TomorrowLawyer: Scholar, Tribune, & Operator

Another day, another court ruling on some litigant allowing AI “hallucinations” to make their way into a court filing. As reported by Eugene Volokh, a federal court in New York discovered hallucinations in a pleading in their court. The case highlights a wide misunderstanding: AI hallucinations are more than completely fabricated citations, they also include modifications and misrepresentations of the language of real cases. The trial judge found:

“Lead counsel reviewed the cited cases in LexisNexis and confirmed that they were actual decisions relevant to the legal issues in this matter. However, the quotations included in the initial draft of the Memorandum were not independently verified against the official opinions word-for-word.

The cases were real but apparently the AI tweaked the quotation from the underlying case.

In another recent case, the counsel’s AI really displayed a full range of hallucinations. Judge Nina Wang in Coomer v. Lindell (D. Colo.) found:

“These defects include but are not limited to misquotes of cited cases; misrepresentations of principles of law associated with cited cases, including discussions of legal principles that simply do not appear within such decisions; misstatements regarding whether case law originated from a binding authority such as the United States Court of Appeals for the Tenth Circuit; misattributions of case law to this District; and most egregiously, citation of cases that do not exist.”

Seemingly, locating and calling out AI hallucinations is the judiciary’s crusade de jour. Sanctions are flowing. Courts are requiring lawyers to report AI hallucinations to disciplinary bodies.

Humans make these same type mistakes. The profession’s sudden vigilance about AI hallucinations exposes a uncomfortable truth: we’ve tolerated human citation errors for decades. The scrutiny itself, not the source of the error, drives the current enforcement surge. For instance,

“A prosecutor recently used artificial intelligence in preparing a filing, which resulted in an inaccurate citation, Wilson said in the statement to The Sacramento Bee. ‘Once the error was discovered, the filing was immediately withdrawn. Wilson’s also said his deputies had filed briefs with inaccurate legal references in two other cases, but that they were caused by human error, not AI.‘”

Why the sudden examination at a “word by word” level of scrutiny? My suspicion is that a judge has discovered a citation to a case that does not exist. Once discovered, the judicial microscopes come out. Every line of a brief or pleading is examined. But as was argued by Nevada County District Attorney Jesse Wilson:

“Just because the Turner case involved AI, ‘it cannot now be assumed that every citation error stems from the use of artificial intelligence,”

I agree.

From my twenty-six years of practice, human “hallucinations” have been occurring a long time before AI ever came along. Many, many times I have observed “misrepresentations of principles of law associated with cited cases.” Once, I even suggested to the court that my opposing counsel were “not reading out of the same law books!”

For example, consider the series of appellate cases involving discovery in felony criminal cases in Alabama district courts. In State v. Brown, 259 So. 3d 683, the Alabama Court of Criminal ruled that a district court merely could not order discovery after the court had bound the matter to the grand jury. Fairly clear and simple ruling. However, for the next three years the Alabama Attorney General’s office and prosecutors across the state “hallucinated” that Brown stood for the proposition that “the district court’s limited jurisdiction . . . did not include the authority to issue a discovery order.” If not conclusively resolved in a later ruling by the Supreme Court of Alabama, the issue was finally been put to bed by Dees v. State, 351 So. 3d 567.

And misrepresentations are limited to attorneys; court themselves have historically been known to “hallucinate.” In an SMU Law Review, Professor Luedeman highlights many examples and different ways that courts actually misstated and misquoted the law. For instance,

“An appellate opinion can also cause confusion by contradicting earlier binding precedent on which it purports to rely. To be sure, appellate courts often have the prerogative to revisit and modify standards from their precedents—indeed, that is often how the law develops over time. But what about when the modification appears not to have been deliberate but instead to have been based on a misunderstanding or misquotation of the earlier precedent? . . .

In addition, inter-opinion contradictions can result from seemingly tiny modifications to wording . . . But in Abdul-Akbar v. Watson, . . .it ended up not only misquoting a precedent but also changing its underlying logic.”

Unfortunately, I could not find any empirical studies which examine the frequency of misquotations and mis-citations by human lawyers before advent of 2023.

My suspicion is that if the same word-for-word microscope were in fact pulled out on pre-AI pleadings, especially in state courts, we would find them replete with similar “hallucinations” as attributed for AI products.

Two simple reasons that these were not caught: first, opposing counsel did not cite check every single citation in a motion or brief (and especially did not pull up the case and read it closely). Secondly, the trial judges frequently didn’t even read the briefs at all! So if the opposing counsel didn’t catch the misstatements, the hallucinations, it wasn’t caught.

Perhaps, now with renewed scrutiny of every citation and the advent of tools like Lexis+ Brief Analysis, an AI empowered tool, human “hallucinations” will be caught and quartered anew. The Lexis+ tool enables you to upload your own brief or your opposing counsel’s briefs into the interface, and not only Shepardize each of the cases cited but also “Use Quote Check to validate that quotes cited in the document are fully consistent with the actual language used by the court.”

I think all of this this will require more work for lawyers, demand greater diligence for precision in every brief you write and require greater scrutiny of your opponent’s filings as well.