data.bythebay.io: Daniel Lewis, Creating Value by Turning Law Into Data
Recording: data.bythebay.io: Daniel Lewis, Creating Value by Turning Law Into Data
thanks thanks so maybe the outline for this talk could be something along these lines which is law is America's operating system it's this fabric that ties individuals and companies together affecting everybody's actions and affecting folks around the world too so the way that we think of it as Ravel is that that operating system needs a major upgrade it needs a new interface it needs a new way of looking at this information and extracting value from it and helping practitioners and general citizens interact with it and the way that we're working with it is is from a very um lawyer Centric perspective I as you mentioned in the intro I'm a graduate of Stanford law school I come from a family of lawyers so both parents are attorneys an older brother is a lawyer a younger brother is a seconde student at Stanford law school so there's definitely a genetic defect in my family um but it's not a genetic defect around technical expertise that's not where the idea of Ravel started it started from a perspective of from people who are interacting with the law day in and day out what are some of the challenges and working with this information and what are some of the opportunities to to take advantage of it that haven't been exploited yet the rest of our team though is a really interesting combination of lawyers engineers and designers all working together and so for this talk and thinking about how do you take Legal Information and turn it into a business turn it into a product we've felt it's really important to approach it from those three different perspectives um the data science doesn't exist in a vacuum the legal expertise doesn't exist in a vacuum and understanding how lawyers or general citizens are going to use this material or product day in and day out requires this combination that's really important but um rather difficult to put together so as a team we're building this new era of analytics um think of it as taking Legal Information and building a whole new set of interfaces and Analysis on top of it that helps people sift through millions and millions of documents to find what's important and understand why it's important and to have a datadriven perspective about how they could use Legal Information more persuasively and more effectively um the other challenge that's really worth talking about here um is where does this data actually come from how do you get legal data as the core building block of building all these other things um when we started Ravel about four years ago we looked around the world looked around the country and saw that government was becoming more transparent access to court and Legal Information was becoming more uh transparent and publicly available and machine readable formats but it wasn't totally there yet it was still a patchwork and as you look around the country to different states different cities um one of the challenges is definitely getting your hands on enough data or enough trustworthy data or a comprehensive set of data so that the things you build from it um have credibility so what we've done is actually form a partnership with Harvard Law School to digitize our entire collection of case law which covers every Federal and State Court decision going back to the beginning of time for all of those courts it's a really U comp it's really extraordinary collection rivaled maybe only by the Library of Congress uh and it includes about 45,000 books that have been sitting on Harvard shelves in their archives Gathering dust that Harvard was excited about bringing into the digital age and that we were excited about bringing into the digital age too so that we could build things on top of it so one of the places that we could could go with this talk but I won't is just to say that we're actually starting to make this data available to other people too and so if you're interested and thinking about Legal Information and Building Things From it you should be in touch with us because we have this collection now that we're making available to the world um New York and California's collection of case law are already fully digitized and live in our system and so we're we're starting to work with that so one of the the places that we started with this of thinking about how do you take a new view on legal data and what does it mean to build a business around it um was from these different perspectives uh the one that I had experienced most firsthand was this 35% of attorney time is spent on research and you can't imagine how sort of horrible those hours are it's sifting through millions and millions of documents trying to co together different pieces of information to make an argument and it's trying to figure out how a judge May rule on a certain case or in a certain certain type of topic um because of that work is so time consuming we felt there was a real opportunity and there was an opportunity not just because it was timec consuming because it was really unloved work and young attorneys who have to do this kind of thing are the unhappiest workers in America uh according to Forbes which is the the group that was right after that sort of second less unhappy was call center workers so that kind of puts it in perspective how unhappy these lawyers are um and then as we we started interact with firms and our customers we learned this additional perspective which is that they're writing off about $60,000 a year in attorney time that's being spent on Research that their clients won't pay for so that adds up when you're looking at even a relatively small firm to millions and millions of dollars a year and the optimistic piece of this though was yeah sorry do you know a population Junior attorney inate the population the population um I don't have an exact number but it's probably 75 to 100,000 something like that to yeah um so the the those are some of the the downers about this space and some of the things that make people interested in new products in the legal world but the optimistic perspective too is that there's uh the ability to take technology and turn this into something more powerful not just to save time but to give people better insights about how the law works and how they can take advantage of it so a lot of what we're doing has relied on technology and I'm not the best person to speak about it the person who is is Jeremy Corbett sitting right here um who will be giving a talk at 110 which you should definitely go to and so he'll talk about the stack that we're using and how we're taking millions of documents and processing them and extracting and classifying information but if you think about what we're trying to do the changes that have happened in the last 10 years of the ability to use the cloud ability to use Technologies like spark the developments in machine learning um even just 10 years ago we wouldn't have been able to do what we're doing today and so a lot of it is is pushing the envelope pretty hard um but it's sort of our belief that when you take machine learning and apply it in a very specific way to a specific challenge uh you can get really good powerful results and so we're not trying to use it in a a Watson type way we're trying to tailor it in a very supervised um trained way to specific tasks and I can give you some examples of that the the other thing that resonated with us was combining data science with how do you actually present that information how do you take the insights that you can gather and make them accessible to whatever the audience is and in our case lawyers are not the most technologically savvy people so the thing that got us interested in how you access this information is the idea that search is really still in its infancy and the 10 Blue Links that you get from Google Google would be the first to tell you that that's not the right approach for every kind of research problem um and so here are some examples of the nuances and search that you've seen everywhere else and shopping and finding a restaurant and searching for flights and the unique challenge in the legal world was how do you take 20 or 30 or 40 different pieces of information that may be scattered across different courts that may be scattered across 50 or 100 years worth of history and tie them together and take bits and pieces of each of them so you finally have an answer and so we use a lot of data visualization as a tool to take statistics to take Network relationships to take insights and try to turn it into something that's really easily digestable and and really actionable the other unique thing about the law is that often times it's talked about as an art not a science and you hear that in law school almost from day one U being a lawyer is an art not a science and that's true but it's not the entire truth so there are elements that it's a it's a professional skill and it requires expertise but um for a long time lawyers have not recognize the opportunities to take advantage of technology in new ways and to take a bit more data driven and a more scientific approach and we have confidence that that's changing in legal profession because because of what we've seen in some other industries that had that same mentality that what they do is an art not a science um so two of the really interesting examples are pro sports where you had this happen in baseball um where people who believed that you could only evaluate talent with human skills um were quickly sort of in the 2000s um replaced or augmented by an approach that used data to look at Players develop new game strategies and in the early days it was a really fierce battle between the people who believed you could only do it with human eyes and the people who believed you should use it with data um but today that debate is no longer really a debate it's been resolved and the answer is you need to strike the right balance between those two things so what we're trying to do in the law is is strike that balance and say how do you use technology to augment the skills of lawyers or of General citizens to understand information not to provide a blackb black blackbox answer that uh isn't really reportable on you can't see the underlying rationale and the other interesting place that this has obviously happened is in political campaigns where for many many years they were driven by pundits who would stick their finger in the air and take a rather unscientific approach to understanding how people would vote and how to get out the vote and most observers would credit the Obama campaign's use of technology for their win in 2012 and so these are you know lawyers are often times big Baseball fans and they're political junkies so these are things that resonate with them too which is good maybe the uh the one one Cel I want to talk about here and and maybe you can think about as a takeaway from this talk um would be that we're really entering in the law a new era of asymmetric analytics and you may have heard of asymmetric information it's a economic idea that when two people have different information one side ends up with an advantage and the example would be you know used car salespeople they have information about a car that the buyer doesn't have and that creates Arbitrage opportunities it creates all sorts of things and for a long time in the law that was the case you'd have differences in the information available to people whether it hadn't existed in digital format yet and they went out and found it or they had different tools what we're seeing as government has made information more available and we're sort of part of that effort is that you have a a leveling of the playing field when it comes to data and now the advantages are to be found through the analytics and the interfaces on top of that and so those who have the best of those tools will have the best Advantage so this is what we mean when we talk about using Ai and machine learning and NLP on on these documents it's taking a case like Citizens United and running it through a system where we're starting to extract and classify bits and pieces of it so it's identifying the lawyers and the companies and the judge that may be involved it's identifying citations to other documents it's classifying language of whether the case as a whole is about a certain motion type and what happened in that motion whether it was granted or denied whether the case was ultimately later reversed and we do each of these bits with their own training data so we have a had a team of attorneys and linguists on site um internal at the company creating this training data uh sometimes for certain of these tasks hundreds of thousands of examples and and we'll create it we'll train it we'll look at the results and we'll we'll tweak and continue iterating so it's a it's sort of a continuous process of improvement until we're reaching scores that give us confidence that they're sort of as good as humans would be able to do it or because we're doing something so new that nobody's ever done before uh you know you could be 85% successful and it's still adding a lot of value here's something um really specific um which is a challenging task in the law and it's an interesting problem that's really important for lawyers which is understanding when a case has been overturned before um so one of the famous examples is plusy v Ferguson which said you know separate facilities are equal and it sort of embodied racial discrimination for a long time and so when Brown verse Bor came along that case overturned that decision and said separate is not equal and so the example of how you would do that with technology is to spot those relationships between the cases and you may often use actually third cases that reference that relationship so you'd have a third case observing what happened between brown V board and plusy v Ferguson and you can use that understanding and that observation to make a determination of whether plus Ferguson was overturned or not you can also use information contained in brownberry board itself to make a determination so you have to approach it from a variety of different angles you have to deal with the nuances of how cases may talk about each other and sometimes they'll use really clear-cut language saying we reverse or we overturn and sometimes they won't sometimes they'll say we distinguish or we criticize or um this case is very different so these are the kinds of problems that are really interesting and that we're working on that are are challenging and they don't get solved overnight they require lots and lots of training data and iteration and they require working with end users and customers to figure out what makes them comfortable that the technology is doing the job well here are a couple different examples of things that we've done so here's an example of the search visualizations that we build when you run a search to take a traditional list but combine it with a citation Network to show you relationships here's a here's one of the the notes that we got that um was really sort of inspiring from an attorney who was using our system and wanted to renew and he had embedded in his renew will notice an email that was a scan of a document he'd filed in court that used one of these visualizations to make his argument saying that the law had changed and here's how the visualization helped explain that so it's been it's been really rewarding to see these applications out in the field with you know Ravel being mentioned in case filings now and being used by lawyers in practice here's a an example of a product that we've built that takes a a case and breaks it down into a page by Page understanding looking at how each section of the document has been talked about by later decisions and highlighting the most relevant language that you can have a quick understanding of what's important about it the last thing I maybe to leave you with a story is about um how some of this is new and how some of this is very old so there's examples of lawyers doing this kind of in-depth analysis and spending hours and hours doing research to find an answer and the story I like most is about Lynden Johnson from his Senate campaign in 19 48 he was running for the senate in Texas and he got caught up in accusations of voter fraud and a a judge issued an injunction keeping his name off the ballot saying he wasn't going to be listed on the ballot until the accusations were cleared up and with the election just a couple weeks away he knew that he needed to get that injunction you know lifted and he needed to get his name on the ballot and both of those things needed to happen it wasn't enough to win eventually two years from now he needed to win really quickly and so he assembled a team of lawyers nobody had a good answer for how how to win fast enough that it mattered except for one guy named Abe foris and Abe said look there's only one chance we have here which is to try to lose our appeal as quickly as possible so that we can then appeal to the Supreme Court where I think it's more likely that they'll rule in our favor so they sent a team of attorneys to where the the Circuit Court was based the appet court where they read through every decision written by the judges to find a judge who had ruled in a similar kind of appeal in the way that they wanted and then they wrote the worst possible argument they could got it submitted to this judge and sure enough within a couple hours he ruled against them and they took it to the Supreme Court where they won and Johnson's name went on the ballot he ended up winning the election years later when he was president he appointed Abe fores to the Supreme Court and so that kind of manual data analysis is now something that we can replicate in large part with a couple clicks using technology which takes the knowledge that a really expert lawyer like an a foris would have and democratizes it across a bunch of different lawyers democratizes it back to those young attorneys who are looking for that knowledge and trying to build their skill set so uh I'll leave you with that and and open it up to questions [Applause] too um do you also work on um I mean we're parallel with the Democracy conference right so do you also work on uh using NLP on uh government bills that are sort of like cases but AR cases that they're written in a legal way and they have legations yeah we've only started scratching the surface of that of bringing in regulations and statutes um because they're often linked to from the case law so the legal system does have this multi-part analysis where it will start with a statute a law but the law needs interpretation by the courts and so you don't have a comprehensive picture of what the law means until you can look at the legal analysis from the court system too so we found those inter relationships are are important to our users and they they sort of enrich this graph of how things are related so we've started to bring in the US code to our system and understand the linkages between cases um but we haven't um focused all of our attention on that space yet it's been a it's been a real sort of um weighty challenge just to Grapple with the case law so far two questions one how do you make money two how's this like are you ultimately trying to get to be what NEX so we make money through premium subscriptions to law firms um and firms are used to paying sometimes millions of dollars a year for Lexus and West law products and the way that we position it is as an additional tool to those existing conventional ways of doing research we don't yet have a full functionality that can replace them but um we are resonating with the idea that this is an additional tool that goes above and beyond what they can do with other other services the piece that we like in our so we like doing business with firms obviously but at the same time we've felt like we can provide free search and access to the entire world too so we make searching and reading for qu La unravel free um oh sorry so the the question was um how do we make money and do we eventually see ourselves as a replacement for Lexus um so where I was and the answer sorry was uh in thinking about these two parts of having a premium paid for product as well as being able to democratize the information um it fits our thesis that the Level Playing Field of information is sort of the the business as usual right now or it's where we wanted it to go and so we'll make that basic information available to everybody and provide layers of analysis on top of it uh yeah I I like this idea of asymmetric analytics I think that's a very curious sort of notion I don't know if you came up with it but I think it's interesting um the same application sort of tying together information in a case I think it seems like like you mentioned the 10 links is not really the way to do everything right I mean academic research gez wouldn't it be neat if you were putting together you know a neuroscience study you could pull from where the corelates are and all these other have you tinkered with in either your project or maybe seen other companies that do this um applications of a similar idea in other domains of information that really need deeper search yeah so the question if you didn't hear it was do we see applications in other spaces similar to what's happened in the law it's it's a really great question we get asked that a lot from folks in the medical space in particular um and in sort of scientific literature more broadly in the medical space there's not a a whole lot going on but there are companies that have been building out and I'm not sure it's with technology but they've been building out things to highlight the latest finding in an article and to tie that back to earlier research but I think there's a really big opportunity to build similar kinds of search and analytics for in particular medical information but I think there's applications outside of that too maybe one more question last question my do you get any customer in terms of how much time up yeah so there were two questions one is um are there applications in the patent world too and there definitely are there's a couple different companies doing things around the filing of patents and thinking about how you construct a patent and then there's other companies that are focused on what happens in the litigation around patents so we cover some of that but not um exclusively as our focus and then um oh sorry yeah so we've heard uh really phenomenal stories actually of attorneys calling us up and saying thank you last night I spent 2 hours looking for something in Lexus I came in raval and found it 2 minutes um so the savings can be really dramatic can't guarantee that that always happens but um we found that it it does have moments like that where it's really transformational [Applause] thanks