 up there. Great. So let's start by just introducing ourselves to see who's with us today. And we'll start in the room and then we'll join. We'd like everybody who's online to join us next. What's up? We've got about seven people so far and three more texting me. They're joining soon. So I'll get us started and go right around. My name is Dazza Greenwood. I am a lecturer and scientist here at MIT Media Lab and a fellow at MIT Connection Science, which is in the School of Engineering. And this class, the IAP computational law course, is now in its fifth season. And it reflects research that I've been doing here with others. We'll be introducing ourselves shortly, mostly under Professor Sandy Pentland's lab, the Human Dynamics Lab here in the Media Lab. And the broad brush there or the broad umbrella, we call it computational social science. And that involves, oh, look at that. Oh, good. That was another fire alarm. Usually the fire alarm starts just when we're getting rolling. So I'm glad that wasn't yet. Anyway, just to come right to it, the computational law is about reimagining law as a, oh, is my mic on? Oops, sorry about that. Let me do a quick sound check here. For anybody online, can you hear my audio now? Anybody? If someone could come off? Yep. Oh, good. It's good. Okay, great. Thank you. Those of you who responded. It's about reimagining law as a system that is data-driven, that can be model-based, and that can be explicitly engineered. So when you think of law, like rules, whether it's statutes or regulations in the public sector, or rules that are enforceable by private laws, such as contracts, licenses, and other areas of law as well, common law, like even potentially torts, and common law of contracts, among other things, and minister of processes, legal processes, legal instruments, these all ready are expressing themselves as data, as software. And practice in a sense has gotten out a little ahead of the law. And so what we're looking at here is a conceptual framework, but also a very practical, you know, engineering framework for that is engineered to translate law and technology. So there's a one-to-one correspondence just at the very highest level. We do that through prototypes. We do that through convening. We convene a lot with bar associations and with law schools and with law firms and with vendors. We're joined by one of our close collaborators in Thomson Reuters Labs today. So the industry connections are really helpful to stay, to make sure that we stay very, very focused on what's happening in the field. And we do that through education. And so this IAP course is part of the educational prong of this research initiative. And the reason we're doing this through IAP partly is to share some of the best of what we've been up to, but also largely to open the front door and to hear from people that are interested from the community to find out more about what you're up to and to bounce ideas and get some idea of flow going. So this is very much designed to be discussion-oriented and open and collaborative session. So with that, I'm going to pass it down the line to the co-instructor, Brian Wilson. My name is Brian Wilson. I am the co-instructor for the IAP computational workshop course in years past. I have also co-instructed, but before that, I was a teaching assistant for this course. And I think one of the things that I really like about this course is it gives, it provides an opportunity to learn about things that haven't really yet happened. And that's kind of sparked a lot of interest for me in the space. I work as a fellow in connection science with DASA under Sandy Pentland and also in serving as the editor in chief for the MIT computational law report where we're kind of taking the ethos from trying to figure out how we can reimagine, re-engineered the law and applying that to a publication so that people can almost reimagine a publication a little bit so that it's more interactive, more dynamic, more available to adapt to the emergent changes in technology and the things that are actually happening in the world. And so I've had a lot of fun doing that. And I think that that does it for my side. And so I'll hand it down the line to cool Brian. Well, it's great to be here. My name is Brian Ulyssany. I run Thompson Reuters Labs in the Americas. And should I just launch right into the what I'm going to talk about or no? Oh, all right. I run Thompson Reuters Labs Americas. We're a bunch of data scientists and machine learning folks, data visualization people and blockchain people, all working for Thompson Reuters, the giant information provider. So we provide information and data and research platforms to lawyers and tax professionals and global trade professionals. And we also have the fabulous Reuters news agency, one of the most We have reporters in over 100 countries for over 100 years. So here's something on it. Question for folks online. Are you now seeing Brian Ulyssany's screen or are you now seeing the screen with like talking heads? I see both at least. Like I see the slides. And then we also see in the corner whoever's actually put their webcam up. Interesting. Okay, let me see if I can click on the right side of the screen. So with that doing like, how do you? Oh, it seems like that upper right thing. There's a thing on the left that just switched to sharing content. So now it's just all the screen, right? Sorry, we were used to using Google Hangout and they recently discontinued their broadcast product. So we're now switching to a new to zoom and having a little bit of a time. There we go. Aha. Okay. Maybe it's because, well, we'll just go with it. So let's move around the room. And then I'm going to ask people online if they'd like to introduce yourself, if you'd like to introduce yourself. If you don't want to introduce yourself, just say pass. That's okay. But it would be useful to know people. And actually, I have a ringer here. So I want to get started with an MIT Media Lab alum. A guy who I went to college with and the guy who introduced me to the Media Lab 20 years ago and a good friend and collaborator in the computational law space, Brendan Marr. Who are you? Beyond what I just said. I'm Brendan Marr, but first, Daz and Brian, I have to say you guys are being way too loud of straight. You should mention, of course, that the MIT computational law report had part of its driving incentive. That's true. But yes. So my name is Brendan Marr. I'm a Media Lab alum from 95 to 98. And back then I was doing actually 3D audio and virtual reality. I was with the original virtual reality group, one of the first 150 people in the world in that. But today, I'm really excited to be here again because this is where all action is happening. In every aspect of everything going forward, there's going to be some facet of legal contracts and the way we communicate and express ourselves at every level. And it is just phenomenal, the things that I'm seeing. And I'm really glad to be here. And I hope that we're going to explore a lot more of that. And this is my fourth year in this class, four out of five. We've only had four classes calling at the computational law course. So let's just say that you've been to every one. And thanks again for introducing me to the Media Lab. I don't know how to do this. I'm sorry. This is not the home audio and video. We're so great at it, of course. Thank you. I'm Linda Sutliff. And I teach here as a lecturer in writing rhetoric and professional communications. So that means that I teach students how to write in their major field. So a lot of it is scientific communications. But I also teach over at Sloan. And at Sloan, I teach in a field where I was a practitioner for many years. So I teach in finance. And the research that I've been doing on my own is really looking at financial disclosures and particularly the communication patterns that you can see in footnotes. Very interesting. Have you looked at the SEC's XPRL? Yes. So yeah, we should definitely follow up and trade like little, you know, spelunking sessions down into those footnotes. Thank you so much. I'm so glad to see you this year. Would you like to? Come on up. Come on up. Come on up. Hey, everyone. My name is Mergen, Mergen Nacin. I am a master's student at the Signum Kinetics Lab at the Media Lab. I was actually a class of 2010. I finished like nine years ago or 10 years ago now. I was a CS major. I was working afterwards for about, at Facebook for about nine years and then came back to school. So happy to be here. Welcome back. Thank you. Oh, again, actually, can I double check? And what interests do you have in computational law? What brings you to this IAP session? I'll just hold it. I don't have much background to it, but I want to learn what I heard about it. I followed the Stanford Law Codex. Yeah, we love them. Yeah, so it seems like something similar. Oh, so did you hear about this through the Codex crew? Not really, but I was interested in the Codex, but it's like, hey, he also has some in some person. Perfect. Great. Thank you. Todd. Actually, can I hold it for you? Oh, sure. Because people keep going like that. Oh, sorry about that. So I'm Todd Wallach. I'm a journalist. I'm currently spending this academic year as a fellow, Neiman Berkman fellow at Harvard. I'm that guy. And I'm on leave from the Boston Globe where I'm an investigative and data journalist. I'm a Boston Globe spotlight team. Last year, I worked on a project with a spotlight team looking at secret criminal court hearings in Massachusetts. And as part of that analyzed court data on thousands of hearings, and we unsuccessfully sued the state court system, trying to get access to more information about those secret criminal court hearings. And I've been involved in a number of suits seeking access to government data, which has long been a challenge, and it's particularly a challenge with the courts, which in Massachusetts and many other states are explicitly exempt from the public records law and selective in what information they disclose. Yeah, very good. Yeah. So while suing the courts like trying to sting a bee, I would think. Yes. Brave soul. We should talk, so one footnote here, I guess. We should bring in the e-discovery and relativity adjacent use case for FOIA and some of the other analytics on FOIA stuff. Hi, come on up. MIT alum. MIT alum. In the house. Richard Amster. Well, I'm actually an independent solo practitioner, a patent attorney. My degree from MIT is in chemistry. I have a computer science degree. I worked for a couple of decades hacking different kinds of things, including generative document-based systems. So, and I worked in CAD-CAM systems with knowledge-based engineering. So I'm quite interested in the process of using rule-based technology to create documents and systems, which is not exactly computational contracts, but you could see the connection. Also, I'm quite interested in how current trends in AI, which are non-representational machine learning, just not going to work in law. So I'm interested in understanding how representational rule-based AI and procedural contracts and adjudication processes could work. Okay. So just to double-check, I'm tuning in right. So declarative languages like prologue or document assembly kind of systems that are based on explicit business rules along the lines? Well, prologue is obscure and nobody understands it. But I do think there is a way for explicit representation. I'm a little bit of an old timer and my uncle, Marvin, helped found the media lab. Okay. So the not, you know, frame-based representations and all the other kinds of things that we thought about possibly happening in the old-style AI, I think they're going to end up having to come back in order for us to make progress in these fields. And doing the conventional law work that I do, which involves technology startups and patents and some litigation, bored to death. And I would like to see much more interesting things happening in what I'm doing in the law. That's why I'm here. Perfect. Okay. Let's see if we can help you scratch that itch. And I think that's everybody in person. Let's go online. And when I said you're in the house, I meant you're just in the room, not that you're in house attorney. Okay. Now then, let's go online and then we'll dig into the curriculum. So I'm going to call people. Alexandra and off. Is that how Yeah, pretty good. Hi, everyone. Hello. And yeah, I'm an assistant professor at the University of Copenhagen. So I'm saying hi to all of you from Copenhagen at this moment. And yeah, so I work and do research and teach within the field of corporate law, finance law, and tech law. So kind of and the reason why I'm very much interested in joining you and learning from you guys is really to better understand the tech background and to connect with even better into the research itself. So yeah, very briefly. Great. And you come recommended from a friend and a member of the advisory board of the MIT computational law report, Elizabeth Reneres. So we're predisposed in your favor and can't wait to learn more about you and to collaborate with you. Also, did you do I remember correctly that you had something related to computational law coming up at your university, some event or something? Yeah, so we have a well different thing. So one is a course that I'm running together with two other colleagues that it's called kind of a little bit funnily digital lawyer, but teaches our students different kinds of set of skills, including a little bit of coding, but also other skills, more business and tech oriented. And then we have a yearly conference that is called Law, Technology and Trust that I organize in September again this year. So yeah, so I would definitely want to have more discourse there about computational law and whether the computer computational law is a new field of law or whether it's part of every field of law. So yeah, number of questions. Outstanding, but welcome. Very good. And let's keep moving down the line here. Farshad, if you could come off mute and say hello. I'm reluctant to unmute people though because I don't know what's going to happen. I've had bad experiences. Farshad, yes, you're quite faint. So if you could maybe speak up a little. So yes, thank you. I should be better now. Yes, we can hear you well now. Thank you. So yes, I'm also an assistant professor in Baltimore, Maryland and at the business school. My research has been mainly on age law, it's resolution and international economic law, but more and more it's shifting towards law and technology. I have some pieces on blockchain, but I'm also working on law and stats and essentially AI and machine learning. So I'm really happy that I'm part of this group. Great, welcome. When you say economics is antitrust or competition law, within your wheelhouse? Yes and no. That's not the main focus, but I've analyzed it in the context of dispute resolution and how close we approach that, but I'm familiar with that area. Okay, great. We made a call on you and again, everyone feel free to just say pass if I have a call on you and you don't feel like talking for any reason, but we may be talking about new ways on Thursday. Data could inform monitoring, you know, what's really happening in markets and maybe new concepts and legal frameworks based on these new information. We have, but we don't have to surmise so much and imply infer so much about what's happening. So great. So thank you for a short and welcome. And who's up next? It looks like Megan Ma. Yes, hi. Hi everyone. My name is Megan. I'm a visiting PhD student at the Harvard Law School. And my thesis is focused primarily on translation. So it's kind of the translation of, I guess, law that's primarily in descriptive natural language and the implications when that's translated to structured data. So I'm interested evidently in computational law. Yeah. Great. Very cool. Welcome. I don't know if he probably didn't mention it as usual, but Brian got his PhD in linguistics here at MIT. So maybe we should arrange a cup of tea or something later while you're in Cambridge and get really geeky. Yeah, come down to the seat, work to the lab. Yeah, definitely. I'm actually doing a PhD in law, so I need as much help as I can in linguistics. So. All right, perfect. Okay, let's hack. Thank you and welcome. And so now let's see who's next is Natalie Nolten. I think you skip Michael. I'm sorry, I skipped Michael Jeffery. Michael Jeffery. Hi everyone. I'm Michael Jeffery. I am a corporate lawyer in Australia, specializing primarily in mergers and acquisitions and corporate law. It's a nice, well, 6am here in here in Australia. So nice right early start for the next few days. In my interest in this space, I look with corporate and M&A being fairly document intensive, most of my, most of my work in the computational law space has really been focusing on contract automation and document generation so far. And we introduced Dock Assemble into our practice about six months ago. And so that's been an interesting, an interesting path. But looking to look into other areas as well, and just generally open to any other ideas of working, working smarter and more efficiently. Outstanding. We love Dock Assemble. Yeah, it's a great system. Have you been experimenting with community.lawyer, kind of no code overlay for that? We're actually using the document overlay at the moment. So which, look, I use that aspect to certainly helps to speed up some of the initial build process of the interviews, but outside that it's sometimes it's custom coding as well. Great. So we'll have some, we'll have a little exercise on Thursday with community.lawyer where we can do the hacks and stuff. So you're really active in that and maybe can help show us how you've been using Dock Assemble at that point. Sure. Great. Thank you and welcome. So next up we've got now, finally, we have Natalie Knowlton. Sorry to scare you twice by putting you on the spot, but no worries. Can you hear me? Yes, you're great. My name is Natalie Knowlton. I work at a legal think tank research institute at the University of Denver. And my work really focuses on access to justice, court simplification and attorney re-regulation. I'm interested in this course because we have this tendency in law, of course, to look backwards and courts are just catching up with what was popular 10 years ago with respect to technology. So I'm trying to figure out what is going to be happening. I think Brian, you said it really well that this is, and I'm paraphrasing, that this is an area where we're talking about what could be as opposed to what might not be right now. And I think that's really important as we start thinking about access to justice solutions. There, here. Great. Welcome. Thank you. And who's next? Neha? Neha. I think you go by Neha kind of as your nickname. Zoom filled in my full name. I don't know how it got it. But yes, hi, this is Neha. I am a former engineering student, former patent agent, former law student. And now I work at a legal tech startup and I'm really interested in computational law. I've also been volunteering on the side for a code for Boston project where we're trying to help people expunge their criminal record. So another facet of approaching how to use technology in law. So I find this whole area very interested and I'm excited to see what's coming up. Great. Thanks. Welcome. And hopefully if it meets your schedule, you can come in person one of the days as well. Yes, definitely. I'll definitely try. Who's next? Is that everybody? Is there anybody online who we didn't, who hasn't introduced themselves? Okay. All right then. So you were promised computational law. Now you're going to get computational law. Shall we dive right in? Should we start with Brian or do we want to do our thing first? It doesn't matter to me, I think. Do you think? What makes sense? I kind of want to hear from Brian. Okay. Is that okay? Yeah. He just says such great stuff. Okay. Now you're going to get a share. So actually I can say something while we're segueing. I asked, so Brian, you list me, thanks, is also a member of the Board of Advisors for the MIT Computational Law Report, this new publication that we launched last month. And he's going to talk about tax and in the computational law telegram channel. We've been having a little discussion about something new that's happening with the internal revenue service here in the United States. They have recently lifted a prohibition that they've had for some years with industry to basically limit free filing software. So there's some limits on that right now. And there had been a voluntary, I think restraint to some type for where the IRS is refrained from putting out their own kind of open public filing software. So it's not to compete with industry. And so they had just read and distributed an article from ProPublica that indicated the IRS is lifting those constraints. And they're requiring, I think, the intuit and others to have more standard naming conventions for their free file software and to make it more conspicuous and easy to access. And the part that caught my eye, which we're talking about is the IRS is also now working on their own public free filing software. So in addition to that just being, I'm just going to say, without getting too political on behalf of MIT, which is educational nonprofit and doesn't I'm not here to take political positions. I think it's a matter of public policy. This is very beneficial. And it also has an interesting dimension for computational law. So just to set the table a little bit for what we're about to hear and talk about, tax is an area of law, internal revenue code and regulation and all sorts of enforceable administrative guidance as well. That's written primarily as narrative right now, like human natural language. Much of it, however, is very amenable to expressing itself as code. And to demonstrate that you could use like turbo tax, for example, and any number of enterprise packages to do tax prep and filing as well for business filings. And so this code is fairly well positioned to be expressed as its computer machine readable code. And when the IRS is writing code to do the filings, it's even closer to a government agency now promulgating law as code. It's a computational law system. And so you could imagine, perhaps a next step might be once the IRS is deploys their software and it's been used for a few years and it becomes more stable them opening up, perhaps, for example, a series of APIs interfaces so that other people could have their own software public, open source or proprietary, but it basically goes through the same APIs and may have the same types of kind of form validation and other types of verification, which could also be public so that you could actually see for a matter of as a matter of interoperability and just to kind of conformance of your code and whether it's conforming with what the IRS expects. This is another way to express tax law as a computational system. And you could imagine not long after that once it's an API creating business types so that it's collecting and preserving all the data that they're financials that would be relevant for tax, but then also for other types of financial management in a way that natively is ready to support the sorts of analytics and data interop like with filing and reporting that's going to be necessary as part of the information. And so this is a little sliver of in one context, which is the tax context of what we imagine computational law is and how it may play out. So there may not be a better fit in tax law for computation than that in value-added tax. And I think if I'm not mistaken, that's what Brian's going to tell us about in part. Brian, take it away. All right. Well, thank you very much. Yeah. So I think this is a, you know, absolutely a great place to start a computational law workshop because, yeah, I mean, tax laws is, if we think if computational law is law as algorithm, then that's what tax law is, right? It's just saying it provides these rules to, you know, that in some deterministic way, hopefully, tell you how much you owe the government based on, you know, for income tax, your income, for VAT, for the value-add in transactions, and so on. So, you know, like every computational system, there's sort of two aspects. There's the ontology part. So here I have some, a little snippet from the 1040 instructions about what constitutes a child, right? So that's part of the ontology that you need to, you need to figure out, you need to specify what things are. So what counts as a child? Well, a child has to be your son, daughter, stepchild, or foster child under age 19 by the end of 2018 for, I guess, this the previous year, and it has to be younger than you or your spouse, curiously. I never would have sort of thought about that, but... So what if I were to go into, like, stasis and stop biological aging? Okay, we'll come back to that. It's like the fertile octogenarian exception to the rule against perpetuities, which is... Yeah, right, that's an oldie, but a good one. So, yeah, so the government gives you these details about these rules of the ontology of what counts as a child, and then there's all kinds of other things, like was your child enrolled in a school? Was he or she a full-time student over the past year, and so on? You have to make all those kinds of determinations about what the thing is, and then there's all sorts of rules. So here's... This is a snippet from Thompson Reuters' guide to the corporate part of the tax law in 2018 that was changed that has to do with depreciation deductions for automobiles. So once you determine what this thing is an automobile, then certain rules apply to it. So it says that certain depreciation allowances are based on what year the thing is built and so on, and then it cites which part of the law this comes from. So you've got ontology, and you've got rules, and together those things then deterministically produce a result for how much tax you owe, and as Daza said, so turbo tax is an example of... So the IRS basically provides this computational artifact, the tax form for you to calculate your taxes, and so essentially it's a paper and pencil computer, so you store numbers in these buffers, these fields, and then you subtract these from this, and then by doing all of this you end up with your final tax determination. Obviously a lot of that can be automated and so people like Intuit have automated this for your personal income tax, and so you upload some of your documentation like your W-2 form, you answer a bunch of questions, it says based on... So I've determined that this is your income, so therefore your tax burden, you know, your sort of prima facie tax burden is this, and then it goes, walks you through doing all these deductions, and then says, you know, what your actual tax bill is this, and then that's what, and then allows you to file it with the government. So similarly, and I'm not going to advertise TR products, I promise, but TR, for example, among other people has a corporate version of this, so it's ingesting data into the system determining, you know, what each account in the, that you're up, you know, interfacing with the data represents, so these are, you know, this is a payroll expense, these are office supplies, these are my automobile rentals, these are travel expenses, and so on, and then it cranks through all of the arithmetic and determines what the corporate tax is going to be. So, you know, so tax is pretty straightforward that way, and that it does lend itself to this sort of computation, but so one question that, you know, we should ask is, why is a human still needed in this loop, you know, wouldn't it be possible to simply upload sufficient documentation, or doesn't the government already have access to sufficient documentation, that it should be able to just, you know, take your W-2 and various other things, you know, register our documents from universities, your automobile registration, and so on, and just be able to calculate the tax that you owe, why should you have to do it, why should any human being have to do this? Well, the answer seems to be that there's, you know, some subjective determination about a lot of, you know, these values that we haven't been able to eliminate yet, but, you know, we should think about to what extent it would be possible to automate more and more of these things that the TurboTax able to do by, you know, by your updating, your uploading your W-2 form and so on. So that's topic number one. So that's about, you know, income tax and corporate income tax as a computational system. Next thing I want to talk about is VAT. Can I just raise a little? Yeah, yeah, sure. So on the one hand, of course, we don't want to have a needless burden of, you know, manual computation for things like taxes, because in addition to in addition to how much of a bummer it is to pay taxes, we shouldn't be tortured with having to figure out, you know, how to apply these Byzantine algorithms. But so that makes sense, and that I think is like a core use case for more automation. On the other hand, when the way that you left it, I thought maybe it was worth just putting another word or two on where the human in the loop would be, especially the human who would be due taxes and maybe wants to double check some of the subtle assessments are made about how much tax was due or whether they might want to look at having a more or less aggressive posture and how they interpret things. So where is the idea that there'd still be the human, like the taxpayer or someone else in the loop, or were you really envisioning like a purely completely automated system, like some jurisdictions, the government and just informs you how much taxes due because they already have access to all the information. Like where, what were you saying? I'm not really taking a position. I'm just wondering to what extent current U.S. taxes could be fully automated. And, you know, and potentially, of course, if things are fully automated, and there's an explanation, then you could look at the explanation and say, oh, you know, no, I object, that's how property taxes generally work, right? So you get your property tax bill, you have the ability to say, no, based on these comparable properties, I'm paying too much. I protest, they come and they check and you can win or lose. So you could go towards that kind of system. Yeah, I think kind of hitting on something you'd said before as well, you know, I think more of those tax kind of like measuring sort of jobs where you're trying to calculate after the fact are going to be shifted to like trying to like figure out how it fits ahead of the fact. So it'll be more like ingesting into the system instead of trying to untangle the system. Or maybe also choosing what algorithms or what templates or what based on how you want to interpret, like how aggressive you want to be with tax avoidance, one could imagine choosing different algorithms or different models or setting parameters or thresholds on software differently. Maybe something like that on the front end. Yeah. Could we could we open up to Brendan before you March? Absolutely, yeah, sure. And then in order to do this successfully, I think everybody should be able to hear Brendan, because he frequently says great things. So, okay, I'm going to do the mic. Chiwala. So, Christmas list is wireless, lavalier. So it's wonderful. So what I hear all three of you saying is it's really about simulation. And what is amazing all this is that when you look at, you know, what gets built on top of all this stuff, it's phenomenal. Because then you can think about, you know, all the meta decisions you can make, the business logic, you know, should you do this and this jurisdiction, should you do that and that jurisdiction, you know, all the fun stuff. That's all I'm going to say. You guys are on some. Todd. Yeah, Todd. I just want to add that there's been some resistance to having the U.S. IRS automatically do people's taxes and present them with a proposed bill and calculations, particularly from companies that make money selling the software. And they've successfully lobbied Congress to pass laws and block the IRS from doing this. The IRS has also been historically reluctant to release data and information about the information it collects. So for instance, it issues private advisory opinions to individual taxpayers and doesn't make those widely available. So some people have access to them. Some people don't. It collects data on millions of nonprofits and has that in electronic machine readable form, but had to be sued to make it available and did so only reluctantly. So there's a lot of resistance, both from industry and from agencies themselves in automating this, some of these processes or making the information available and transparent. So that's some of the obstacles as well. Did you hear my reference to the pro-publica piece? Are you familiar with that? Yeah. So the pro-publica piece, if you distributed it, goes into more detail about how industry players like Intuit have lobbied Congress to block the IRS from essentially taking over tax computation and giving people the option of accepting the IRS's calculations or modifying them, which most people would probably do and significantly reduce the amount of money that industry players make. Thank you. What a great point. There's some online points as well. It really brings up the point that when we're talking about this in a public law context, in addition to the algorithms, we really have to talk about architecture and how many layers of that stack need to be open in public and then where is it appropriate to have things private? You know, some individual financial things should be private. Should some of that in the aggregate be public? Should the algorithms that are being used to assess your tax be public? These are some of the core questions, and this is all within the scope of computational law. We have somebody online who is seeking to speak and go, or are we supposed to read? Okay, we're going to read it. But why don't you read it? Okay, so this was from a little bit ago, but it was Andre talking about how adoption, oh yeah, so that was to the point about how can you have somebody who's you know, a child that's older, or I think it's a different, right? It was the term? It was a child. So you can imagine that if you marry someone, you're 18 and your spouse is 30 that they could have a child that's older than you. Fair point. Yeah, and also obviously adoption, like straight up adoption. Yeah, that's true. Yeah, and then sticking with Stasis also. He has another point as well that I think is really good where this notion that you know, when an individual submits their tax forms, it's kind of a voluntary act by the taxpayer. So there's a question about what do you think about, you know, the voluntariness of this and how that relates to kind of automating it and you know, declaring it down upon everybody? No, I mean, there's a whole, there's a whole lot of policy implications, a whole lot of policy choices to be made here. You know, so currently I think I've read on the way over here that only four-tenths of 1% of individual tax returns are going to be audited this year or this will have been audited this past year. So we kind of made this decision basically to audit nearly no one, but then you require, you know, have everyone do their own taxes. And you know, there's a from a policy perspective, would the revenues be greater if you know, the taxes were just done for people and they could object and so on. And also, forever it's worth in the United States if you, you know, voluntarily choose to not file taxes and the tax authorities at the federal and state levels at least believe that you had taxes owed, they will eventually, although it may take, you know, a couple of few years, they will assess what they think your taxes owed are and they'll send you a form and they will inform you of that, give you a period of time to object and then if you don't object, then they will consider them to have been owed and not paid and then, you know, what happens after that, like the processes take, take what, but it does, it just let's just put a pin in for now other than to say this raises basic questions about the social compact and the role of citizens in government and and how do we translate forward, you know, autonomy and freedoms as well as the obligations in administrative state and appropriately, appropriately when they're all digital and that does change some, that will change some things. Oh, yeah, we have one more in the room and then and then let's forward. Go for it. I just want to raise the concern that I just want to raise the concern that you really have to think about the fox in the henhouse problem. Could you just to connect the dots? Well, the IRS does have a regulatory function and normally you don't allow a person who is the regulator to also be the read on the other end of the transaction. So we just really want to be careful that those functions are ultimately separate or you lose or at least you run with risk of losing an important kind of oversight. Beautiful. Thank you for saying that. It was it may have been a brief comment but a very important one and I was trying to get at that a little bit with the question of autonomy and the role between, you know, citizens of the governed and the governed. I mean the quintessential definition of a totalitarian state. It's holding all roles and all of the cards and so that, you know, that really does raise the question of how far we want them to go and maybe next year or in next semester or two we should explore what an API would apply for taxes to see, you know, how much of this still provides, you know, proper decision making for the tax filer versus expropriating a lot of all of those decisions by the regulating entity because I could build an API where you can build a software but like it or not it's never going to process unless you do it my way or I can build an API where you're literally just reporting the results of your tabulations based on your decisions which may be more subjective and so the questions don't go away with computation in some cases there they may just become more explicit but computational these systems are becoming and so now is the time to begin to come to terms with that and understand how to express law computationally and then how should we be expressing it but first things first here we wanted to see if we can get really good at expressing different rules and different systems computationally so that we can distinguish and then you know perhaps uh engineering a wiser way Brian got another chat oh another chat one last chat okay clicky clicky Natalie uh regarding her point about the regulatory issue this is exactly what we encounter with lawyers the lawyer in the hen house right thank you Natalie i'll be back to you Brian cool Brian okay so so the second topic I want to talk about was uh computational issues with bat so value added tax uh so not not uh such a big not a big thing here but uh in Europe and now in the Gulf Cooperation Council and the Middle East and other jurisdictions you don't have to worry just about property tax and income tax but also value added tax so just you know very very quickly so value added tax means that at every step as a you know a product is enhanced to bring it to market there's a tax burden that's that's passed along and then as you pay the tax forward you get the tax the the person of the previous step gets the tax back so that that's a well-known thing in the EU and I'm just going to quickly show this video from 500 million europeans that's 50 billion euros every year and when he scores with that paper how do the criminals do it they set up shady companies in Europe just to give you a tax money here's an example company a sells mobile phones to a company in another country in the EU no vat charge because cross border sales shouldn't incur vat still with us good across the border company b sells the phones to company c in its own country and charges vat to company c he's supposed to pay this vat back to the treasury but it doesn't protect the vat and disappears with the money but if company c then resells the same phones back to company a then applies the tax credit the cycle is complete told you it was complicated the point is they all disappear before we can catch them it's called vat carousel fraud the same goods go round and round like a carousel your money is stolen and it's happening a lot and because of the current rules it's really not easy to catch them some of these guys have been caught but that's just the tip of the iceberg are you mad by now you should be some people use your money to buy sports cars and billets it may be used to fund other criminal and perhaps terrorist activities the rules of the game need to change that's what the european union is doing changing the european vat system so we can stop criminals in their tracks and put this money back in your pocket let's stop the carousel okay so i'm carouseled oops women so there we go all right so um so that was a little video about what's called carousel fraud in vat so i won't be labor that but so here's some thompson writers has invested in a company and and done some work with a startup called somito which is an amsterdam based startup which has a blockchain based solution to eliminate carousel tax fraud that it is piloting with the dutch government so i'll just quickly play this oops oh i see um it's fairly easy just fill in the amount of vat you wish to pay or receive not only is it easy but it's also sensitive to fraud worth 50 billion euros per year in e you alone now that's a lot of money so what can we do to solve this in the battle against vat fraud blockchain technology might be the solution we've been waiting for our blockchain solution named tx plus plus allows companies to register an encrypted fingerprint of the invoice i'm not storing actual invoice data the risk of data bridges is eliminated and because of the decentralized network architecture there is no single point of failure so risks of system failures are minimized can you see the bigger picture imagine a vat system which provides benefits for both the public and the private sector it can save millions without risking confidentiality a blockchain based invoice registration system has the capacity to significantly tackle missing trader fraud while at the same time guaranteeing taxpayer confidentiality so how does it work this invoice registration system can be coupled with any existing accounting package companies can also easily register invoices through a web portal a fingerprint of the invoice is generated timestamp and encrypted as a result of this simple procedure fraudulent reporting cannot go detected any more encrypted fingerprint of the invoice okay so uh so i think it's um so so the idea there is is clear enough if you understand a little bit about blockchain so by having all of these invoices uh be encrypted and just uh and just uh putting on this on the blockchain the parts that are needed in order to you know basically the parties and the amounts um in order to calculate the vat uh because of the immutability of the blockchain and because of the decentralized nature of the blockchain then everyone can be assured that all of all of the the correct vat has been paid uh there's no possibility of the you know the one party not paying their tax the the vat into the treasury and then getting a refund for that tax and making and disappearing so um you know this seems like a promising solution that solves some of the problems that we saw with uh you know the income tax meaning that because all of the all of the data that's relevant to calculating the tax is is computationally available on the blockchain uh that's not the case with us income tax and so on because there's all these external data pieces that we need to know about like when was the car when did the car originated and so on but uh with vat all of this can be calculated simply on the base of these invoices which can be cryptographically just fingerprinted on the blockchain and used as the basis for this complete calculation so that's that seems like a promising idea yep oh no this is the the next topic so that's perfect yeah sure great um so what's pretty interesting that i've come across is the idea that with uh zero knowledge proofs you could prove the you know the identity of of an existence of a value right so you'd be able to know that a tax is owed without knowing anything else about the transaction and I think once these things get scaled up you know the way we conceptualize uh transactions and their taxation is going to dramatically change because the information that we need to know about taxable things is going to be fundamentally different in the future then can I can I um augment that a little sure so with zero knowledge proof this is a perfect example one of the elements that you would look to prove for the amount owed would also be the identity of the parties that owed it um and I wonder how would how would blockchain for example get us to like this vision of a legal entity identifier or some other way to identify like in the carousel fraud what was the legal entity the business or the person um who had the obligation in the first place and then how do we make sure that that whole identity system doesn't become you know itself like a tool of a tyrannical um you know entity that might use it against us yeah I can't speak to the tyrannical use case but Tyrannosaurus um Fox in the henhouse Tyrannosaurus Lex I mean so the identifiers you know as you as you know very well are you know these pseudonyms right so on the one hand there's the anonymity that's that's all pseudonyms provide there's systems like these decentralized identifiers that not only provide this pseudonymous identifier but also provide the ability to to verify that that is your identifier cryptographically and that you control that doesn't rely on some uh intermediate like the DNS system that says you know in order to um uh de-reference and identify like your LinkedIn profile URL you depend on this whole infrastructure of the DNS system to say oh to find out what's behind that URL I need to I rerout it to this you know server and it and it produces this information which may or may not be correct and depending on whether someone messed up the DNS routing right uh with the with the decentralized identifier I control what information is de-referenced and I that's always under my control and it's always verifiably me so that seems like also a giant step forward so just for uh for the so you can do this at home um you could um learn more about what Brian was just referencing by searching DIDs or decentralized identity and there's a worldwide web consortium um set of specifications and how this could happen and the mechanism by which people can acquire you know kind of sovereignty in a sense over their own individual IDs in a decentralized way would be with public key cryptography so um you would have a key pair and initially and that's how you could prove um that you were an entity um that had a pseudonym um that connected in this case to say a tax um transaction uh without giving up your individual your ID to start with and then one other little distinction here which is important we were talking about architecture and stacks earlier the DNS system the domain name save the domain name server system um is distributed um you know there's a lot there's a number of domain name servers but it's not decentralized it's very centralized it's it's very hierarchical um by contrast and so in the United States um if it was if it was an identity system that would start looking a lot like a national ID um we are allergic here in the United States to national IDs same in Great Britain uh for for various reasons um the this concept of this um alternative way to have decentralized and distributed identifiers may be a way to to um create a capability of having verifiable and like you know provable identity uh but in a way that's decentralized and maintains um autonomy and like things like civil liberties and and other uh other you know beneficial kind of um attributes uh to the individuals and companies just want to unpack some of what you just said there cool yeah I know I should also point out that Thompson Reuters is on the working groups for both the decentralized identifiers and uh verifiable credentials yeah and of course we we don't uh endorse um individual companies at MIT but I'm just gonna say I'm proud to be wearing the Thompson Reuters hat today because that comes close to an endorsement because that because that's you know having big companies as well as you know civil libertarians and academics think about um the decentralized things like decentralized identity is critical to getting it right and then also for to being acceptable and adoptable at scale so kudos to you and your company for for being involved in that thank you very much all right so topic number three so uh the last in the last topic we we talked about how blockchain could potentially be used to solve uh some tax fraud problems uh this is the counterpart uh what are we going to do about taxing blockchain-based uh assets of cryptocurrencies and so on so in uh you'll notice in when you fill out your 2019 tax form that question zero the top the question at the very top of your tax form is going to ask you at any time during 2019 did you receive sell send exchange or otherwise acquire any financial interest in any virtual currency so I don't know if that means s and h green stamps um as well that's an ontology question but uh that is uh interestingly the very first thing that you have to fill out in this form we can leave that to the linguists can we though yeah exactly there's there's a lot of things that uh the question was what about frequent flyer miles the uh the uh instructions say but in any case so so there's a lot of you know so uh if you've got uh cryptocurrency assets oh actually I'm sorry let me just say um there is there are working definitions now I think in a in a cross cross agency regulatory group for virtual currency digital currency um digital assets um so we're not mistaken sec cfdc irs um other components of treasury um are starting to get together with um with um common definition sets for these and so I am almost positive the irs has a formal definition of of uh of that term and that it's increasingly being used as a common glossary set across regulators and just forever it's worth um you can yourself go to mit.edu forward slash blockchain and see an initiative that um we did um this computational law initiative like three years ago now or so with congress where we um convened a bunch of regulators and congressional types and a lot of businesses and blockchain people and standards organizations to look at coming up with common terms um so what would be a legislative and regulatory what would we mean when we said blockchain what would we mean when we said smart contract and we didn't get down to virtual currency but in subsequent years uh that they have yeah i'll just wait oh that's that's Brian next segment right yeah tax and crypto tax and crypto yeah that's all right um so anyway so just wanted that there are emerging solutions so thompson Reuters just partnered with a company called verity a startup that um does basically accounting of crypto assets so that to enable you know this next step where the those kinds of transactions are going to be routinely taxed and so you're going to need auditable uh records of uh you know to do capital gains on crypto transactions for example you need to know like how much it was worth to begin with and how much you sold it for and all of that and tracking in a reliable way is going to be increasingly important so the verity does this across they they have this uh legible technology that tracks uh you know a variety of cryptocurrencies and across different exchanges and so on and then finally so so that's uh that's the the counterpart to blockchain is solution uh cryptocurrencies as problem uh then there's just the my last topic is just keeping up with computationally keeping up with changes in tax laws so um uh an intern that we have last summer a student at duke uh did a very interesting project for us so one of the things that um the businesses have to deal with is changes in sales tax regulations so uh you know states and cities and uh frequently update their what's what's uh has sales tax applied to it uh then there's also this whole idea of sales tax holidays and all of those changes need to be then you know computationally implemented at your cash register so that the sales tax is correctly applied to this category of products and not this and so then you need to have the sort of whole ontology of products and what falls under what category that's specified in the legislation and so on so what this what this intern did for us was used um vector based word encodings as a way to identify uh when the legislation spoke about different categories of things that would be taxed or not taxed during sales tax holidays mapping those to an ontology of of sales tax um of product categories for sales tax that comes from where it's maintained and then ships to its customers so that they can you know do that calculation at the point of sale so here's for example this was a um sales tax change in for Iowa and it said simply things like digital audio visual works digital audio works ringtones digital books you know digital greeting cards would now be I forget if this is that they were going to be taxed or not taxed I think this meant that they were going to be taxed so then you have to figure out well you know what what counts as a digital audio work and so on um and so so he used um you know current techniques in in natural language processing to map these natural language terms to uh these product categories which enables uh the the people who are doing this mapping to be much more effective um at uh at doing these these mappings so that's that's really all I have to say here so um you know on the one hand you know uh tax is obviously very computational um how much of it can be automated in part depends on how much information is available to the system a lot of the income tax information is sort of external to anything that you might be you know sort of conceivably um brought to bear in an automation can blockchain help prevent some tax fraud what we've seen in in that that it seems like that's a very good use case uh what about tax and crypto assets that's sort of a technology that's that's just starting and then are there also computational techniques that we can use to help us keep up with all of these changes in the tax law so algorithmic we keeping track of updates to the tax algorithm is a second order sort of task right here so could you imagine for example in the future um if let's say the IRS had an API um for um people that wanted to um roll their own tax prep software or wanted to create a open source project or or like the next version of turbo tax um could you imagine that um they might publish their API and eventually um do it formally as regulation of the way the sec has published their um xbrl um filings for um for publicly traded companies and then could you imagine to your point about looking forward um that if they were going not if but when they up version the API that they might publish an nprm a notice of proposed rulemaking and identify what the new code would be and maybe even like the test suite yeah how you could come up the software and then have you know however many months for comment from people to say you know we think this is fair or unfair this is this breaks um you know previous version or whatever they would say before they actually finalize the algorithms and the uh and the apis for for that and couldn't and of course couldn't some of that testing be done algorithmically is what you're getting at yeah absolutely yeah no good points oh what a world that would be yeah great well thanks for your attention everybody that brian you're listening computational um okay so um let's see um in the just in the afterglow of that presentation are there any any um comments or uh questions or or ideas about about any of that uh anyone online or okay you're getting some kudos online um or in the room okay we've got one this if i hand this to you would you hold it at a steady distance from your mouth okay thanks uh this question this question is for uh for cool brian and the question is can you give us a sense at this time you know in you're in a very unique position and the question is what are you hearing from uh the corporate culture in terms of you know their beginning beginning to uh grapple and understand these kinds of issues you know is there an outcry is there are there interests by some who are the who who are the players i don't know anything tell me tell me a story um so yeah so interestingly i think that uh the whole idea of blockchain the whole idea of uh the usability of blockchain is uh is definitely being embraced by government so we've been we've been doing some interesting stuff with uh customs and border protection for example about um you know tracking um uh entries of um shipments and so on via blockchain and the government is sees immediately how much this would make their lives so much easier so in in that sense it's it's kind of interesting that you know the government's always supposed to be way behind everything that they're they're very much uh i mean at here at MIT everyone thinks well blockchain oh that's ancient news but uh in in the government world you know i think the certain agencies like the cbp are actually pretty far ahead of things um i don't know that too many people are super worried about uh you know in the corporate world about taxation issues around um cryptocurrencies and things you know people are aware that there are issues but i don't think that anyone's too um worried about it uh it's not really a mainstream thing for most multinational corporations to hold cryptocurrencies so yet yeah yeah exactly um yeah and then there's a question online i i believe it was about algorithmic discrimination so like to to what extent can some of these you know tax systems be have you know almost like a discriminatory effect on um uh the way that taxes levy do you think it's like greater than it would be as kind of like this static process i mean i think this is just a general problem for ai right um poorly designed ai systems can uh lead to discrimination um and you know and there's a there's a role for policy here obviously uh there's a role for um you know simulation and so on and and seeing what would the effect be of of automating people's tax returns versus voluntary voluntarily filling them out and so on um so i i don't this is not really a particular issue for tax i think this is just what we have to face with ai and automation everywhere and we have a clarification on that and then you've talked um it wasn't uh discrimination it was demonstration and the clarifications explaining how exactly the algorithm works in a certain system or software we can imagine situations in which the automation um of the tax activities by states could lead to algorithms that the taxpayer could not understand so this gets to um i think technically some people call it interpretability of algorithms like especially with these black box algorithms that you don't have access to which Todd was mentioning uh some of them like proprietary or classified or something or or ones where maybe have perfectly good access but nobody can figure out what's going on and it gets down to uh it can be comprehended much less replicated well i mean it's certainly true that there are a lot of ai systems where you can't figure out how it's making the term you know the determination but you know automating tax is not like a deep learning system it's very straightforward arithmetic right so it it should very well be explainable how you how the calculation is if you have if you have if you have trouble explaining it you're doing something right yeah right the world needs a lot more high school math teachers and science teachers thank god for david call the rooso and people like that um todd did you want to okay we have a microphone for you and i just for you to keep an even distance from your mouth all right i had a follow-up question uh currency and blockchain uh so there's widespread use by bitcoin and some other cryptocurrency by ransomware black hat hackers and there's a real concern by a lot of government agencies that it makes it easier for people to hide their identity and get away with fraud so i would be really interested in how if you could explain how the implementations of blockchain that you're thinking of wouldn't allow people to hide their identity and then disappear as they are with a vax carousel that carousel and instead there would be more transparency and ability to track down people who don't pay their fair share right um you you seem like you want to say something i was going to say i i think with uh you know they you look at the silk road marketplace they've been able to kind of uh take the uh entries into the bitcoin blockchain and actually figure out um you know different people who have like continuous uh use on that blockchain or uh in the kind of illicit marketplace and and so i i think there there's that way to do it but it involves knowing what the the public key is right and so once you know the public key or once you can see oh this uh this public key continually engages in the carousel fraud you you would have a way that you could identify that public key as being like a bad actor for continued use right um so i in in all these blockchain schemes there's sort of you know different differing levels of anonymity i mean some blockchain bitcoin is meant to be completely you know to allow this these anonymous transactions um someone in in the vax uh scenario for example is is the the government and they have their they have the right to know who's who whereas for everyone else that's um you know these these encrypted values so they so it preserves privacy in an outward facing way because people who look at the blockchain can don't know who's trading with who so it preserves those trade secrets but the government has to know who to send the the bill to or to go after so they're they're able to you know identify those people um so so different blockchains will have different roles and you know that that's how you could address those those problems i think that my assumption is all the best identity fraud of these new systems hasn't yet even been invented um and so we're going to be in a in a in a time of escalating you know measures and countermeasures until until we figure out how to do this right and identities near the bottom of the stack of what we have to get right as we transition you know the economy and governance in the society into a computational footing yeah and i linked in the telegram channel to a wired article that goes over how you can kind of go back through and look at uh transactions and kind of start identifying people from bitcoin it's amusingly titled your sloppy bitcoin drug deals will haunt you for years not mine not yours uh i mean the metaphorical you um okay so um in your another one okay don't worry uh cool brian you're off the hook this question actually one of us this question is is really for the folks online oh and especially the folks who are online and uh don't pay us taxes so i'm very curious as to uh how you're paying your taxes uh in your respective countries and also a cryptocurrency taxes my understanding is that paying taxes in every other country in the world except for america is much more streamlined yeah i can jump in there from the austrian perspective this is michael jeffrey um the we've taken a bit of a hybrid approach here where you look the the austrian tax office doesn't it doesn't make a uh a determination of what your tax is but it does pre-populate what your information would be and if you are if you have a very simple tax position um basically like you know you've got a job you get a salary and you don't have a whole lot of um uh investments or deductions then most of the time they get it precisely right it's uh so you log into the tax system um people have everything pre-populated for you and the position is just do you agree and submit or use it or do you want to challenge or modify things so it's uh you do have the ability to modify it but it is most of the time i've never changed it uh that's been around for the last two or three years and on the cryptocurrency front the big uh the big question mark with cryptocurrencies the last few years has been about whether it is really treated as a currency and so in austria we have a value-added tax as well which is called the gst um puts a good services tax um and until about two years ago dealers or dealers of cryptocurrency would also have to charge if they were doing it legitimately they'd have to be charging their vat or gst of 10% on cryptocurrency sales as well so it wasn't really treated as a as a currency um so it really didn't wasn't really effective as a currency because if you were doing it legitimately you'd be paying paying tax on your transactions as well um an extra 10% so um that that all changed about i think it was about 2007 so now it is look there i don't i don't see too much in the corporate sense of people using cryptocurrencies but it is uh becoming more recognized at least from a tax perspective as being a legitimate um a legitimate currency nice yeah i like that aspect of um pretty of sharing particularly sharing all the sources of information that it's being used i know that the irs um frequently before i do my own taxes i'll go and ask for my tax transcript and all of the all the income that um like consulting clients have reported just to make sure that connects with what i thought and that um no no curves have been made and and so forth yeah i think the the pre-population notion actually also gets to the idea of the the demonstration or explainability piece too so that would be a good solution there yeah michael did uh do you do you know what affected the uh you know pre-populating or pre-calculating people's taxes have on the tax revenue did did it go up or go down um not not sure actually but like i think it's one of the few areas where i think that uh people aren't challenging the government that much um most of the time most systems that particularly particularly here that have been introduced from uh where the government is trying to get into a a techno it launched some sort of technical play have been a quite a quite a failure but um this one just seems to work um it's very simple and i don't hear of many people challenging at all having issues with it um from a from a privacy or provide privacy perspective the the tax department here largely has just become a big um um most of their job is now um crawling over data and um there's i don't hear of many people having an issue with it that's awesome yeah cool australia is the future yep so we should probably in some senses we should probably do the class from australia next time right i think that probably good good for it oh no you're supposed to say yes we invite you um to okay um so so we move forward i reckon okay so thank you very much brian awesome job okay so now so we did this a little bit out of sequence um logically um you know some like this newspaper people would know even better than i how to present and and returitions um things that i was taught you start with something broad and then you go narrower and narrower it just thought it might be nice to start with it with a concrete example that hopefully everyone can understand and what we're going to do now is broaden it out a little bit because computational law isn't just computational tax um in some ways like taxes the worst kind of um um you know poster child for a system that we think actually could be quite exciting and could could make things substantively better um and so this uh next presentation is based on a overview that brian and i presented at relativity fest we did a version of it too for the computational law blockchain festival yeah that's right that's not so so this is somewhat iterative but it's kind of like our this is the stump speech now for uh basically answering that existential question what is computational law and why might i want to love it too so um with that i'm going to now do a little screen share oh i can do it on mine if you need um okay how do we do that um i'll go the presentation slides just one moment it's having some thinking let me do it yeah you should do it it's uh i'm getting some because i had to open it in a different browser since i'm doing the ipad okay all right one moment getting the spinning beach ball of death why does it want me to open system that's what i had to oh how do i make this not big if you stare at the beach ball too long it hypnotizes you so just be careful be careful okay it stopped now what do i want to do to allow this oh zoom i want to click it or something oh i'll i'll do that on mine real quick so that we can right i'll slog we're great at computers obviously this is what you're about to see you got it it's making me quit and get back in okay these computers are a little too smart oh we're trying to reconnect all right get back here stop that i heard the consequences of design um so for those of you online we are now we're trying to get into presentation mode there's another way to do this you can go to mine i've got it now perfect okay quick quick check uh with um those online uh can you see the slides this will also tell us what's recording okay very good any anybody online yeah so nai has said yes oh yes very good okay off we go okay so this is uh our talk on the future of legal service delivery where we're kind of looking at the role of interoperable absent services within the context of law and uh i don't know that we can run through this in 15 minutes so we may go a little bit over time um but everybody will be able to watch this at their own convenience later on and we'll have uh some little q notes uh to where the to where uh cool brian's lecture ended and where art started so that it's easy to kind of jump back in um so you already know about us uh and what we kind of uh what our uh assertion is is that technology kind of increases our potential for uh delivering services especially legal services between people every day um there's a lot more data out there now that data can be used combined configured in ways that uh enable more efficient more transparent uh legal processes and legal services and um so uh kind of speeding through that um you know yelp data has been shown to predict health code violations more effectively than uh the the existing system and so you know what would that look like for law if we used um different sorts of legal data to start uh you know addressing problems before or in the in the case of tax you know what if we started using data that already existed to kind of pre-populate some of those tax systems um you know what if we could use twitter data to track the spread of contagious disease a little bit better this is uh they this is a project that had been done where they were actually able to show you know they were able to track the spread of a contagious disease about uh 80 faster than they were able to with some traditional measures um so the question really becomes how do we optimize data for law um so last year we had a session on uh data for legal apps and services with Juan Ramirez from the company in stereo they're uh they work with a lot of e-discovery providers to build custom solutions and workflows um around the way that that data that the data from e-discovery gets reviewed so one of their uh projects involves creating like a GIS layer to visualize where um I think in a in a pipeline uh in litigation about a leak in a pipeline they were able to you know visualize all the different places where people had gone in and worked on the pipeline over time so that they could show that you know they were handling that with the right amount of diligence and they had another app where it was like a almost like a slack bot so that they could automatically do request do request for um you know some of this different data natively in slack and so it gets a lot to this idea that if you if you turn law into data you can do a lot more with it and uh and they're doing a really great job of that um in another context uh this has been used to evaluate non-disclosure agreements so this is a company called LogGeeks and uh LogGeeks trained they trained this AI to review uh non-disclosure agreements for errors or you know not errors but uh things that would cause concern and then they kind of took this they kind of created this challenge where they pitted the lawyers against uh they pitted the AI against a bunch of uh senior lawyers at big firms uh academics uh senior GCs at tech companies and one of the things that they found was you know the AI spotted issues nine percent better than the lawyers did um but the I think uh most uh the the craziest fact is the AI could do it in about 26 seconds where the uh average time taken by a lawyer was 92 minutes and so uh so you know that that kind of shows uh the potential of uh kind of representing law and legal processes as data and so and this one I think it's especially important to point out that um the creation of an AI to review a non-disclosure agreement is is kind of like a different process from like representing taxes data because this is like representing legal knowledge as a kind of model that um eight data set of like a non-disclosure agreement goes into and then subsequently is processed um so what we're gonna kind of go over here now is you know where have we been uh actually could I just say just one quick thing about where we would have been like 10 seconds ago um so on the NDA thing where you could imagine that fitting say in a life cycle of a legal process might be um for a business person like the first place people go is uh robot lawyers like could could how much life could we get successfully through without lawyers at all how much legal knowledge and and um well um let's call it like um skill and practice could be reduced to to um code um so you know for small businesses and so forth doing an NDA that could be a real challenge or even for for big businesses uh it could be a big spend um it then moving further down the spectrum toward um human centrality um I could imagine when I was an in-house attorney uh if I had a tool like this um to get me to the first drafter to the first um first um human analysis of an NDA I'm being presented with to do issue spotting um that is another way you could look at this so it's not replacing the human but maybe it's super charging uh the human and then there's a matter of judgment about how what what the clients interests are what their priorities are and whether you know they care so much about um you know how maybe for some clients um only having uh a restriction on date on uh on disclosure for two years is much more important than the range of certain types of information that are subject to the restriction on disclosure so anyway some some judgment about legal practice and representing the interests of a client um could be advantaged by this or you we could easily imagine kind of um you know jobless uh you know hellscapes of the future where the AI could actually do a worse job and we have over dependence by using the same tools that we were just seeing so um part of what I think is interesting about where we've been is we're now in an exploratory almost field building stage I'd say in the economy with computational law apps and um services um and we want to show you what some of those are but we're not yet presupposing precisely how they should fit within the life cycle of legal practice or um or or within um you know business and and and social processes yeah and I think that's a great point or I think that's a great thing to point out um especially because they're especially when you get to like litigation there's a certain amount of uh gamesmanship that takes place that you know I think an AI would be very ill equipped to to handle something like that um oh do you mind speaking of practicing lawyers who have actually have MIT degrees so so let's consider a couple things one is that um you know what you invest in the technology is definitely going to be influenced by um what the improvement is what the costs and risks are so you said um we're looking at this chart and we see in 94 maybe say it's a 9% difference that's actually the inverse of the way I looked at it as a lawyer which is that in the in the AI case there's only 6% error rate and in the human rate it's 15 it's three times worse because there are three times more there's three times higher likelihood that something's going to come back to bite you so that's one number two um the context of litigation and things like that there are already lots of tools that people use to manage things like decision trees for working out you know what's the potential progress of litigation and which way it goes so I think it's the computational law applications can't even be identified without thinking about what the milieu is the environment and so I think we can maybe we can start here by talking about what kinds of technologies can be applied but I think there'll be a more fun discussion if at some point which we're not going to do in the next three days to continue with this we say you know look at the kinds of apps that people have used in the legal industry and how could you take that technology what kind of what kind of deep learning you might could you be able to apply to what the average decision tree looking under litigation would be to in fact seems to me to be very possible to do much better on strategy yeah so I don't accept the proposition that we're not going to get there in fact I think it's opposite thing I think um by going in this direction we'll be able to outsmart lawyers yeah I mean I didn't mean to suggest that we weren't gonna ever get there or that it would be impossible to get there I think my my assertion was that there there's so much kind of unique kind of almost quirky behavior that goes on that that lawyers over a period of time learn right that you know some of this you can't represent with a a simple a straight away and so I think having this is a you know more like uh as we'll get kind of later on like a an iron man suit instead of like a c3po sort of thing where it's not fully automated but it's you know this combination of human plus uh machine gives you better results than human alone or machine alone now that's certain that that that makes total sense I also see that you know not falling into pitfalls sure is uh is one of these things and I think in the history of AI there were a lot of things where people talked about sort of this mythology that humans could do something better turned out to be false so I think it may turn out that the the assisting thing is good too but there may be it may be better than c3po or yeah it may be that all of a sudden like for very routine stuff it would be right like for uh I think you jump me on I'm sorry I'm Richard Richard two MIT alums um and then uh just as a placeholder slide 17 which we should not skip uh we'll get us to the um this tantalizing idea you had can we extrapolate across a bunch of areas of law and uh see where people are using apps and services right now and what we can learn from that so we've got a bit of a breakdown in the market that we should use to do some brainstorming so let's I guess that's a note for us let's not skip slide 17 so we're together uh great so um Brian I think you don't want to scare anybody but I think but I think yeah I'm with you Richard um so what's really fascinating here I think is that uh and what I think what you're getting at Richard is that when you when you look at any kind of system right when you go down to a very low level it's the very very tiny changes that make the biggest differences in architecture and what happens you know in terms of usability and interaction of the doubt and I think what you're getting at Richard is that you know when we take the data and we encode it in a way in which we can uh apply AI to it then even very very tiny changes in that encoding process makes really large changes at higher level systems that are AI based yeah and and when one starts to think about smart contracts and the data as encoded yeah as contracts etc etc you get to realize that the design of these things becomes extremely important oh 100 in terms of in terms of what comes out the other and enough to maybe leapfrog you know it can new ways of of you know doing analysis like you said the decision trees and you end up in some place entirely different I think I think that's just like nice and close to your mouth yeah I think it's true um so if you can very quickly size up your options in a way that's more concise and the and and in terms of design I mean there are two there are a couple of independent aspects of what we're talking about for design in terms of applications one would be sort of the data architecture and another one are the designs that allow us to integrate to them with human factors and the human factors like the user interfaces the quality of the documents how they come out of them are the documents close enough to being drafts for a particular purpose a particular court a particular phase in litigation the negotiation sure and the ability to do those things with a high level of accuracy and then as as does it was suggesting like or you were also suggesting like as this thing becomes an assistant to you all of a sudden your time is being spent not in fact um doing something else but rapidly internalizing what the what the salient issues are in in the law that we sometimes call that um with technology practicing at the top of your license and not you know incessantly like playing with margins of Microsoft I wouldn't know about that well and I think more like assembly yeah document assembly at the bottom like assembly oh oh right assembly language so there is one point I'd like to make but like uh in so in the job I had before this we were I was at a company called risk genius and we were breaking down insurance policies and we had developed a taxonomy where we could categorize the clauses of the policy you know by line of business by whether it was you know a certain type of clause so definition exclusion ensuring agreement condition so on and so on and then within each of those you could go on down you know a few levels and one of the things that we were able to do that was like very meaningful for the insurance industry was we were able to figure out you know so say a new regulation comes out and that impacts one clause in every policy but you don't know how many policies it in it's in because so many of these uh insurance carriers have only gone digital very recently like how are you able to go through that uh the the 200 000 policies that could have that language in it and quickly identify exactly where that clause is uh you know get an endorsement so that that clause is changed and then reissue the policy to that individual and what we were able to do is based on this taxonomy that we developed we were able to go and have the policy uploaded you know run it through this machine learning algorithm and identify it much quicker than you know a human ever could and so I I definitely agree it comes down to the to the you know almost like developing a taxonomy for how that data is represented developing the kind of like populating the leaves on the tree the the computational layers around it so that it is much more amenable to those sorts of processes but then it all comes back to you know you have to have somebody look at that and make the determination of oh what do we need to issue in order to replace this what do we need to issue in order to you know make sure that we're adequately covered it there is still going to be that that role it's just going to be a different role and and uh uh kind of glib analogy with somebody uh mentioned to me one time was that if a calculator hasn't replaced an accountant you know one of these ai's isn't going to replace the lawyer and so I think it's it comes down to like a tool selection issue and you know each tool has its strengths and weaknesses and so I I hesitate to like look at ai as you know the the kind of the other analogy of like to to the man with a hammer everything looks like a nail I don't think we need to kind of we certainly have enough of that with ai but I do think in the example you gave was a really good one because the chances are that it's vastly more efficient but not only that more accurate yeah yeah yeah all right so yeah please yeah because we have people online that aren't able to um you know like uh see each of the bench here so uh Michael um I believe you you had something yeah look I don't disagree I don't disagree with anything that's been said here I just probably more from the perspective of a from a practicing lawyer and I spent a lot of my time doing contract analysis and drafting um I I certainly think the tools that are available at the moment there are a lot of useful ones out there I haven't used uh little geeks out there but um but certainly I think there are a lot of tools which can be useful for flagging issues at the moment but for me I think the largest part of my role in any sort of contract analysis or drafting perspective drafting role is more about a it's more of a linguistics and semantics aspect of looking at looking at language and whether they're drafted in a way which is clear and concise things like consistent terminology and um ensuring that say defined terms are used in the correct in the correct order and if there's anyone in the in the in the room who's uh I don't know there's a few people who've got some linguistic and AI backgrounds as far as I'm aware like that's sort of a task that look at sort of higher level cognitive reasoning process where it's a action linguistic semantics of a spoke document which something which is very different than what a system has otherwise already been trained on it's not really possible to do with any sort of AI system at the moment and that would be the big breakthrough of a time consuming and reducing volumes of work is frankly that cognitive understanding of of the and of readability as opposed to purely rule-based systems and flagging say contracts got an indemnity so that's a bad thing flag that um that is where the bulk of the time is actually state yeah and I think that also gets to the the notion of having like standards and like standard clauses and you know uh sally sali they have the legal mat legal matter specification standard um that uh purports to uh you know represent common legal matters as uh as though it were like something in supply chain like a code a transaction code in a supply chain and so I think uh I think that's the direction things are heading um and and we I believe we touched on that later perhaps in this we should we should circulate that maybe after class but sally just came up with a new standard which is pretty complete for identifying parties and the type of legal matter and a whole bunch of other stuff and uh it's I think it's like a um honeypot for analytics and automation a couple of things that come to mind um for what um it's Richard yeah uh well it's only that Richard said about um you know doing initial analysis better um people heard of Kira systems Kira so about half in the room have uh we can send a link but there's a several vendors on slide 17 will do a breakdown that do a pretty good job of initial um issue spotting in contracts in particular in some and other kind of expert areas there's also this really interesting tool that I saw at um um future law conference at stanford this is um annual codex conference that they do um called lot to mation um and I want to see if we can get those folks in to do a demo have you seen that yet it's pretty remarkable it's really cool stuff and so like you can it can among other things um get you to the first draft of like pleadings and um and like a complaint or or response to a complaint and litigation again back to something that Richard had said lot lot to mation also kind of a cool name it looks cooler than it sounds I think but um but it uh but what's really interesting is they go really deep jurisdiction by jurisdiction so they have a handful of jurisdictions right now um and and uh when you look at it like it knows kind of what to look for in terms of like the and what the points are you're going to want to prove what types of causes of action you might want to be looking at based on the facts and then it and it does a really good machine job for doing a response to a complaint so that you make sure that you've you know you've alleged everything you have to or state everything after you to make your best case for summary judgment or to find any affirmative defenses and to kind of say everything in the right way and so it's just kind of like it's really remarkable uh what it can do uh or what at least based on their demo um at least uh so uh where are we now we are one place we are now is at 337 p.m eastern time which is uh precisely seven minutes after we promised that we would end we've always done this class up to five p.m um and then frequently actually ended at like six or one time near seven p.m i think with people really cranky and babysitters you know sending emergency beacons and things like that and so we thought okay let's pretend to end it at 330 this time and then see if we could beg people to see how many people would be willing to go a little extra like extra innings like baseball because it's fun uh so we wonder um are people um online um would people be willing to go do you think we can whist through these before four yeah we can we can do really quickly like the skip across the top of the waves is anyone online able to go another 20 minutes or if you're not uh let us know and people here in the room if like unanimous consent can we go till 4 p.m we're gonna have to speed is that possible you have to go right hit the road did this explain anything about what you're interested in with what computational law yeah i said uh one thing that was interesting if you have this competition a lot you have this code yeah there's a there's a really cool company called lexon that uh they they have almost like a it's not a compiler but it uh it's almost like an editor where you can type in uh contract obligations and it represents them as though they were like smart contract code for like solidity and so you can see like what changing like one clause of like human text does to like the machine text and so it's kind of a really fun way to to get to uh get to that and i'll slide 27 okay um so just before you go since since i now this is the first time i found out what you were interested in so just before you go here's a beautiful slide just for you um so one i'm gonna call it a design pattern that we favor uh is um for short we call it like blt like a bacon lettuce and tomato sandwich uh but it stands for the business uh legal and technical let's say dimensions of like information uh and so we think uh when there's a contract or a legal process even a statute um that um we want to see it expressed um in machine readable code for sure don't miss that or it won't be computational in human readable code and by that let's just say plain language like maybe twelfth grade like um filter it would it would qualify as like um um twelfth grade i think is about the right level for human readable plain language and then legal code which is not twelfth grade um plain language it's actually of necessity fairly dense um and um uh it's dense and um um it has a lot of jargon in it and the jargon has meaning um and so um creative you know creative commons license um so the creative commons license is not a bad example of that um it has the legal code where it goes through all the edge cases and intellectual property for that type of copyright license um it's much longer it has the so-called human readable plain language code which is like less than a page and which can even be reduced down to an icon um and then it has the machine readable length layer which people don't use much but it they've got an interesting um uh what is that triplets language rdf uh they have an rdf version which is machine readable so leaving aside you know their specific machine readable like i would choose something like json or something that may be more commonly used maybe xml is a good good match for machine readability or whatever the right machine readability is for the legal thing that you're doing but by um identifying each one of those layers of the the rule or the legal instrument or whatever it is and then um um harmonizing them so that they're they have equivalents aligning them and harmonizing them and then ultimately integrating them so that they're three aspects of the same thing you know what you can do with like section headers or you can do some of that metadata but it's one thing we think is is a good way to ensure that um uh that the that the computational legal system serves the correct serves a full purpose for the machines and for the humans both kinds of humans the regular humans that you know that need to operate with law as well as the legal system um so effective translation yeah so on different projects i've been on mostly in consulting over the years but also in MIT um it depending on what what it's about it may be emphasizing one or more of those layers dependent and that's very context specific but we actually think there ought to be better reusable common like design patterns and templates and frameworks so that you always explicitly i um solve for each of those three layers in a unified integrated way does that sort of that's hard um so you like increasingly have to be tri-lingual or or but do you remember remember joey ito um and so he used to uh you know he had that phrase we want to be anti-disciplinary or you know maybe cross-disciplinary trans-disciplinary interdisciplinary i think like one of the ways that i that i tried to make sure that they're equivalent when i write rules for um systems so i've done a lot of supply chains payment networks and other types of federations is i will if it's if it's a big system i will actually have a single document and i'd put this in the scope of work in the beginning when i'm working with somebody to help them architect and write it up uh the first section is called is the business stuff so um so it's like you know like who's in and who's out who's who's paying um you know business practices a few things like that the second section is the legal stuff that's like 80 liability usually but ip and you know other lots of little legal things notice whatever uh indemnification and then the third section is the the fun stuff like the technology stuff what standards are we using how do we connect things how do we you know up rev what are the testing what are the security requirements and so forth and then i have a single table of con a single table of contents for the single glossary that's the most important thing and so to the extent so that and then i'll try to break it up so that um there the it's like that the technical people have primacy over what's in the technology parts they're answerable to a cio or cto or whatever the legal people are answerable to like a general counsel or risk management officer compliance the business people usually go up to the ceo or c o or cfo but that there's someone that's really in charge of each one of those and before they up rev their section of the document they actually show it to the other groups and we kind of circulate it back and forth so that if something's changing in one place we understand the implications in other places and to the extent possible of any people from any one of those sections areas should be able to read the whole thing and have a real good idea of what's going on um even if some of it's quite dense and then sometimes we end up using the same word to mean different things and that's okay we define that when we do uh you know but the idea is that all that there's a single unified glossary or um and that there's one table of content so it you don't so that the others the other parts of the system aren't invisible or opaque or confusing to people but even then it's like there's a lot of like that there's a lot of it takes a while for people to get the hang of things that they don't yet know or understand uh and so people that are in the business or the technology kind of the aspects of a project aren't kind of become like legal scholars and lawyers aren't going to become computer scientists and so forth uh but they need to be conversant enough to see the interdependencies and to be able to come uh and part of the way I try to facilitate that is by putting in a single document and circulating it a lot until we get alignment harmonization and integration won't you please here you go so so this is a bit technical but this has to do with also the the translation notion that one of the participants was talking about and so you can have um you can have translations at the sort of term level where uh between technical languages and natural languages where you think things are aligned but it's in addition uh to the terms at the on the technical language side it's the rules of inference that also matter that can lead to strange consequences so there are there are well known things like um so there are the deontic logics logics of like obligation what you must do what you can do and so on and there are well known things like the uh the miners paradox where translating these you know these uh innocuous looking sentences in English into certain deontic logic frameworks so the technical expression of those things it looks like they're lined up so at the term level you can say okay yeah this this sentence says what that sentence says but because you've embedded in this language that has these rules of inference you infer these weird consequences uh that are not possible in that you don't infer in English so it's not just the alignment at that sort of term and meaning level that matters it's also uh this um these inferences yeah in the background logic such great point not so fast mr stay right there um but uh just a uh some one so i i kind of blended a lot of stuff together really fast but alignment that i'd use in the beginning because just so we can begin uh and then when i say harmonization i actually mean kind of i don't want to be too fancy with where it's beginning to a more of a semantic level and then one of the ways that we test that i've tested that is um and i'll share this in our telegram channel i've got a great rule set in from the insurance industry for id federation where they permitted me to publish it under creative commons so we can look at it as an example but basically we have a set of use cases that are the approved authorized use cases that describe what we think will happen when this set of circumstances comes together so basically we have this it is not like a formal verification by any means but we have a way of testing what the probable legal and technical functional results will be against like situations so or um i think richard used the word frames uh from but anyway we use like use cases and legal fact patterns like a blend of them and then we run the ones that people care about against each one of those layers to see what would we think the results will be and then we try to harmonize that is that word i use but then at the end of the day though a lot of this stuff is so squishy because we're not dealing with an entirely constrained context like things are dynamic and we may not have thought of all the use cases that will become relevant in a year or two from now um at all and so there's i think there's a lot of unknowability at the edges which i you know it's largely unknowable at the edges but what we want to do is reduce the darkness you want to increase the likelihood of um of correctly assessing what the legal results will be you know what the business results will be and what the technical results will be in the system so here's sort of like a way to begin to get there one fine day and this is where i wanted to ask you not to leave yet uh i i wonder if we couldn't if we actually had an area of law that let's say through like federal legislation was like complete so anything else was preempted so this is like the complete area of law to just start with that assessment uh and then if it was fully computational like let's just say like all the statutes the regulations even the case law was computational if we couldn't actually then create like a legal instrument or structure of transaction in a way where we could formally verify every permutation of things that could possibly happen and um and know in advance what the legal results would be is you think is that doable brendan just uh noted that this is where simulation against use cases becomes very interesting just so um yeah so one day so tomorrow as a quick look ahead we're going to have sandy pentland come and talk about his um i'd say like groundbreaking article of perspective on legal algorithms and part of that's going to be talking about literally how do we architect and engineer systems so that we could um meaningfully measure everything um and start to instrument the system and um and then be able to adapt in real times on it yep all of that great stuff but but when you said yeah do you think it's a realistic goal to have like as part of computational law in our program here formal verification or do you think that's just too far out and we shouldn't even bother talking about it and could you grab the mic yeah i don't have any you know i don't have a concrete area of law in mind um but i mean it does seem like you know tax is one of the most more likely uh candidates that some of the more you know um objective kind of parts of the tax code seem like they can be completely formalized like that and so and then do you think there would be a way to express it so that through software and people's understanding that we could surface and harmonize the like second level inferences you're talking about in machine logic so that it matches what we're saying in natural language or is that even a realistic or desirable goal uh yeah i think it's a i think it's a desirable goal and i mean you know so in logic there's you know this these notions of soundness and completeness so completeness means that everything that you think should be inferred can be inferred and soundness that everything that can be inferred is a valid inference and you know you can prove both of those informal systems so um so it gets down to being really explicit about our priors right and uh making sure that they're correct the axioms and the rules of the axioms thank you um did somebody uh okay so i think we're we're now we're now back to zero we have no backlog um i think i answered your the question that you had before you stuck away you're free to go if you want to uh but uh we actually have a delightful seven more minutes to like the best of well we we have to we have to go back as well thank you cheers um so the initial adoption um as we can kind of see represented in the next slides uh looks at kind of like duplicating processes as like a paper-based paradigm um and so you know digital formats obviously enable a lot more than that but the uh you know the really the direction that we're headed is you know to have things that are model-based algorithmic um these kind of like modular little containers of services that can be configured and compiled together in interesting ways so it looks like going from paper to you know the notion of uh oh and this gets to re-engineering so um so going so there may not be an equivalence this is very much in tune with what um cool ryan just said about you know the things that may appear to be the same but there may be real differences with the outcomes when in the beginning of aeronautics when we were trying to you know make things that would allow people to fly people there's a lot of evidence of designs that look like birds of people some people can we need like wings that would flap because see that's how birds do it and um i don't know if you've ever seen like clips from like the tens to 20s with people you know crashing at the bottom of cliffs trying to get these flapping wings and then and then the great insights were made about fixed wing and you know we made a lot more progress so we found a way to engineer in a sense re-engineer based on the materials and the resources that we had in hand ways to achieve the result we were seeking but it wasn't a one-to-one like extrapolation from you know from some um natural system and so you know some business process and other re-engineering is going to be needed for sure just to get the intended results for computational law and click so where are we now and oh yeah this is kind of i think the slide that does it was referencing earlier but Codex has this legal tech index that characterizes the state of the marketplace for a lot of legal innovations and they have broken it out into these nine categories and i think they may have a few more but you know you can see marketplace document automation practice management legal research legal education odr you discovery analytics and compliance and there's some blur so there's a number of products and services and projects that we'll cut across a lot of these categories but these do seem to be if we just start from an analysis of the marketplace of apps and services and projects these are pretty solid categories and they built up from the ground up by literally going through and do hundreds of thousands now i think of entries of companies and products and projects and so the the kind of next direction that we're headed is you know expressing law as standard data in these interoperable service interfaces so that you know these great things can happen like updates are provided alerts are provided you know they're internal external linking of you know these different applications you can start to imagine you know chains of these apps and services that are connected to form you know a network of computational systems so that it's easier for people to um interface with the the legal system and with a better user experience so what does this look like you know this is kind of a good representation as a sample json schema and xml this also looks like uh uh you know having an api with you know different documents you could have like a rest api and you could be able to like get and post all of these uh so this is like some documentation on like a generic api so you can imagine like what would that look like for the irs like you know get the formula for depreciation on something and plug it in um and then i've been the right as the xbrl um so the extensible business we love that rule yep extensible xbrl extensible business rules language okay language um and it's a dialect of xml that's really designed for financial accounting and when you're analyzing scc filings that have been filed um and tagged basically with xbrl it's it's awesome including for investigative reporting but also just to get inside of your investor or if you're a regulator or an law enforcement of any type um it's just great it and especially for the tango of footnotes part of the reason why scc adopted xbrl was a reaction to the enron debacle where you know i think technically some of the worst successes of like the shell games were technically disclosed if you could piece together you know the Byzantine chain of like uh footnotes that referenced other footnotes but but humans really could barely comprehend that and that was somewhat deliberate as it turned out but when they're all tagged um you can very rapidly see how these things fit together and and and and you can go you know faster further and deeper and especially when there are standards as well about you know what what can happen um right number one two portions number two um um it's voluntary still um and so further um filings that are xbrlizable um uh only us a minority of publicly traded companies um a small minority actually do mark it up that way although um hester remember her last name pierce pierce hester pierce one of the sec commissioners he's great having young brain uh uh is came to mit not long ago and we also saw her in chicago uh like a month before and i had a chance to ask her um about xbrl and whether they were going to be pushing this to make it more widely used and um she um was happy to hear that people cared i don't think they've heard a lot from people that are seeking xbrl so like this isn't high on their list for user demand but but she noted it um and i said it twice in two different cities so i don't know uh so everybody tell hester and the other sec commissioners how much you love xbrl and let's see if we can't get it more broadly used um cool brian understood that it's going to be mandatory at some point in the future well you know from your mouth to the regulator's ears fingers crossed i guess um and then alexander pointed out that there's a link in the zoom chat that we can throw into the telegram channel that has a list of the market for nordic legal tech as well so we can kind of do a comparison that might be interesting um so getting kind of a little bit down down the stream you know they're they're um also this is kind of just a pictographic representation of what we've been going through where it's you know these interoperable services with an api layer communicating to these different different pieces um you know you start to have sensors sensors trigger rule tool to trigger workflows and workflows trigger actions i think that's kind of the essence of you know how do we get from like something that exists as a static process to something that achieves a real-world outcome through the use of line computation in a sensor here could be as simple as like a listener and a web server so it's like did somebody do a filing oh here's a new file or did a new client come in and just give me a new nda to look at or whatever it is so something happened uh in time and space you applied rules you had some kind of workflow to go from this intake and then i put it into my process and i did a thing um and that took an action like i sent a letter of representation or i sent a markup of the nda or i sent a yep that looks fine sign it or whatever so and you could generalize this very generic um kind of like explanation of automation to all sorts of legal instruments and processes we think and but looking at how to put things in boxes to make them modular we think is is a very important way to make this something that can be architected and where we can start to um get more power out of doing more defined capabilities and sequencing them correctly without having like one big fat code base do a lot of things and not being able to change the rules especially notice how the rules are in a box oh how i want rules in a box so that you know when the rules change we just change the rules and when other things change we change those things and we can model stuff right now a lot of legal software is a lot of software this legal tech and just a lot of software in general combines all this stuff in ways that are totally interdependence like spaghetti so we think more modularity is critical for um for like a usable unscalable um uh adoption of computational law yeah and i and i think it's important to point out that sensors in a certain sense could be actions that are triggered by you know the attorney themselves so you know send a client engagement letter that could be a sensor to create a new file and then the new file once it's created has a has a set of rules that indicate that you should you know check off one of you know four or five boxes that are relate to the type of service that it is so you want to form a company and then that you know forming the company triggers the workflow for okay we need to get an EIN we need to file articles of organization with the secretary of state we need to create a operating agreement we need to um you know do have we have these optional triggers for amending the operating agreement or filing the annual filings and then even a safety valve for dissolving the entity completely and and so once you know you have that workflow you can start to call on those different pieces and it gets you to those actions that are at the end of the end of the chain and when I was making this slide I was like I was going to get super fancy making a circle because some people's actions are other people's inputs you know uh or like you know some outputs or other inputs so uh so that's where we get into the idea of chaining these things okay let's move forward okay um yeah so sprint yeah so the the desirable destination for this is I mean that you have get your version control systems where you're actually able to go through you're able to edit you're able to see all the changes that take place and you're able to develop a greater context for why we have the things that we have now yeah so just for the publishers in the house and others like right right now one of the okay let me start by saying um the Microsoft Office suite obviously is a triumph of our species and Microsoft Word is amazing in terms of what it's made available so you know hats hat tip and it's also become somewhat of a prison um so having legal documents and um and and everything locked in PDF and Microsoft Word is not it's digital but it's hardly computational what we really need to be able to just break it out treat as data and also have version control on it um and and so something like the get protocol really gets us there so when we are do the um when I do legal contracts now if they involve a lot of parties and I'm involved in it I will put it in in get hub uh for example or uh increasingly pit bucket uh and that way we can actually see exactly what version we're working on we can branch it so if people are looking at like well what if we did this a little differently we can put that in a different branch and play it out and if people really like it and they want to agree to that we can merge the branches later but um but the first like 20 minutes of most meetings I've ever been on with a bunch of parties negotiating a contractor when I was doing government stuff like a regular version of regulations wait what version are we on oh I emailed it last Tuesday oh is it final final 3b um no it's the other one which one oh it's no final final 3a but I actually changed it but I didn't change the wording and look at the date and then by the time we have an idea of what document we're even talking about that's a real bad time uh but then also you have to look at the comments and there are all these comments on the side it's okay if you have two or three comments but it's not actually data it's hard to export those differently it's hard to check them off um what we really need to do is treat these things as as like the same way people treat software uh put it in a code repository so we know the versions we could see issues we can do branches and forks we can actually manage the information appropriately and make sure it's all hashed so there's no absolutely no question about you know what version it was this is like old obvious simple things if you're doing you know like a ten dollar you know JavaScript app but this is brand new territory for the law and this would turn out a lot of the underbrush that is holding us back now so that slide on get it's like they're just it's boring it we shouldn't have to talk about it so much in a class like this but it's almost a matter of theology that we must spread the good word about the get protocol and it's and it's criticality for computational law okay so in the so picking up from there the the the goal that we see is uh to make it so that legal content is created collected in standard formats and data structures that can be displayed as legal processes rules and that can be understood by anybody in plain language usable by lawyers and processable by machines so getting back to that uh uh slide that we talked about earlier where it's like got the three layers the BLT layers and then you know from there there are a lot of interesting things that can happen so we have uh oh good good point uh it's now 407 and we we're we're coming in for a landing um and so uh you know the notion of uh how law is practice is going to change with the um kind of advent of uh Alexa and these human computer interfaces um it's also you know the the way that we imagine it is a little bit different and this is something that I referenced earlier where you know you think about autonomous systems you know that's not really the goal that we've gotten in mind we've got really this extended or augmented intelligence kind of idea where everybody is able to kind of leverage the technology in order to practice at the top of their license we call it um extended cognition at the media lab and so the idea is it's extending the capabilities of both not replacing people uh and and so some of the success measures of what computational law or what ideally computational law could do include achieving predictable legal outcomes um simplifying identifying um kind of these rules and processes so that they're easier for humans to understand uh simulating you know what happens uh can we you know fully comprehend some of the formal logic um of all of this and then um measuring the effectiveness and verifying that the the results are what we think they are and so the the the kind of architectural stack that enables this to be something like the OSI seven-layer um stack of uh you know you have an application you have a presentation you have a session um you know so one and so on and then the one on the this is the last slide the one on the right is um is an sort of a extrapolation of the OSI seven-layer stack uh to business and this was a group that used to participate with a lot in the in the 90s called commerce net uh and this one represents um a pretty clean high-level architecture what we thought e-commerce was going to look like but it's that's sort of adjacent to law and so kind of decided there's like networks like the internet or something and that contains markets you could say and markets contain businesses like a business would have identifiers and then businesses conduct like service have services and services uh uh can uh conduct interactions or transactions and other interactions and those have documents I would call those better record now probably as opposed to document so it's a little less paper concept and then instead of information items I would just say data in 2020 as opposed to 1996 or whenever we did this ancient times um but but it's not so different so here's another way to look at modularity or I'd say layering a stack and to understand the kind of when we're creating computational legal systems which once fit in what stack and keep kind of keeping those swim lanes um able to move independently we think is is a good clean architecture yeah and I think um oh I thought that was the last slide we've got there are a lot of slides in this deck okay let's make the next one the last slide okay well I think we should have made the last because these are so these are the slides that are from uh the presentation that I did last year so oh that's on um uh doc assemble yeah on computational on doc yeah so let's save that for Thursday I think with the community lawyer people so we're gonna stop here then so da da there you have a computational law and a sprint yeah so yeah that was uh approximately half the slides yeah and these slides are actually like a very high level summary believe it or not yeah so um so with that um I want to say thank you very much for allowing us to go over time tomorrow um there is readings which we'll send in the telegram channel everybody should read it's a short wonderful read five pages sandy petlins um uh ground baking breaking article a perspective on legal algorithms um and it's in the uh blog dot mit dot edu it's the first article I'll I'll throw it up there right up there uh and then we'll actually have sandy with us um uh and he'll go over um the the essence of of that um vision of for computational law and then what I think we're going to do for tomorrow is is make the tables into a circle um and uh make sure that everybody online can see everybody here uh and then we're going to spend most of the time in a dialogue with sandy um for a lot I think we'll have them for a full hour so we will start on time tomorrow and then um and then the second hour with the group is oh a second hour is going to be uh talking about economics and computational law with uh with the digital economist that group yep um a lot of fun yep thanks for your flexibility uh I don't know to answer Michael's question I think we'll probably have a different zoom link um and we'll be sending that out uh five or ten minutes before the session starts so that everybody can access it whether you're here or remote stuff um if uh you know being remote is the easiest option definitely do that and even uh even for the people who are watching this on delay uh we know there are a few people who are um who are in time zones where it was even more inconvenient um than 6 a.m or 5 30 a.m like it was with Michael um so uh they'll be watching this after the fact and so you know feel free to pose questions in telegram as you have them and we can uh kind of continue this conversation a little bit asynchronously yeah and we and we decided to do a bonus this year too so because some of this is going to be new and may have ideas we're going to encapsulate the sort of big questions that we identified together over the next couple of days and then pose them in our telegram channel and on the class website and then you could just call it a day at that point or if you're interested and you've kind of met people or you yourself want to think about these things a bit of an opportunity to um then write something up like a blog post or write up some code or something that is your take on or write up better questions or even a critique um and then to kind of present that on right now tentatively it's January 24th but in the last day of class we'll check with everybody what a good day would be uh including Sandy uh and see if we can uh then have people do like five or eight minute presentations of their thoughts on this uh either their idea or their question or what have you and then we'll have a chance to have discussion around each one of those things so we'll have like a cap off section or something like that at the end of the IAP at the end of January and hopefully that will even the playing field a bit for those of you in uh in all the different time zones of the earth okay with that I think uh I think that does it yeah we need to gavel this session is adjourned thanks everybody