The Fifth Circuit issued a notable order to parties in an appeal today. They instructed the attorneys (including Mississippi’s fantastic deputy solicitor general, Anthony Shults) to be prepared to address whether or not the case should be assigned to a different district judge due to the current judge’s “use of AI.”
You may recall this instance of Judge Wingate offloading his work to robots from when Judiciary Committee chairman Chuck Grassley sent him — and Judge Julian Neals in New Jersey — oversight letters asking about their apparent use of AI in their published opinions, hallucinations and all.
It also coincides with an interesting report by Josh Morrow, a partner at Lehotsky Cohn, about the prevalence of AI in appellate opinions. He ran 2,250 opinions through an AI detector software and found dozens of opinions with “signs of AI writing.” Gentleman that he undoubtedly is, Mr. Morrow doesn’t name names.
He probably should. Or, if he doesn’t, the Judiciary Committee should. Chairman Grassley should consider asking Mr. Morrow for his dataset so his crackerjack staff can replicate it and, if they agree with its findings, either publicize it or, better yet, send it to the Judicial Conference of the United States and urge them to take action lest he publicize it.
The judiciary needs to take the threat posed by legal AI seriously. No, it’s not going to usher in Judgment Day (probably). Yes, it absolutely has economically and socially beneficial uses — such as in the employment space. But when it comes to judging, which is — as Justice Barrett explained at length in her book — a fundamentally humanistic enterprise, it poses real dangers.
First, there’s the problem of AI judging itself. Let’s call it Data Center Judging — the act of outsourcing the resolution of a case or controversy from men and women in robes to algorithmic tools powered by the massive humming boxes popping up across the fruited plane. The law is a human enterprise, written by people to regulate interaction between people as applied by people. We go to great lengths as a society to pick the people who will do that interpreting, whether via contentious elections, expert appointment, or the advice-and-consent process.
Are they perfect? Far from it; like all humans, judges have their failings. They’re also not interchangeable. They are selected with an eye toward their subjective views of the law and justice, and we expect them to bring those views to bear in their resolution of cases and controversies. While they make mistakes, simple professional pride would almost always prevent them from simply, say, making up a citation to justify their opinion or order.
Beyond that, we know what we’re getting with human judges doing actual judge work, biases and all. Outsourcing legal interpretation to Silicon Valley algorithms hides those biases in the black box of machine learning.
Unlike in, say, employment screening where the machine learning is simply a more efficient way to analyze objective inputs to produce desired outputs, the inputs in legal interpretation are necessarily subjective. AI can no more interpret legal texts than it can interpret Shakespeare. When you superimpose the reality of actual justice on legal texts in the instance of a case or controversy, the proposition of AI judging becomes even more obscene. Again, unlike in the case of many machine-learning applications, where the imposition of “bias testing, transparency mechanisms, or corrective constraints” entails the capture of a quantifiable process by alien, subjective goals, here they would be necessarily baked into the subjective process itself.
From where do we want these subjective legal judgments emanating? Judges selected through a constitutional process who put their names to what they say or write? Or data-center powered algorithms applying recursive “bias correction” traceable back to faceless data scientists and their tech founder overlords?
Second, there’s the looming problem of slop on the legal system. Already our courts are beset by an endless stream of frivolous litigation, often brought by pro se parties. At least sovereign citizens, strike-suiters, and cranks, until recently, needed to type out their own vexatious pleadings. Once they start using AI to do so, expect the volume of pro se and other filings to increase geometrically.
Once that happens, there will be a vicious alliance within the judiciary supporting the expansion of slop pleading. On one hand, liberal do-gooders will insist that this is the best way for the disenfranchised to really have their day in court. No, slop pleadings aren’t good, they’ll argue, but they’re probably better than what the pro se and other indigent litigants would produce otherwise, and, being cheaper than hiring a real lawyer, they’ll increase access to justice. On the other hand, sober-minded judges, clerks, and staff attorneys will agree that at least the slop is intelligible, so it’s better than the alternative.
The only solution for wading through the slop will be Data Center Judging. Slop in; slop out. This won’t benefit pro se or indigent litigants in the long run, and it certainly won’t benefit regular Americans looking for their day in court. It’s a foreseeable problem that should be nipped in the bud, and the way to nip it in the bud is by courts putting clear, defined, and strict restrictions on the use of AI lawyering.
Data center judging is a world of (cheaper) slop briefs ruled on by (faster) slop opinions, based on parameters set by (private-sector) black boxes. You can see why big business may very well like Data Center Judging and why it will likely balk at any attempts to curb it.
But the courts — especially the federal courts — have the luxury of life tenure and Congressional appropriations. They’re in a unique position to draw a line in the sand and maintain both the humanistic and democratic integrity of our legal system.