Agree with the elasticity comment, completely. That said, I do think people misapprehend what lawyers spend their days doing, at least what I spend my day doing as a lawyer. It’s not simply producing contracts that could be ‘mad lib’d by an AI. I’m not trying to argue for my value – I think lawyers are generally a value suck. It’s just that most of the questions and much of the job is about exercising judgment related to novel circumstances, and probably not something that an AI would be particularly good at.
That said, I do think something like a trust and estates attorney could do really well with AI. If you took all of the answers from a client questionnaire, and correlated those with the documents that were output, you could probably train an AI model to do a very competent first draft. But, again, the outcome of such a system might very well be to lower the price of providing these services, which would make those services available to more people, which would increase the size of the market, which would result in more rather than fewer trust and estate attorneys. These things are hard to predict, i think, and the impact of technology on markets is often surprising.
One of the most surprising anecdotes I’ve heard is that the cotton gin lowered the price of cotton, which lowered the price of cotton clothing, which increased the demand for clothes because people wanted more than one set of wool underwear, etc., which (paradoxically from my viewpoint) drove an increase in the demand for slaves to grow the cotton. The claim I heard was that slavery was generally petering out before the cotton gin. I don’t know if it’s true completely, but it’s fascinating that the cotton gin could have been the “cause” – at least “a” cause – of the Civil War. So, again, hard to predict how these things will play out.
"which (paradoxically from my viewpoint) drove an increase in the demand for slaves to grow the cotton." It is a paradox. Jevon's paradox. Belsky mentions this as well, "In some industries where we start paying less for something, perhaps we end up making up for it in volume?" I don't know how this will work out with the various areas of the law, but every time I hear someone say teachers are going to be replaced by AI I think that Jevon's paradox might apply because teachers are part mentor, daycare worker, etc. and we will change some of the things a teacher does, but even with increased productivity from learning AIs in classrooms or whatever we may end up with more teachers, not fewer.
The more I thought about this the more wrong I convinced myself I was because Jevons was talking about resources like water, or sand, or some such. So I asked chatgpt and it said that it was labor displacement, labor displacement of the task of picking seeds out of cotton, and this caused increased demand for the growing and picking of cotton. Under slavery there wasn't any difference in who performed the three different tasks.
Legal services are elastic! I have a friend who spent a year in law school before becoming an economist. He told me that the one thing he learned there was “there are many towns too small for a lawyer, but no town is too small for two lawyers”.
Given the massive "access to justice" (terrible phrase) crisis in Canada and the United States, I wonder if dramatic improvements in lawyer productivity will actually *increase* demand for legal services. Right now, if you have a small matter worth below say $10k, it may simply be uneconomical for you to get legal assistance because a lawyer may spend 25 hours on your file for say $300/hour, which, after other costs are taken into consideration, wipes out any residual value. But if LLMs dramatically improve lawyer productivity, the same lawyer may be able to complete the file in 5 hours for the same per hour cost, so it now may be economical for the person to go the lawyer for legal assistance.
If hourly billing is antiquated, any lawyer can go ahead and start practicing in a practice area that uses it heavily and profit by not using hourly billing. Like go ahead, a giant pot of gold is waiting for you and no one is stopping you if it is such a good idea. We use a mix of fixed price and hourly billing. There is a competitor small firm in our specialized practice area that uses SaaS-style pricing (he started his solo practice immediately after graduating law school), but it does not work well and results in many dropped and underserved clients.
GPT is not as disruptive to many legal practice areas that many people assume because they don't really understand what absorbs a lot of the time involved in a lot of practice areas. A lot of the things that people assume would be killer applications of ChatGPT are already well served by practical law manuals, templates, brief banks, and other things like that. In theory, ChatGPT should be able to help lawyers who are unfamiliar with a specific practice area to get up to speed more quickly, but unfortunately its particular weaknesses make it the most hazardous to use in unfamiliar practice areas. You are more likely to be able to use it "safely" in an area that you already know very well or as a preliminary research tool.
The other major issue with trying to get away from hourly billing is that it is most useful for handling issues for which it is impossible to estimate ahead of time. You do not know what opposing counsel is going to do during discovery ahead of time, for example, and while it more possible to provide an estimate when you have been opposite a particular lawyer/client combination before in different contexts and on the same type of issue, people are always throwing curveballs or doing stupid things, or illegal things, or unethical things. Your client may also be full of surprises that cannot be estimated ahead of time (e.g. "oh yeah, I forgot to tell you, our supplier here just put a different date down on that document" etc.) . And a lot of tools for making discovery in particular more efficient are more specialized and useful than ChatGPT could be at the same task.
The other big issue with using ChatGPT outside of some rumored enterprise edition is that if you input your client's facts, you are violating your client's privilege, because both ChatGPT systems and their employees can review these inputs at any time and in detail deanonymized to you (read the ToS). Even if you anonymized your client's facts, this is really no bueno, because it would not be that hard to figure out whose facts belong to whom. This would be different on an entirely self hosted system that was just using someone else's model that did not phone home. But that is not what people usually use .
One thing I will say about current LLMs is that they seem less good at identifying and framing legal issues. Once you can figure out how to prompt the LLM, it can add a lot of value. But, even now, a lot of the value of legal work (at least the kind of legal work I do) comes from actually being able to spot issues, come up with the right vantage point and level of generality to look at the issue, etc. Put another way, law school exam skills may become *more* valuable thanks to chatGPT, since a lot of what exams are about is taking a set of unstructured, disparate facts and find a way to "translate" them into a body of structured legal doctrines.
Having said all this, I will now undercut my point with three observations. First, I have sometimes been very surprised by an LLMs ability to spot an issue or else trigger something in my brain that helps me spot a real stealthy issue. Second, while I say that LLMs won't eliminate the value of issue spotting, creative solution-making, etc., I think lots of people underrate how rare this skill is even among lawyers. Pre-chatGPT, the average lawyer was probably somebody who, if they could spot the low hanging fruit and apply the approximately correct legal principles, would get by just fine. For these people, LLMs are either going to force them to develop skills they don't commonly flex or else will push them out of the market. Third and finally, all of this is based on existing LLMs. Just based on the massive improvement I saw from GPT-3.5/Claude to GPT-4/Claude 2, my guess is that LLMs that come out say in 2 - 5 years from now may even be able to do the things I'm saying lawyers still have some advantage in.
Agree with the elasticity comment, completely. That said, I do think people misapprehend what lawyers spend their days doing, at least what I spend my day doing as a lawyer. It’s not simply producing contracts that could be ‘mad lib’d by an AI. I’m not trying to argue for my value – I think lawyers are generally a value suck. It’s just that most of the questions and much of the job is about exercising judgment related to novel circumstances, and probably not something that an AI would be particularly good at.
That said, I do think something like a trust and estates attorney could do really well with AI. If you took all of the answers from a client questionnaire, and correlated those with the documents that were output, you could probably train an AI model to do a very competent first draft. But, again, the outcome of such a system might very well be to lower the price of providing these services, which would make those services available to more people, which would increase the size of the market, which would result in more rather than fewer trust and estate attorneys. These things are hard to predict, i think, and the impact of technology on markets is often surprising.
One of the most surprising anecdotes I’ve heard is that the cotton gin lowered the price of cotton, which lowered the price of cotton clothing, which increased the demand for clothes because people wanted more than one set of wool underwear, etc., which (paradoxically from my viewpoint) drove an increase in the demand for slaves to grow the cotton. The claim I heard was that slavery was generally petering out before the cotton gin. I don’t know if it’s true completely, but it’s fascinating that the cotton gin could have been the “cause” – at least “a” cause – of the Civil War. So, again, hard to predict how these things will play out.
"which (paradoxically from my viewpoint) drove an increase in the demand for slaves to grow the cotton." It is a paradox. Jevon's paradox. Belsky mentions this as well, "In some industries where we start paying less for something, perhaps we end up making up for it in volume?" I don't know how this will work out with the various areas of the law, but every time I hear someone say teachers are going to be replaced by AI I think that Jevon's paradox might apply because teachers are part mentor, daycare worker, etc. and we will change some of the things a teacher does, but even with increased productivity from learning AIs in classrooms or whatever we may end up with more teachers, not fewer.
The more I thought about this the more wrong I convinced myself I was because Jevons was talking about resources like water, or sand, or some such. So I asked chatgpt and it said that it was labor displacement, labor displacement of the task of picking seeds out of cotton, and this caused increased demand for the growing and picking of cotton. Under slavery there wasn't any difference in who performed the three different tasks.
Legal services are elastic! I have a friend who spent a year in law school before becoming an economist. He told me that the one thing he learned there was “there are many towns too small for a lawyer, but no town is too small for two lawyers”.
Given the massive "access to justice" (terrible phrase) crisis in Canada and the United States, I wonder if dramatic improvements in lawyer productivity will actually *increase* demand for legal services. Right now, if you have a small matter worth below say $10k, it may simply be uneconomical for you to get legal assistance because a lawyer may spend 25 hours on your file for say $300/hour, which, after other costs are taken into consideration, wipes out any residual value. But if LLMs dramatically improve lawyer productivity, the same lawyer may be able to complete the file in 5 hours for the same per hour cost, so it now may be economical for the person to go the lawyer for legal assistance.
LOL- legislatures are full of lawyers. They will make sure lawyers stay employed with healthy billing hours.
If hourly billing is antiquated, any lawyer can go ahead and start practicing in a practice area that uses it heavily and profit by not using hourly billing. Like go ahead, a giant pot of gold is waiting for you and no one is stopping you if it is such a good idea. We use a mix of fixed price and hourly billing. There is a competitor small firm in our specialized practice area that uses SaaS-style pricing (he started his solo practice immediately after graduating law school), but it does not work well and results in many dropped and underserved clients.
GPT is not as disruptive to many legal practice areas that many people assume because they don't really understand what absorbs a lot of the time involved in a lot of practice areas. A lot of the things that people assume would be killer applications of ChatGPT are already well served by practical law manuals, templates, brief banks, and other things like that. In theory, ChatGPT should be able to help lawyers who are unfamiliar with a specific practice area to get up to speed more quickly, but unfortunately its particular weaknesses make it the most hazardous to use in unfamiliar practice areas. You are more likely to be able to use it "safely" in an area that you already know very well or as a preliminary research tool.
The other major issue with trying to get away from hourly billing is that it is most useful for handling issues for which it is impossible to estimate ahead of time. You do not know what opposing counsel is going to do during discovery ahead of time, for example, and while it more possible to provide an estimate when you have been opposite a particular lawyer/client combination before in different contexts and on the same type of issue, people are always throwing curveballs or doing stupid things, or illegal things, or unethical things. Your client may also be full of surprises that cannot be estimated ahead of time (e.g. "oh yeah, I forgot to tell you, our supplier here just put a different date down on that document" etc.) . And a lot of tools for making discovery in particular more efficient are more specialized and useful than ChatGPT could be at the same task.
The other big issue with using ChatGPT outside of some rumored enterprise edition is that if you input your client's facts, you are violating your client's privilege, because both ChatGPT systems and their employees can review these inputs at any time and in detail deanonymized to you (read the ToS). Even if you anonymized your client's facts, this is really no bueno, because it would not be that hard to figure out whose facts belong to whom. This would be different on an entirely self hosted system that was just using someone else's model that did not phone home. But that is not what people usually use .
One thing I will say about current LLMs is that they seem less good at identifying and framing legal issues. Once you can figure out how to prompt the LLM, it can add a lot of value. But, even now, a lot of the value of legal work (at least the kind of legal work I do) comes from actually being able to spot issues, come up with the right vantage point and level of generality to look at the issue, etc. Put another way, law school exam skills may become *more* valuable thanks to chatGPT, since a lot of what exams are about is taking a set of unstructured, disparate facts and find a way to "translate" them into a body of structured legal doctrines.
Having said all this, I will now undercut my point with three observations. First, I have sometimes been very surprised by an LLMs ability to spot an issue or else trigger something in my brain that helps me spot a real stealthy issue. Second, while I say that LLMs won't eliminate the value of issue spotting, creative solution-making, etc., I think lots of people underrate how rare this skill is even among lawyers. Pre-chatGPT, the average lawyer was probably somebody who, if they could spot the low hanging fruit and apply the approximately correct legal principles, would get by just fine. For these people, LLMs are either going to force them to develop skills they don't commonly flex or else will push them out of the market. Third and finally, all of this is based on existing LLMs. Just based on the massive improvement I saw from GPT-3.5/Claude to GPT-4/Claude 2, my guess is that LLMs that come out say in 2 - 5 years from now may even be able to do the things I'm saying lawyers still have some advantage in.