7 ms·

I'm currently working on a startup which supplies automated legal document analysis for the legal industry. [1] So I definitely see the power for AI for transfo

I'm currently working on a startup which supplies automated legal document analysis for the legal industry. [1] So I definitely see the power for AI for transforming the legal industry, but also encourage caution in reading the report. One thing to note is that a huge amount of legal work done today is actually quite possible to automate. It's a lot of reading through documents and finding the few clauses that actually matter to a contract or document, then focusing in on those. It's defining an overall strategy and then fleshing out (mostly templated) documents to implement that strategy. Another factor in favor of legal AI is that legal documents have much more restricted grammar than standard English text, leading to less ambiguity. This makes them both easier to parse and to create automatically (you don't need to make the text sound "natural"). By 2030, AI will be able to easily summarize a full legal document and provide relevant references. (In fact, probably much sooner.) That being said, it's important to note that this report isn't predicting the disappearance of lawyers (nor am I). Instead, it's the collapse of traditional law firms, which depend on a pyramid of partners doing high-level sales and strategy work being supported by associates and paralegals doing the grunt work. The grunt work is rapidly becoming automated, but I don't think AI will ever replace the client interactions and recommendations component. We've seen a lot of success with AI-lawyer pairing and I do think we'll see much more of it in the future. This definitely challenges the organizational structure of law firms. In the future, I think we'll see a lot more small and individual practices where clients work with a partner they trust who can get a lot more done thanks to AI doing all the grunt work. [1] https://docnavapp.com/ https://docnavapp.com/

jbhatab 11y ago
Why isn't what you described possible with AI?
- morgante 11y ago
  The high-level work? Let me first say that I don't think we will have human-level AGI by 2030. If we somehow do, all bets are off. The high-level work requires a large amount of creativity in developing nuanced strategies for a particular domain. It also involves integrating knowledge and ideas from a lot more than the law. Consider that the vast majority of intelligent humans who attempt to make it to partner don't make it. Top lawyers are highly skilled, and automating them away is not trivial. Also consider the trust factor. We've done some customer discovery in this space and most people would just not be comfortable working directly with an AI. This is especially true because a good portion of what partners are doing is human contact work with clients, courts, and opposing counsel. There's a lot of leverage which comes from being able to read and manipulate people which AI can't replicate easily. Note that I'm referring to the 2030 date specifically for this. I don't think lawyers (or, really, any current profession) will exist in 2100.
  - TheOtherHobbes 11y ago
    I don't see why "Uber for lawyers" wouldn't work. I'm imagining a combined semi-automated service where the paperwork side of law - simple wills, notarisation, simple divorces, and so on - is completely automated, and the creative elements are farmed out to individual attorneys with a public track record of success who compete on price. It's going to kill the traditional law firm business model stone dead. Lawyers are basically paid to logic-chop precedent and written law, and to be persuasive and/or intimidating in more or less creative ways. IMO eventually all of that could be automated. I don't see any reason why AI couldn't potentially have super-human powers of persuasion and manipulation. It won't happen by 2030, but the next step are document searching is to bundle up the basics of law as a low-cost automated service, and then start automating the process of argument, cross-examination, and persuasion - which will lead to an arms race, which will lead to very rapid advances, which will lead to complete automation of civil and criminal law by (I'd guess) the end of the century.
    - morgante 11y ago
      > I don't see why "Uber for lawyers" wouldn't work. Primarily because people generally want to choose and talk to their attorney. It's basically the opposite of taxis, where people didn't care who the driver was and (generally) preferred minimal interaction. For Uber, not having to interact with your driver or choose them is a benefit. In law, it would be a detriment. That's why I think the future isn't a monolithic application where you interact with a faceless half-human, half-AI lawyer. It's small individual attorneys accomplishing a lot more without needing expensive support staff thanks to AI. > complete automation of civil and criminal law by (I'd guess) the end of the century. Okay, sure. In the longer term, every profession will be automated though.
      - fauigerzigerk 11y ago
        >Primarily because people generally want to choose and talk to their attorney Yes but why do they want that? The reason is not just some emotional desire for hand holding. The main reason is that AIs are simply not (yet) very good at language related tasks, and there is not nearly as much progress as the current bout of hyperactive AI reporting suggests.
tensor 11y ago
Systems that use machine learning to do document review actually already exist. E.g. https://kirasystems.com https://kirasystems.com. They currently still need a lawyer in the loop to review and adjust the results, but eventually will be accurate enough to produce summaries that are more accurate than a human without any intervention. Traditional law firms are starting to invest in software like this. While it can be used to reduce hours, it can also be used to sell more by increasing the value of the hours spent. For example, instead of doing a review of a subset of a clients documents, they can offer to review all the documents for the same, or even slightly higher price.
- morgante 11y ago
  Yes, I'm familiar with Kira (they're in a similar space to us). The technology isn't really doing summarization yet though, it's more like extraction/TOC building. I'm definitely biased, but I do think that's the direction that legal technology is going. Right now it's about making it easier and faster for lawyers to review documents: eventually it will be about them not even having to review documents at all.
  - semi-extrinsic 11y ago
    I'm curious: this discussion is all about improving software that understands legal language. Are there any (long-term) efforts to improve legal language to make it easier to understand for software?
    - rwmj 11y ago
      It's my understanding that this project is doing that: http://codex.stanford.edu/ http://codex.stanford.edu/, trying to turn law into something like computer code. See also: http://logic.stanford.edu/complaw/complaw.html http://logic.stanford.edu/complaw/complaw.html
      - conceit 11y ago
        There seems to be a bunch of projects. Lacking an AI to do the job let me extract what caught my eye and seems to be answering the question. >The Hammurabi Project is an experiment in converting legal source material to the C# programming language. [...] The idea is to grow a body of source code that mirrors the structure of U.S. federal and state legislation, regulation, and case law. https://mpoulshock.github.io/hammurabi/ https://mpoulshock.github.io/hammurabi/ >The Stanford Computable Contracts Initiative (SCCI) is working on legal technology that will help move the world from natural language based contracts toward a world of computable contacts. > The project [Designing and Understanding Forensic Bayesian Networks with Arguments and Scenarios] ’s new approach is to link the successful statistical modelling technique of Bayesian networks to models that effectively dovetail legal argumentation and scenario construction in the legal world. http://www.ai.rug.nl/~verheij/nwofs/ http://www.ai.rug.nl/~verheij/nwofs/ >Instead of automating legal work, I intend to focus on augmenting legal work. Specifically, my project [Computational Linguistics and Effective Legal Drafting] focuses on using advances in computational linguistics technology to help lawyers draft more precise and error-free legal documents like contracts, regulations and statutes.
HillRat 11y ago
Fascinating! I do wonder what happens to the traditional white-shoe partner track, where associates make their bones do sixteen hour days of grunt work (albeit with fantastic compensation) as a method of gaining experience and reputation. The skills of a high-level attorney, regardless of the practice, are honed through decades of experience. How would systemic AI affect the development of the profession?
- knughit 11y ago
  The skills of a high level lawyer is in sales and negotiations and trials, not in the document analysis that junior lawyers do.
- rms_returns 11y ago
  Don't worry, the AI isn't going the replace the "Perry Mason" kind of lawyers, they are irreplaceable! AI is designed to replace the "Junior Grunt Joe" kind of lawyer/associate who is mostly involved in documentation and assistance.
  - rodgerd 11y ago
    And where do the next generation of Perry Mason lawyers come from, if there are no rungs on the bottom of the ladder?
    - js8 11y ago
      Maybe they could actually help the poor?
    - chongli 11y ago
      Hence the article's thesis: Artificial intelligence will cause “structural collapse” of law firms by 2030.
    - rms_returns 11y ago
      Well, that logic applies to all fields, not just legal. Lower rungs are disappearing from everywhere. In accountancy, junior accountants used to manually do "debit-credit" first, now Excel does it for them, so what will they learn? In IT, junior programmers used to toil hard with algorithms and data-structures to come up with barely working programs. Now there are IDEs, libraries, frameworks, CMSs all built and ready for them, so what will they learn? In Physics and Math, Scientists worked hard on just the theoretical framework first and scratch their heads in classrooms. Now there are computer software that do it for them. Perhaps this explains we didn't get another Newton/Einstein since last few decades?
dinkumthinkum 11y ago
Well, you certainly have a "dog in this fight." =) It is a little self-serving to say "Yep, this is possible and we are the ones doing it." But since you have such a startup, you'd be crazy not say that! But as I have stated elsewhere, document summarization and classification is not even in the ballpark of an AI lawyer. Excel didn't replace the Accountant. It's like a fancy Excel for clerks and paralegals (and Excel was a wildly successful product). You sort of admit that but I think it is more than just the interactions with clients or making them "feel good." I don't think it really addresses what lawyering is or how positive legal outcomes can be achieved. This ma be a weak comparison but it's like we all know that the point of programming is not writing code and cleaning up memory; it's to create solutions to problems. Also, I think the traditional structure of the firm serves many more purposes than just specifying a hierarchy for task assignment.
- morgante 11y ago
  > It is a little self-serving to say "Yep, this is possible and we are the ones doing it." Fair enough. :) Though I tried to be a little more nuanced than that. To be clear, I don't think AI will replace lawyers any time soon. It's just making them more efficient. Our customers are layers, not people looking for legal services. > Excel didn't replace the Accountant. Great analogy. We don't think AI is going to replace lawyers any time soon, but it will make them more efficient (and potentially eliminate some of the lower-tier staff, like what technology has done to pure "bookkeepers.") > I think it is more than just the interactions with clients or making them "feel good." Sorry if I didn't make it clear, but I do think lawyers still provide a lot of value which AI can't easily replace. There's a reason we didn't have an AI write our equity plan. > I think the traditional structure of the firm serves many more purposes than just specifying a hierarchy for task assignment. That might be true, but remember that associates are expensive. If the budget for them has pressure applied on it and with AI you can get away with 1 associate instead of 5, that's going to change the legal track substantially.
rhino369 11y ago
The true value of a lawyer is their judgement, analysis ability, and their power of persuasion. To replace that you need a true Strong AI. So I expect lawyers to be replace shortly after everyone else is replaced. Give or take a few years, since lawyers have some protectionist options, like banning AI from being a lawyer. Weak AI will eat around the edges. Better spelling, grammar, and cite checking software could reduce a lot of billable time. But most associates spend less than 10% of their time doing that. For big litigation, document review has been mostly outsourced to legal service companies. I've been a junior associate (the ones traditionally responsible for reviewing documents) for about 2 year. I think I've billed ~200 hours to that (out ~4,200 hours I've billed total). Client won't pay our rates. So we hire contractors for 25 dollars an hour. There is still room for disruption. Predictive coding will probably become the norm within 15 years. But it's only going to put the temp "doc reviewers" out of job. Westlaw and Lexis already summarize cases, index them. ALJ and Mathew Bender sells practice guides. Firms subscribe to proofreading, document processing, and printing services. They hire paralegals and assistants to do the less taxing work. It's also important to note that business model of firms only relies on associates billing hours because it's convenient. Clients don't pay 500 an hour for an associate to make sure commas are correct because it costs that much to provide. It's just an add on service to get the high level talent's analysis and judgement. Nickle and diming essentially. Associate costs are fairly small portion of current billing rates. Most of the grunt work I do is bespoke. Stuff like comparing one patent to another. We have google patent and questal to search. But it takes a human to do the analysis. Even humans of average intelligence with an engineering degree aren't very good at it. If you can build an AI that does it for any patent you feed in, all of our jobs are fucked.
- morgante 11y ago
  > The true value of a lawyer is their judgement, analysis ability, and their power of persuasion. I agree. We're not going to be seeing AI replacing lawyers any time soon. That being said, I think you are downplaying the efficiency gains which come from AI and lawyers working together, especially when it comes to bespoke work. For your example, diffing two patents is indeed something which still requires human intelligence. But there are ways of making the process of comparison more efficient by automatically correlating similar claims, providing references inline, etc. This goes beyond the easy wins which come from "better spelling, grammar, and cite checking software" (which are themselves non-negligible).
ryporter 11y ago
You make a great point about the success of AI-lawyer pairing. We need to avoid the Lump of Labor Fallacy[1], in which automation only replaces lower-level work. Advanced tools will allow less experienced (but, let's not forget, intelligent and highly educated) lawyers to perform more advanced work at an earlier point in their careers. [1] https://en.wikipedia.org/wiki/Lump_of_labour_fallacy https://en.wikipedia.org/wiki/Lump_of_labour_fallacy
zekevermillion 11y ago
If you've read Kasparov on playing chess with machines, he has some really interesting observations on practical and philosophical aspects of man vs. machine. His most interesting observation is that it's neither the best human grandmasters, nor the best chess machines, but rather the best algorithm for humans working with machines that tends to win. It does make for nice headlines when a machine like Deep Blue or Watson trounces the best human in a particular artificially-constrained test. But in the real world of unconstrained problem sets and complexity, my money is on the proper pairing of human and machine rather than one obtaining any lasting advantage over the other.
lifeisstillgood 11y ago
Can you recommend a reading list on the NLP / parsing / etc approach needed in this area - I recently had an argument along the lines of "of course you can breakdown a document in five lines of python" and was surprised to find it's harder than I thought (!)
dragonwriter 11y ago
> Another factor in favor of legal AI is that legal documents have much more restricted grammar than standard English text, leading to less ambiguity. That's...really not true. Legal documents pretty much admit the full scope of the English language -- both grammar and vocabulary -- though there are constructions and vocabulary that have special meaning in particular contexts within legal documents that would either not be found in English outside of the legal context or would have different meaning. (And ISTR seeing cases where construction within the particular documents led to significant questions of whether elements that exist both in general English and in the specialized language of law but have different meanings had their typical legal meaning or their more general meaning in the particular document.)
conceit 11y ago
>legal documents have much more restricted grammar than standard English text What are the degrees of freedom that are enabled by grammar and differ?

jbhatab 11y ago

Why isn't what you described possible with AI?

morgante 11y ago
The high-level work? Let me first say that I don't think we will have human-level AGI by 2030. If we somehow do, all bets are off. The high-level work requires a large amount of creativity in developing nuanced strategies for a particular domain. It also involves integrating knowledge and ideas from a lot more than the law. Consider that the vast majority of intelligent humans who attempt to make it to partner don't make it. Top lawyers are highly skilled, and automating them away is not trivial. Also consider the trust factor. We've done some customer discovery in this space and most people would just not be comfortable working directly with an AI. This is especially true because a good portion of what partners are doing is human contact work with clients, courts, and opposing counsel. There's a lot of leverage which comes from being able to read and manipulate people which AI can't replicate easily. Note that I'm referring to the 2030 date specifically for this. I don't think lawyers (or, really, any current profession) will exist in 2100.
- TheOtherHobbes 11y ago
  I don't see why "Uber for lawyers" wouldn't work. I'm imagining a combined semi-automated service where the paperwork side of law - simple wills, notarisation, simple divorces, and so on - is completely automated, and the creative elements are farmed out to individual attorneys with a public track record of success who compete on price. It's going to kill the traditional law firm business model stone dead. Lawyers are basically paid to logic-chop precedent and written law, and to be persuasive and/or intimidating in more or less creative ways. IMO eventually all of that could be automated. I don't see any reason why AI couldn't potentially have super-human powers of persuasion and manipulation. It won't happen by 2030, but the next step are document searching is to bundle up the basics of law as a low-cost automated service, and then start automating the process of argument, cross-examination, and persuasion - which will lead to an arms race, which will lead to very rapid advances, which will lead to complete automation of civil and criminal law by (I'd guess) the end of the century.
  - morgante 11y ago
    > I don't see why "Uber for lawyers" wouldn't work. Primarily because people generally want to choose and talk to their attorney. It's basically the opposite of taxis, where people didn't care who the driver was and (generally) preferred minimal interaction. For Uber, not having to interact with your driver or choose them is a benefit. In law, it would be a detriment. That's why I think the future isn't a monolithic application where you interact with a faceless half-human, half-AI lawyer. It's small individual attorneys accomplishing a lot more without needing expensive support staff thanks to AI. > complete automation of civil and criminal law by (I'd guess) the end of the century. Okay, sure. In the longer term, every profession will be automated though.
    - fauigerzigerk 11y ago
      >Primarily because people generally want to choose and talk to their attorney Yes but why do they want that? The reason is not just some emotional desire for hand holding. The main reason is that AIs are simply not (yet) very good at language related tasks, and there is not nearly as much progress as the current bout of hyperactive AI reporting suggests.

tensor 11y ago

Systems that use machine learning to do document review actually already exist. E.g. https://kirasystems.com https://kirasystems.com. They currently still need a lawyer in the loop to review and adjust the results, but eventually will be accurate enough to produce summaries that are more accurate than a human without any intervention. Traditional law firms are starting to invest in software like this. While it can be used to reduce hours, it can also be used to sell more by increasing the value of the hours spent. For example, instead of doing a review of a subset of a clients documents, they can offer to review all the documents for the same, or even slightly higher price.

morgante 11y ago
Yes, I'm familiar with Kira (they're in a similar space to us). The technology isn't really doing summarization yet though, it's more like extraction/TOC building. I'm definitely biased, but I do think that's the direction that legal technology is going. Right now it's about making it easier and faster for lawyers to review documents: eventually it will be about them not even having to review documents at all.
- semi-extrinsic 11y ago
  I'm curious: this discussion is all about improving software that understands legal language. Are there any (long-term) efforts to improve legal language to make it easier to understand for software?
  - rwmj 11y ago
    It's my understanding that this project is doing that: http://codex.stanford.edu/ http://codex.stanford.edu/, trying to turn law into something like computer code. See also: http://logic.stanford.edu/complaw/complaw.html http://logic.stanford.edu/complaw/complaw.html
    - conceit 11y ago
      There seems to be a bunch of projects. Lacking an AI to do the job let me extract what caught my eye and seems to be answering the question. >The Hammurabi Project is an experiment in converting legal source material to the C# programming language. [...] The idea is to grow a body of source code that mirrors the structure of U.S. federal and state legislation, regulation, and case law. https://mpoulshock.github.io/hammurabi/ https://mpoulshock.github.io/hammurabi/ >The Stanford Computable Contracts Initiative (SCCI) is working on legal technology that will help move the world from natural language based contracts toward a world of computable contacts. > The project [Designing and Understanding Forensic Bayesian Networks with Arguments and Scenarios] ’s new approach is to link the successful statistical modelling technique of Bayesian networks to models that effectively dovetail legal argumentation and scenario construction in the legal world. http://www.ai.rug.nl/~verheij/nwofs/ http://www.ai.rug.nl/~verheij/nwofs/ >Instead of automating legal work, I intend to focus on augmenting legal work. Specifically, my project [Computational Linguistics and Effective Legal Drafting] focuses on using advances in computational linguistics technology to help lawyers draft more precise and error-free legal documents like contracts, regulations and statutes.

HillRat 11y ago

Fascinating! I do wonder what happens to the traditional white-shoe partner track, where associates make their bones do sixteen hour days of grunt work (albeit with fantastic compensation) as a method of gaining experience and reputation. The skills of a high-level attorney, regardless of the practice, are honed through decades of experience. How would systemic AI affect the development of the profession?

knughit 11y ago
The skills of a high level lawyer is in sales and negotiations and trials, not in the document analysis that junior lawyers do.
rms_returns 11y ago
Don't worry, the AI isn't going the replace the "Perry Mason" kind of lawyers, they are irreplaceable! AI is designed to replace the "Junior Grunt Joe" kind of lawyer/associate who is mostly involved in documentation and assistance.
- rodgerd 11y ago
  And where do the next generation of Perry Mason lawyers come from, if there are no rungs on the bottom of the ladder?
  - js8 11y ago
    Maybe they could actually help the poor?
  - chongli 11y ago
    Hence the article's thesis: Artificial intelligence will cause “structural collapse” of law firms by 2030.
  - rms_returns 11y ago
    Well, that logic applies to all fields, not just legal. Lower rungs are disappearing from everywhere. In accountancy, junior accountants used to manually do "debit-credit" first, now Excel does it for them, so what will they learn? In IT, junior programmers used to toil hard with algorithms and data-structures to come up with barely working programs. Now there are IDEs, libraries, frameworks, CMSs all built and ready for them, so what will they learn? In Physics and Math, Scientists worked hard on just the theoretical framework first and scratch their heads in classrooms. Now there are computer software that do it for them. Perhaps this explains we didn't get another Newton/Einstein since last few decades?

dinkumthinkum 11y ago

Well, you certainly have a "dog in this fight." =) It is a little self-serving to say "Yep, this is possible and we are the ones doing it." But since you have such a startup, you'd be crazy not say that! But as I have stated elsewhere, document summarization and classification is not even in the ballpark of an AI lawyer. Excel didn't replace the Accountant. It's like a fancy Excel for clerks and paralegals (and Excel was a wildly successful product). You sort of admit that but I think it is more than just the interactions with clients or making them "feel good." I don't think it really addresses what lawyering is or how positive legal outcomes can be achieved. This ma be a weak comparison but it's like we all know that the point of programming is not writing code and cleaning up memory; it's to create solutions to problems. Also, I think the traditional structure of the firm serves many more purposes than just specifying a hierarchy for task assignment.

morgante 11y ago
> It is a little self-serving to say "Yep, this is possible and we are the ones doing it." Fair enough. :) Though I tried to be a little more nuanced than that. To be clear, I don't think AI will replace lawyers any time soon. It's just making them more efficient. Our customers are layers, not people looking for legal services. > Excel didn't replace the Accountant. Great analogy. We don't think AI is going to replace lawyers any time soon, but it will make them more efficient (and potentially eliminate some of the lower-tier staff, like what technology has done to pure "bookkeepers.") > I think it is more than just the interactions with clients or making them "feel good." Sorry if I didn't make it clear, but I do think lawyers still provide a lot of value which AI can't easily replace. There's a reason we didn't have an AI write our equity plan. > I think the traditional structure of the firm serves many more purposes than just specifying a hierarchy for task assignment. That might be true, but remember that associates are expensive. If the budget for them has pressure applied on it and with AI you can get away with 1 associate instead of 5, that's going to change the legal track substantially.

rhino369 11y ago

The true value of a lawyer is their judgement, analysis ability, and their power of persuasion. To replace that you need a true Strong AI. So I expect lawyers to be replace shortly after everyone else is replaced. Give or take a few years, since lawyers have some protectionist options, like banning AI from being a lawyer. Weak AI will eat around the edges. Better spelling, grammar, and cite checking software could reduce a lot of billable time. But most associates spend less than 10% of their time doing that. For big litigation, document review has been mostly outsourced to legal service companies. I've been a junior associate (the ones traditionally responsible for reviewing documents) for about 2 year. I think I've billed ~200 hours to that (out ~4,200 hours I've billed total). Client won't pay our rates. So we hire contractors for 25 dollars an hour. There is still room for disruption. Predictive coding will probably become the norm within 15 years. But it's only going to put the temp "doc reviewers" out of job. Westlaw and Lexis already summarize cases, index them. ALJ and Mathew Bender sells practice guides. Firms subscribe to proofreading, document processing, and printing services. They hire paralegals and assistants to do the less taxing work. It's also important to note that business model of firms only relies on associates billing hours because it's convenient. Clients don't pay 500 an hour for an associate to make sure commas are correct because it costs that much to provide. It's just an add on service to get the high level talent's analysis and judgement. Nickle and diming essentially. Associate costs are fairly small portion of current billing rates. Most of the grunt work I do is bespoke. Stuff like comparing one patent to another. We have google patent and questal to search. But it takes a human to do the analysis. Even humans of average intelligence with an engineering degree aren't very good at it. If you can build an AI that does it for any patent you feed in, all of our jobs are fucked.

morgante 11y ago
> The true value of a lawyer is their judgement, analysis ability, and their power of persuasion. I agree. We're not going to be seeing AI replacing lawyers any time soon. That being said, I think you are downplaying the efficiency gains which come from AI and lawyers working together, especially when it comes to bespoke work. For your example, diffing two patents is indeed something which still requires human intelligence. But there are ways of making the process of comparison more efficient by automatically correlating similar claims, providing references inline, etc. This goes beyond the easy wins which come from "better spelling, grammar, and cite checking software" (which are themselves non-negligible).

ryporter 11y ago

You make a great point about the success of AI-lawyer pairing. We need to avoid the Lump of Labor Fallacy[1], in which automation only replaces lower-level work. Advanced tools will allow less experienced (but, let's not forget, intelligent and highly educated) lawyers to perform more advanced work at an earlier point in their careers. [1] https://en.wikipedia.org/wiki/Lump_of_labour_fallacy https://en.wikipedia.org/wiki/Lump_of_labour_fallacy

zekevermillion 11y ago

If you've read Kasparov on playing chess with machines, he has some really interesting observations on practical and philosophical aspects of man vs. machine. His most interesting observation is that it's neither the best human grandmasters, nor the best chess machines, but rather the best algorithm for humans working with machines that tends to win. It does make for nice headlines when a machine like Deep Blue or Watson trounces the best human in a particular artificially-constrained test. But in the real world of unconstrained problem sets and complexity, my money is on the proper pairing of human and machine rather than one obtaining any lasting advantage over the other.

lifeisstillgood 11y ago

Can you recommend a reading list on the NLP / parsing / etc approach needed in this area - I recently had an argument along the lines of "of course you can breakdown a document in five lines of python" and was surprised to find it's harder than I thought (!)

dragonwriter 11y ago

> Another factor in favor of legal AI is that legal documents have much more restricted grammar than standard English text, leading to less ambiguity. That's...really not true. Legal documents pretty much admit the full scope of the English language -- both grammar and vocabulary -- though there are constructions and vocabulary that have special meaning in particular contexts within legal documents that would either not be found in English outside of the legal context or would have different meaning. (And ISTR seeing cases where construction within the particular documents led to significant questions of whether elements that exist both in general English and in the specialized language of law but have different meanings had their typical legal meaning or their more general meaning in the particular document.)

conceit 11y ago

>legal documents have much more restricted grammar than standard English text What are the degrees of freedom that are enabled by grammar and differ?