna break ground. And so I vibe coded a bunch of versions of a complaint.

I also vibe coded a bunch of versions of a temporary restraining order. And I still brought in a lawyer. It's a good friend who's been a friend for twenty five years. His name is Jake.

And so I said to Jake, I'm gonna file the complaint, and I'm gonna file a temporary restraining order. He said, yes. File the complaint, but don't file a TRO. I said, wait.

Why wouldn't I? I've got a gun. Why wouldn't I shoot it? He said...

Jake said, well, you're...

The temporary restraining order has a high burden that you have to show irreparable harm, and they can always tear out the fence. So you might lose that motion, and then forevermore, that judge is gonna look at you as the person he ruled against in the thing. So why don't you file the complaint? And what are the odds that your neighbor is gonna build a fence in the middle of an active lawsuit?

Almost not gonna happen. So you get the result that you want without the risk. So anyway, that's something that was in Jake's brain that I, with the fifteen years of litigation experience and arguably the best AI legal tool, Vincent, on the planet, I didn't think of that. Jake did.

That is a creative argument that is not yet in the dataset. So when you think about the process, like your question about process, I said, that's why I hired Jake. Right? That's why I want Jake to do my work.

But what if Jake put that thing that was in his head into an artifact that the AI can then run through saying, if you're thinking about filing a temporary restraining order, think about just filing the complaint. Because the temporary restraining order might...

You might get the result without the TRO. If you put that into an artifact that the AI can then go through, all of a sudden, that creativity of Jake turns into another data point.

And so your question about process, I think a lot of the process gonna happen for lawyers and law firms for the next, you know, five years or so is taking the stuff from our brains as lawyers and putting it into artifacts, taking the processes, taking the insights like JCAD, and putting them into, you know, documents, whether they're markdown documents or otherwise, that the AIs can be able to go through to be able to, you know, make processized our... Is previously called our creativity.

Thank you so much for having me. I'm a big fan. And sometimes people listen to podcasts at 1.5 speed, but they say that listening to mine, it's point seven. So if you're gonna listen to it again, listen at point seven.

So...... 1.5 anyway.

Yeah. Thank you for having me.