Fifth Circuit Comments on District Judge's Discussion of Using AI in Judicial Decisionmaking

From La Union del Pueblo Entero v. Abbott, decided today by Judge Edith Jones, joined by Judge Kurt Engelhardt and District Judge Robert Summerhays (W.D. La.):

The district court gave an interview to the Wall Street Journal explaining how he had used artificial intelligence as an adjunct to his work on some aspects of a case "involving Texas[ ] election law." Whether it was this case is uncertain. However, as one distinguished U.S. Senator [Grassley] has commented, AI "must not be a substitute for legal judgment," nor must the public perceive that federal judges outsource our judgment to AI tools.

Thanks to Michael Smith (Smith Appellate Law Firm) for the pointer.

Comments

[TwelveInchPianist]

"...nor must the public perceive that federal judges outsource our judgment to AI tools."

Of course they don't. It's their clerks that do it.

[Bubba Jones]

Why do federal judges give interviews? What is the upside?

[not guilty]

Provided a judge is circumspect in what he says and what he declines to talk about (such as pending litigation), it can promote greater public understanding of the judicial system and of the legal process.

[Lee Moore]

FWIW, which is not much, I am only slowly getting into using AI and I get variable results - surprisingly so.

In the last few days I have been asking a couple of the AI engines :

(a) about DNA paternity (and remoter) testing including going down rabbit holes on autosomal testing, X chromosome testing, Y chromosome testing and mitochondrial DNA testing, including even tinier holes on sex differences in DNA inheritance from grandparents, and the mysteries of X inactivation

and

(b) can I have a list of German generals killed in 1942 on the Eastern Front please ?

You know - the kind of stuff everybody wants to know.

The results on (a) were spectacularly good and detailed, and "intelligent" and they checked out with non AI sources.

The results on (b) were spectacularly bad, with a majority of AI listed German generals killed in 1942, in fact having survived the war.

I haven't asked those same AI engines why their results are so spectacularly good on one subject and so bad on another subject, and one which seems a lot easier to get right.

But if I had to guess I'd be guessing that AI bots are trained mostly on English language sources.

[Stupid Government Tricks]

That sounds like a suspiciously bad conclusion.

For (b), it's amazingly easy to verify the results -- look up the names.

(a) is nowhere near as easy to verify. It's interesting that you reported the obvious method for (b) but no details on how you verified (a). I find that itself suspiciously unreliable.