The AI Apocalypse has Likely Already Hit Maryland’s Appellate Courts—What Mischief Can Be Expected, And What if Any Rules Should Apply?
Claude Monet, was considered the Father of the Impressionist movement, beginning in the latter half of the 1800s.[i] Impressionism went against “classic” principles of painting in many renaissance and baroque style works, usually by painters with academy training. Artists of the caliber of Michelangelo or Peter Paul Rubens, would begin art projects, taking years if necessary to painstakingly captured them with full details. Impressionism was on some level, the opposite. The best impressionists like Monet, could do a painting quickly onto a canvas, and without the level of detail that, at least through the end of the 19th Century, was necessary to be considered a masterpiece.
Well, good or bad, (mostly bad), impressionism, to use a metaphor, has come to appellate litigation, in the form of “Generative” AI. And with that sea-change on the horizon, at least without time-sensitive concerns of specific applicable Rules, Maryland appellate courts should expect a rapid increase in the number of appeals, potentially rising to a level reminiscent of the increase in appeals from the 1960s, and overall decrease in quality of appellate briefing. (It was this increase on expanded application of Federal Constitutional rights, that lead to the intermediate appellate courts in the United States, including the Appellate Court of Maryland being formed, to address the upsurge in criminal and post-conviction appeals.)
There have been recent efforts primarily in some Federal appellate courts, to establish norms for disclosure and ethical Rules for the use of AI in appellate briefs. There has been, at least publicly, zero specific AI regulations adopted in the Maryland appellate courts, to do the same, though Chief Justice Matthew Fader is one of the eight members of the “AI Rapid Response Team” of the Conference of Chief Justices.[ii] The lack of formal Rules are concerning, as explained further below, because (1) Maryland is less well equipped than Federal Court counterparts which are more obviously dealing with this concern from multiple Court decisions nationwide, and (2) the Maryland appellate courts will be significantly worse off and adversely impacted for when the AI storm has hit a critical mass, which might not allow for additional time optimistically contemplated, to formulate guidance for attorneys and the judiciary. Maryland should begin efforts, to add specific guidance, standards, and transparency, post-haste, before this storm hits.
Two previous blog posts of mine addressed this topic somewhat obliquely, in the context of legal research, not legal writing. Before the “rise” and prominence of ChatGPT began in late 2022, the posts back from only 2021 explained, (“Problems with Modern Electronic Legal Research—“Googlization” Versus “Boolean Logic” and a second part on the “Rosetta Stone” “exception” whereby even “Boolean Logic” searches would fail to discover existing precedential authority), the recommendations on the topic from the “Googlization” Post was as follows.
“Presently, I recommend algorithmic searches as a primary in narrow circumstances. First, is the rare circumstance of it’s unclear where to start, on an obscure or very new topic area, that I’d be surprised if there were any caselaw, which may justify first checking secondary sources like law review articles. Second, is when a “Boolean-logic” search does not provide a satisfactory answer directly on the legal issue, after tracking conventional leads from cases, statutes, and secondary sources. “
The recent advent of ChatGPT, and other “Generative” AI systems, will likely make the concerns of “algorithm” uses in legal research, a comparably small concern. This is because the legal research concern will be subsumed by appellate briefs themselves mostly written and delegated to and by AI, will also include legal research, with no specific ethical or Rule requirement attorneys be familiar with appellate work at all when doing so, or fact-checking them fully or properly. Thus, the basic suggestion when using algorithms approaching AI-levels, was it should be done primarily to supplement legal research, not become a presumptive use or foundation for research. This still remains my present recommendation, but it’s now more complicated, because the Generative AI will allow at least give the appearance on a potentially complete-looking Brief, for when more attorneys are doing regular appeals especially on their own cases. When in reality, these can be quick “impressionist” versions of the type of work most appeals demand, and which clients are paying for.
This could theoretically work, if the practitioner fully understands the ins and outs of what makes a good brief. I won’t focus on that concern in this post, but Questions Presented, Standards of Review and properly persuasive facts are usually quite helpful. However, without a necessary practical understanding by the practitioner, to recognize when the work product is wrong, and potentially structural appeal issues unexamined, (so that the final product ends up being mere gloss), this has the risk of greatly expanding the numbers of appeals in the short term, at least with the present state of Generative AI.
After a few more noteworthy examples of seemingly absurd delegation of AI in June 2023, (such as in New York when the attorneys used Generative AI in opposition to a Motion to Dismiss, which manufactured the names of fake cases, and even after the Court asked for the origin of the cases the same attorneys still relied on the fake cases before admitting it was ChatGPT), the Federal Courts initially appeared to quickly address these concerns. The Fifth Circuit Court of Appeals out of New Orleans, in November 2023, sought to establish a Local Rule, on using AI on appellate Briefs, but recently decided against a specific AI Rule after attorney pushback, saying existing ethical and 5th Cir. Rules were sufficient.[iii] This year, the Third Circuit and Ninth Circuit Courts of Appeals, set up committees to examine adopting new AI Rules. The Fourth Circuit and Second Circuit has, also begun more informal conversations on the topic. Some States, like Florida, have seen it necessary to explore new Rules adopting the Federal Rule 11, and have had at least one Bar Counsel referral, but as of yet, have zero affirmative disclosure of AI use by Florida Bar members. Perhaps precognitively recognizing the enormity of the challenge for Federal Courts, Supreme Court Chief Justice John Roberts’s Year-End Judiciary Report, focused on the development of the law from quill pens, to typewriters, to computers, and ending with a quixotic discussion of how Generative AI will likely play a pivotal role in the future of the judiciary. But notably, the Chief Justice did not address what Rule changes are important or necessary to consider implementing.
This present ethical and rules void, in combination with the rapid development of ChatGPT, has a number of law firms, increasingly publicizing the use of AI in their work products, and that’s from the firms that want to disclose they are using AI at all, which may be a negative for some clients. (I won’t cite to these news articles or press releases here specifically, as it’s becoming increasingly non-novel). The Maryland Law schools, seem to recognize AI use can be potentially positive, on one hand, saying it’s a good thing that more people can use AI to bring to the forte more genuine legal cases that were cost-prohibitive, but comes with it a recognition that law students will need to have significant guidance on ethically using AI. [See Maryland Daily Record, “How Maryland’s Law Schools Approach AI in Curriculum, research,” August 21, 2024, Konieczny, Rachel) (Behind paywall)
And to be fair, it’s not a one-sided worry about lawyers and their staffs using AI, on appeal briefs—judges and law clerks also may be tempted to try and use it, especially if done in response to an increase in legal cases and work, that itself may trace to generative AI, beginning a sort of “AI Cold war” with ever increasing use by attorneys, necessitating increased use by the judiciary. In Nevada, the backlog for the agency responsible for unemployment appeals, has led them to conclude it is best to start using Generative AI in deciding cases, concluding this was fine as “[a]ny AI decision will also be double checked and reviewed by a human referee before it is handed out.” Judge Kevin Newsom, a Harvard Law graduate, presently on the federal Eleventh Circuit Court of Appeals, recently publicized and admitted to using and being somewhat impressed with using ChatGPT for future appellate cases, in a Concurrence on one of the Court’s Opinions.
So with this background, while the Federal Courts are openly worried, and at least in the 2nd, 3rd, 4th, 5th, and 9th Circuits claim to be publicly exploring the matter through possible new Rules on use and ethics of Generative AI, why are Maryland appellate courts more suspectable of becoming significantly overwhelmed through Generative AI now? It’s not as if there’s not some appreciation, at least from an evidentiary perspective that the “growth and proliferation of artificial intelligence, especially generative artificial intelligence, are real and pressing,” as Chief Justice Matthew Fader, noted in a brief concurrence in Mooney v. State, Sept. Term, 2023 (Dec. August 13, 2024). (Chief Justice Fader also noting there was zero indication or “hint” AI played a role in that particular case, and in a case with such a concern, “may ultimately require us to adjust our rules and procedures for authenticating electronic evidence.”)
The Maryland appellate courts may be relying on some of the initial conclusions by a Stanford paper on AI use, which were summarized in the recently published August 7, 2024 “Artificial Intelligence: Guidance for Use of AI and Generative AI in Courts […] from the AI Rapid Response Team at the National Center for State Courts” and noted:
“The following are indications that GenAI may have been used to create a document” which includes:
• References to cases that do not sound familiar, cannot be found through traditional legal research, or have unfamiliar citation formats.
• At first read, AI text may sound impressive and well written, but there are often structural issues. AI content tends to be overly formulaic and lacks natural transitions between topics. Once you strike out all the words that are meaningless filler, there may not be a lot of substance left. AI is also not mindful of grammar rules or basic punctuation although that is improving.
• AI is designed to recognize patterns and replicate them as accurately as possible so look for repetitive patterns in the writing. Perhaps the most obvious sign of AI-generated content is the use of repeated words, phrases, or the same sentence structure used regularly in different paragraphs within the same document.
• Often AI generated content is written in the general sense, glossing over facts and figures and may be lacking details, unnatural phrasing, lack of natural transitions between topics, or errors that a human is less likely to make. It often uses alliteration to articulate an appealing word arrangement.[iv] […and other indications]”
These conclusions are relatively good indications the appellate Brief, is being primarily drafted and conducted through Generative AI. I expect the “hallucination” stage of early ChatGPT AI, which led to the New York Sanctions, to be a correctable, and less prominent, after all the negative publicity. (It’s as simple as actually putting the cases through a designed legal database to “Cite-Check” and while the reading of the case by AI might be utterly wrong in every way, I would expect that these at least will be real case names and citations in the future.) But consistent with the Stanford Report, practitioners and members of the judiciary should expect: (1) an unnecessarily large percentage of citations to not be proper and primary, (2) unexpected citations to other states or federal cases that seem on-point, but fail to basically cite the main citable decisions in the jurisdiction involved, (3) inexplicable reliance on unpublished opinions, not distinguishing between Federal and State or between State Supreme Court and intermediate appellate court cases, and (4) this should continue for a time period while ChatGPT and other Generative AI systems with a legal tinge, continue to progress and better establish themselves.
Questions Presented and Standards of Review are critical to most successful appeals. Expect, at best, pedantic and unhelpful Questions Presented to become more and more the norm, some that basically lose the case by say conceding an “Abuse of discretion” standard in the QP (which may be accurate, but are hardly persuasive to locating potentially reversible errors of law),, but others, being no more illuminating than “Did the trial Court commit reversible error, in granting the Motion for Summary Judgment?” Expect there will also be no citations to Shakespeare, T.S. Eliot, or anyone else interesting or poignant, to make the judiciary audience interested, for this strain of AI-generated Briefs.
Record reviews and obtaining a Record Extract, are another major difficulty for AI at the present. They will completely rely on a human to understand and do these, taking out a major step in the traditional work done on appellate Briefs. There needs to be basic facts and evidence admitted, to understand and appreciate what legal arguments should be argued. An AI might be able to help with the “legal component” but most appeal cases, aren’t purely a legal issue. Grammar and legal writing have been in my opinion, on a downslide, but oddly, it’s not necessarily because an AI-adjacent system like Grammarly can’t be helpful, if done properly.[v]
The reasons why Maryland is most susceptible to negative AI use on appeals, and will create an excuse for greater quantity, and not quality of appeals, are three-fold. First, is there are different Rules between the Federal and Maryland Courts, and there are actual District Court decisions that are exploring abuses of AI by some attorneys already. As the only Federal Circuit to come to any sort of conclusion on a proposed AI disclosure Rule, the Fifth Circuit ultimately declined “at this time,” after most attorney firms objected. [vi] The primary reason given was because the Federal Rules have FRCP 11.[vii] Ultimately, the Fifth Circuit relied on their Local Rule, which is a more specific version of what’s in Maryland Rule 1-311 on “Signing of Pleadings and other Papers.” There, however, is a large difference and rub. Sanctions derive from the signature (now electronic signature), that there has been a “certification that the attorney has read the pleading or paper; that to the best of the attorney’s knowledge, information, and belief there is a good ground to support it and that it is not interposed for improper purpose or delay.” A natural reading of Maryland Rule 1-311, merely notes that the attorney has read the paper, and “best of the attorney’s knowledge […]” there’s a good ground of support. But that appears to be easy to comply with, at present, when an attorney can simply say they relied upon the AI system, that may be ChatGPT related, or the more modern incarnations of Lexis and Westlaw, that are implementing a generative AI system in conjunction with their legal database. At the very least, the Fifth Circuit’s “I used AI will not be an excuse for an otherwise sanctionable offense” is not consistent with, in any shape or form, present Maryland Rule 1-311.
Second, is that the Federal appellate Courts, especially the Fifth Circuit, which I happen to be a member of, simply do not have the same audience of attorneys as Maryland appellate courts. Both have their fair share of pro se litigants, who both now mostly use the informal briefing method. However, the Fifth Circuit, which arguably has the strictest use of local Rules amongst Federal Circuits (including a precise “electronic record” system,), like most Federal Courts of Appeals, are predominantly done by larger law firms that have multiple attorneys and/or paralegals familiar with federal appellate practice. This is partly due to the more strict and complex federal practice rules for appeals, but also the premium for law firms hiring former clerks of Federal Judges.
So there does not appear to be at present, with this combination of (1) stricter compliance rules, (2) not too many attorneys who aren’t already well familiar with appellate practice in the federal Courts, (and those that do, coming from larger law firms that have a number of appellate specialists). Thus, there shouldn’t be too many Federal Appeal attorneys, who would take the step of completely delegating an appeal, to a Generative AI-writing system.
Maryland, however, is not set up the same. There’s a very small number of regular “appellate specialists.” The lack of rules, legal, procedural, or ethical, on the use of AI, could easily lead to abuse, especially when there are financial motivations, of “quantity over quality.” Imagine, a small to medium law firm, churning out appellate brief after appellate brief, that’s 90% ghostwritten by AI, charging 30-70% less than the standard brief, but capable of being done by paralegals or recent law school graduates. There isn’t really hard to imagine, because this is now quite doable, but only in the past year, and there have almost certainly been appellate briefs that are now primarily the work product of Generative AI, argued in the Maryland appellate courts.
Third, is the issue of “ghostwriting” and that this issue perhaps should be more clearly examined to address the threat of Generative AI. In the Federal Courts, including Maryland Federal Local Rule 102, it appears that most “ghostwriting,” especially for self-represented parties, needs to be disclosed to the Court, which is consistent with most Federal Courts. Most states, including Maryland, are the opposite, following the ABA model, but with a strong minority of about 20 states having disclosure requirements for substantial assistance in ghost-writing. [viii]
While AI-Generated assistance is newfangled enough, it would not appear to fall under the definition of “ghostwriting” in the Federal and State Courts, perhaps it should be. (Though when Artificial Intelligence, is the “ghostwriter,” the issue becomes more meta, but perhaps that’s what is most concerning.) Regardless, the Federal Courts having general disclosure and even prohibition requirements against ghost-writing, and Maryland is the opposite, with zero ethical prohibitions. Thus, if it doesn’t occur to Maryland attorneys, as there are no prohibitions per se against most ghostwriting, and particularly on disclosure, making such a disclosure on AI, would seem more foreign in Maryland State Court, when required. Regardless, one of the ways the issue should be tackled and examined, are the differences on “ghost-writing” in practice, and in the established rules of the Federal versus Maryland courts.
If one doesn’t know to look out for them, the strange citations that upon examination are wrong or unpersuasive citations (to other jurisdictions, federal courts, unreported decisions, not being the actual holdings), and other clues outlined by the Stanford paper on AI use, may be helpful to keep in mind. However, as there’s presently few if any legal, practical, or ethical Rules, explicitly set up on the AI topic, expect AI-use in appellate briefs to increasingly become routine after going unaddressed by the Court system. (And thus, the argument could be made, that disclosing AI use, which might have gradations on the amount, could be educational, as to their utility, and success on appeals, especially amongst Appellants) And expect, after a few test runs, this newfangled AI-harvested appeal work, to possibly increase exponentially, requiring more appellate resources to address them, including by appellate staff attorneys, and not abate without standardization of actual rules on the topic.
While one day, Generative AI might be fully usable, to allow “Impressionist” attorneys familiar enough to do appellate law without AI-assistance to begin with, more quickly organizing and doing appellate Briefs, to conscientiously increase their appellate work caseload with AI, without sacrificing quality and originality. However, at this juncture, the AI-Appellate Court Apocalypse is a more likely outcome, because of the negative effects of worsened quality, and significantly increased numbers of appeals that may be coming to Maryland. I therefore would recommend Maryland require what the Fifth Circuit originally proposed (and partially adopted), and/or to adopt some of the Federal Rules in existence, which makes the Federal system more capable of addressing the matter in due course, but not dawdle on the topic. The Maryland appellate Courts should also endeavor, to not penalize, non-AI Briefs, should they end up handling an ever-increasing number of AI-produced Briefs. Furthermore, while awaiting a full exploration by the Federal Courts on the topic first is generally a good rule of thumb—Generative AI, should be considered an exception, as its introduction may end up being exponential, and the Maryland appellate courts may not have time to first carefully examine the issue, to stem a rising tide of AI-generated appellate briefs, that do not have to disclose even the objective fact, they originate from ChatGPT, Lexis, Westlaw, or other Generative AI-services marketed to attorneys.
Michael Wein is an attorney in Greenbelt whose practice concentrates on appellate, civil, and criminal litigation, and can be reached at weinlaw@hotmail.com.
[i] The term “Impressionism” itself comes from Monet’s “Impression, Sunrise” painting from 1872, showing a rising sun above Monet’s seaside hometown in France. They were to be done quickly, to help guarantee the “Impression” of moving targets, like a sunrise with Monet, or a landscape in the moving daylight.
[ii] There are a couple of States that have Judiciary-led and/or established AI Work Study Groups or Taskforces. Maryland does not have a Judiciary task force listed on the CFJ Web site, though there does appear to be a request for an AI Task Force group being set up through the Maryland Judiciary’s Court Technology Committee. The MSBA also appears to have a Task Force, though not apparently publicly, and no web site is linked to them.
[iii] With no small amount of irony, the Fifth Circuit’s Statement on AI, which stressed “I use AI” is not a tolerable excuse for improper arguments or citations in Appellate Briefs in that Court, has a citation error, on the sole citation included in the document. It references a reliance on Federal Rule of Appellate Procedure (FRAP) 6. However, FRAP 6, is on bankruptcy appeals. Reviewing the public comments solicited by the Fifth Circuit, finds the reference should have been to “Form 6” in the Fifth Circuit Local Rules, and that it’s merely a “Certification” requirement for attorneys filing cases electronically. Basically, part of the “check box” of certifications made when appeals are electronically filed in the Fifth Circuit.
[iv] To be accurate, having written shelves of appeal briefs, I have noticed my writings naturally have a lot of alliteration. That being said, I have never used generative AI on anything.
[v] What’s particularly very helpful for legal writing and limited word space, is to constantly include active voice, and removing passive voice in legal filings. However, this is no longer part of the “free” system but requires a monthly premium subscription to Grammarly. The free system isn’t much better than what Microsoft Word offers and will not correct the significant run-on sentence or when the passive voice is 30 words, that could be expressed in ten words with proper active-voice legal writing. My recommendation, for those who insist on using the product, is to understand most grammar rules, and take the time to do further edits, so the use of Grammarly is limited to a final review and/or rendered completely unnecessary
[vi] Some of the attorneys in public comments to the Fifth Circuit, noted they were already making use of Lexis and Westlaw AI programs, that offer a competitive program amongst themselves, but at much greater cost to attorneys, because it portends to do a substantial amount of the legal work in briefing and complex legal motions with “ease.”
[vii] Though the Federal Rule 11 has a “safe haven” clause, which makes that Rule actually easier to avoid sanctions (though with perhaps better objectivity) than the corollary Maryland Rule 1-341 on sanctions on “Bad Faith; Substantially Unjustified Proceedings.”
[viii] At least one prior Maryland Bar Counsel used to recommend in favor of ghost-writing letters to the AGC. This perhaps confused the practical application and ethics of the issue even more, by giving the appearance, ghost-writing is considered ethical and beneficial in all circumstances. This full-throated recommendation that attorneys “ghost-write” matters without disclosures, gives short shrift to the contrary position, and although Maryland’s position is consistent with the ABA position, should be considered controversial, against the majority of Federal jurisdictions Rules, as well as a twenty-State minority, having some sort of “disclosure” requirement, particularly when substantial assistance is involved.
Maryland Appellate Blog 