AI will make lawyers cheaper, faster, and better
(If we're allowed to use it.)
Already several people have tested ChatGPT’s legal abilities. Reviews are mixed.
SCOTUSBlog gives ChatGPT a score of 21/50 on a Supreme Court quiz. It nails some hard ones but thinks RBG dissented in Obergefell.
ChatGPT takes some University of Minnesota Law School exams and averages a grade of C+.
Law360 reports that ChatGPT makes some mistakes when you ask it to draft a complaint and opines it would be unethical to file a ChatGPT-drafted complaint without reading it over.
These results should engender wild confidence in AI. ChatGPT is at the level of a law school graduate despite having never attended law school! It has not been trained on the vast compendium of legal materials in Lexis and Westlaw and has never been given human guidance on any legal issues. Eventually it will. I am skeptical about how much lawyers actually learn in law school, but surely reading and memorizing all legal texts ever written will improve ChatGPT’s performance as a lawyer. Also, there are plenty of competitor AIs under development, some of which will target lawyers. It will not be that hard to merge Large Language Models with custom-made outlines to help the AI issue-spot. The technology is there, it is just a matter of refining it.
Meanwhile, the good folks at DoNotPay are trying to help people fight traffic tickets with AI, but are apparently being threatened by state bar associations who call this “unauthorized practice of law.” Apparently, people are better off going pro se or bankrupting themselves with attorney’s fees? The good news is that state bar associations do not get the final say. No one benefits from banning AI except for (some) lawyers, and no one likes lawyers. I am optimistic that, once the technology improves a bit, state legislatures will permit the use of AI throughout legal practice, including in the courtroom.
Also DoNotPay has also offered $1 million who will use AI to give a Supreme Court argument. While I doubt the Supreme Court would permit this in the near term, AI will be particularly useful at oral argument. Humans are bad at split-second extemporaneous responses and AI is designed to give split-second extemporaneous responses in every case. DoNotPay’s idea is a gimmick, but also a good idea when a court is brave enough to permit it.
Turning to brief-writing…
Why would ChatGPT be good at writing briefs? Well, let’s ask ChatGPT:
This is a decent answer! I’d add a few things:
Lawyers are extremely expensive, so cheap AI adds a lot of value.
AI is good at making logical deductions and teasing out logical errors.
AI is good at searching large evidentiary records and finding factual authority.
AI can search for obscure case law quickly.
AI is good at writing in standard formats, and most briefs tend to follow standard formats, particularly in lower courts.
AI will be fanatical about compliance with deadlines, local rules, and citation formatting requirements.
What about stuff AI is bad at? ChatGPT has an admirable willingness to introspect about its own faults:
These do not bother me so much. #1 and #2 simply require a bit more learning by the AI. #3 is not a big deal, a legal argument is much more likely to win when it is supported by existing legal authority than when it is “creative.” 100% of the time, when a court characterizes an argument as “creative,” the court explains in the next breath why the argument loses (or should lose).
#4 is not a problem if the AI is trained on a sufficient number of texts. #5 is a problem for humans too.
Many of AI’s other deficiencies are not a big deal in the context of brief-writing. AI is not very funny, but briefs should rarely be funny. AI is bland, but briefs require a certain measure of blandness, plus the marginal benefits of making a brief “sing” may not exceed the additional costs, at least for price-sensitive litigants.
The upshot is that pretty soon we’ll be able to submit requests to the AI like this and get good answers:
Under Illinois law, what does it take to pierce a trust? Find the best case for the proposition that it should be really, really hard.
Write a paragraph with three policy arguments in favor of an “ascertainability” requirement in class action cases.
Find the trial testimony most clearly establishing that the client suffered emotional injuries as a a result of her fall.
Tasks that current require hours of legal research and writing, with multiple levels of review, will be completed in a few seconds. A couple of years later, we’ll be able to submit more sophisticated requests like this:
I live in Florida. A New York-based venture capital fund called me, persuaded me to invest, and lost all my money. What are my possible legal claims against the venture capital fund? What facts would I need to prove for each claim?
Write a complaint against the venture capital fund with all those legal claims. Include placeholders for specific additional facts I’d have to allege for particular claims.
What are the three best arguments that the district court’s order contained reversible error?
Write a detailed outline for a brief with the following arguments: The district court misapplied the hearsay rule and the error wasn’t harmless; there wasn’t substantial evidence to support the damages award; and the injunction is too broad.
Take this edited outline and prepare a draft of the brief, including a facts section that heavily emphasizes trial testimony about our client’s emotional injuries.
It is easy to imagine tasks that took 200 hours for a human taking 20 seconds for an AI.
That said, I suspect that AI will complement rather than replace human lawyers. The human will have to know the right questions to ask, make appropriate edits, maybe write a punchy intro. The AI might be more like a junior associate with superlative researching skills. So maybe 200 hours will turn into 20 hours rather than 20 seconds, but the amount of money saved will still be staggering.
I’ve focused on brief-writing because that is what I do for a living, but AI will have benefits for all lawyers. For example, it is pretty obvious how AI will assist in deal work, just tell the AI what kind of deal you want in general terms and the AI can write the documents, subject to light editing.
So what’s going to happen to civil litigation?
First, most obviously, AI will improve access to justice. It will be possible to produce half-decent work product with a busy lawyer, a bad lawyer, or no lawyer. Criminal defendants, low-income civil defendants (like defendants in eviction and collections actions), and civil plaintiffs will benefit immediately. They’ll bring the right claims and never waive anything. (This is one reason I think state bar associations, which purport to champion access to justice, will either back down or be overruled by legislatures.).
Also, AI will make litigation go faster. If AI allows lawyers to move faster, judges can make deadlines tighter. As I explained last week, faster litigation is better in many ways.
The dynamic effects of AI are harder to predict. Will AI increase or decrease the amount of litigation? Well, if lawsuits are cheaper to bring, perhaps more people will bring them. However, AI might also ward off bad lawsuits by warning inexperienced litigants of their lack of merit. Also, AI will make lawsuits go more quickly, which will incentivize good lawsuits while deterring bad ones that rely on the slow pace of litigation rather than the merits of the claim.
With respect to the cost of legal services, there is something of a Red Queen effect here. If the plaintiff’s lawyer is suddenly better, then defense counsel might want to spend more money getting better. On the other hand, there are diminishing benefits to throwing money at lawyers, especially with cheap AI that can prepare serviceable first drafts. Also, in some cases a good lawyer can save both sides money, by foregoing meritless filings that waste the time of both sides. Faster litigation will also save both sides money.
Ultimately we should not overthink this. If lawyers are faster, better, and cheaper, Occam’s Razor says that litigation will be faster, better, and cheaper, one way or another.
More speculatively, how will AI affect law firms and the legal profession? Here are some possibilities…
AI will bring down costs a lot, even relative to other fields. AI will bring the cost of writing a brief way down (both because legal services are so expensive right now and because AI is well-suited toward brief writing), whereas it will bring down the cost of, e.g., conducting basic scientific research or putting on a new roof a lot less.
AI will disrupt the traditional structure of law firms. If AI will replace anyone, it is the junior associate that prepares the first draft of a filing for the partner’s review. It may be unrealistic in the short term to forego all human review of an AI’s draft, but it is more realistic for the AI to prepare the first draft. Clients might say: I’ll pay for the partner’s time, but not for the first draft. Sometimes the draft will be great, sometimes it will be way off, but that is also true of first drafts by humans.
For each of the prior two reasons, law will become a less lucrative field, which will affect everyone’s incentive to go to law school.
The demand for lawyer-hours will decrease, so marginal lawyers—the type who barely get jobs in the status quo—will forego law school and do something else.
Meanwhile, at the highest levels of profession, lawyers will make less money (in relative terms), so the type of high-powered college graduate who currently goes to an elite law school with the intent of obtaining a lucrative law firm job will also choose a different career path.
Perhaps this will lead to a more efficient social allocation of talent, as there is reason to doubt society benefits from having so many bright people suing so many other bright people.
One can only hope! Next week, I will turn away from AI and focus on some current litigation.
(This post reflects my own views, not those of Jenner & Block.)