Some Random Thoughts on AI and Legal Practice

About a decade ago, I remember reading about the singularity, AI and automation and whether it would leave everybody jobless and we’d need UBI. Matt Yglesias wrote:

That said, this is why I’ve been saying that yoga instructors have the job of the future. . . .The people of the future will be richer than the people of today, and therefore will more closely resemble annoying yuppies. Nicer restaurants are more labor-intensive than cheap ones, and the further up the scale you go the more specialized skills (think sommelier) come into play. Annoying yuppies take yoga classes, or even hire personal trainers. Artisanal cheese is more labor-intensive to produce than industrial cheese. More people will hire interior designers and people will get their kitchens redone more often. There will be more personal shoppers and more policemen. People will get fancier haircuts.

The idea is that we won’t be jobless, but that with automation doing the tasks we used to do, leaving humanity much wealthier, more and more jobs would be about interacting with other humans providing a service at more a bespoke scale. Baumol’s cost disease on steroids.

When I first asked ChatGPT to write a legal brief and it churned out something reasonable, I knew that the legal profession was forever changed, it is just a matter of when. I assume this is coming for all sorts of business services but I think something about the practice of law gives some insight here, both positive and negative. Consider these some scattered thoughts.

In the future, all white collar professionals will be yoga instructors. I think the prediction about yoga instructors gets some things backwards. I am a civil rights lawyer. While a lot of my work involves writing briefs and there are really interesting legal questions, a lot of it is applying pretty similar facts to pretty similar legal questions, with pretty set answers. A better trained LLM will give pretty great writing soon. Lawyers will need to review it, particularly for ground truth, but we already do that. It’s Shepardizing.

But a lot of the work of being a civil rights lawyer is the personal and emotional work. My clients are people who have been illegally sent to jail, they have been abused by the police, they are in jail and have been slashed. The work is often managing their emotions about the case, guiding them through the case, preparing them to relive their experience to give testimony. I also need to find witnesses and track them down and convince them to give testimony when they’re reluctant and do not want to be involved. The work is emotional and it requires trust.

For the Big Law firms I used to work at, the skills of the incredibly talented litigators who can write great briefs and analyze complicated concepts and business dealings, are about to be tremendously devalued. The skills of witness preparation, client management, settlement negotiation and strategy, will go way up in value. Spreadsheets and accounting software have made the math skills of accountants decline in value, what you need to be is good at organizing information and at being trustworthy to clients and guiding them through stressful processes like audits.

The more analytical work that is done by computers, the more social skills will be valued. In the future, professional services will be increasingly about coaching clients through what the computers are spitting out and making the concepts legible to clients, judges, and juries.

The First Uses for AI Will Be Where Ground Truth is Given. I was listening to Chris Hayes’s excellent podcast on AI with Kate Crawford and they discussed the issue of “ungrounded” vs. “grounded” LLMs and the problem of “hallucination” from LLMs, i.e. making up citations to cases that don’t exist. I’ve seen LLMs do that in the legal context. Ungrounded LLMs aren’t trained on search results, there’s no ground truth function. In the legal context, that means the best place to use LLMs for legal work is where the lawyer is providing the ground truth. I imagine that civil rights lawyers doing intake will be able to take detailed notes on the incident and asked an LLM to draft a complaint based upon those details. In that instance, the world of fact is limited by what is provided by the attorney.

To test this out, I asked GPT-4 to draft a complaint based upon a news article about a civil rights violation at a New York jail. The complaint that came back contained factual information from the article and a decent first draft of a complaint in ten minutes.

Indigent clients just got much better representation and public defenders offices will benefit greatly. If you work in solo practice enough on the plaintiffs’ side, as I do, you will frequently hear the following lament “the liability is good, but the damages just aren’t high enough for anybody to take the case.” Plaintiffs’ lawyers, who work on contingency, have to think about how many hours a case will take to litigate, how much the expected damages are, and accordingly how much money they will make if successful. The Federal Bar Council, for example, has an access to counsel program for the many individuals with reasonable cases that deserve justice, who cannot get a lawyer to take their case. If a case starts to take half the amount of time to work on, cases that were once uneconomical will become economical. This is to say that while it is certainly the case that there will be reduction in the need for lawyers in certain respects, there will be a big increase in the demand for cheaper legal services, and that will be a boon to people who deserve a day in court.

Anybody who knows a public defender knows that they are overworked and underpaid. But what if they get access to high quality briefing on an issue that takes a motion from 10 hours to 1 hour? What if an LLM can review trial transcripts before closing and spot issues and appellate issues to raise with the court in order to preserve them that an overworked defender might miss? I think defenders augmented by an LLM may be able to provide significantly better representation than indigent defendants once had access to. There are many concerns I have about LLMs and the displacement that we will see soon, but the access to representation effects are real. It is bad that legal services are out of reach and too expensive for many people and the threat to lawyers is a societal good, including me.