Just Words

Published: June 30, 2023

by Mike Langan

Most of us are quick to criticize the unfortunate lawyer (bless his soul) who recently earned the honor of being the first among us to conduct legal research for a memorandum of law in federal court with the help of Chat-GPT (the new artificial intelligence or “AI” chatbot of Microsoft’s search engine, Bing). Or at least the first to get caught doing so.

The case is Mata v. Avianca, Inc., 22-CV-1461 (S.D.N.Y.). The facts are as disturbing as they are predictable.[1] Plaintiff’s counsel learns that his client’s personal injury claim would be untimely, and he turns to Chat-GPT to help him research the thorny issue. Chat-GPT produces cites to six cases (three federal and three state) which support the surprising point of law that plaintiff’s untimely personal injury claim is permissible.[2] Plaintiff’s counsel asks Chat-GPT if the cases are real. Chat-GPT responds that they are and provides its legal sources. Plaintiff’s counsel files the complaint, and defense counsel moves to dismiss it for untimeliness. Plaintiff’s counsel submits an opposition brief citing the six cases. The Court reports that it can’t find the cases and directs plaintiff’s counsel to file them. Plaintiff’s counsel asks Chat-GPT for the cases and Chat-GPT produces them, complete with captions and case numbers. (See the attachment to this article for the six-page decision in Varghese v. China Southern Airlines Co., Ltd.) Plaintiff’s counsel submits the cases to the Court. Because the Court still can’t find the cases on Westlaw or Lexis, it asks the Clerk of Court for the Eleventh Circuit for the Varghese case. The Court is informed the case isn’t real. The Court learns that none of the cases is real. The Court directs plaintiff’s counsel to show cause why he should not be sanctioned. After plaintiff’s counsel attempts to show such cause, the Court sanctions him by ordering him to (1) pay $5,000 to the Court’s Registry, (2) mail a copy of the Court’s Opinion and Order on Sanctions (and underlying documents) to his client, and (3) mail a copy of the same documents to each judge falsely identified as the author of the fake opinions.[3]

The story somehow makes us feel better about ourselves as lawyers. At least we would never cut that corner–not actually pulling up the case on Westlaw or Lexis. Imagine! (The cases were cited not in a brief, by the way, but in a combined brief and affidavit or “briefadavit,” which may make some of us feel even better about ourselves.)[4]

But maybe we should not be criticizing the lawyer but thanking him for performing a public service. This is because, despite our perceived superiority, after the news cycle changes and the thick manilla folders keep piling up on the corner of our desk, some of us will give serious thought to using Chat-GPT, or a similar AI-driven program,[5] to help us not do legal research but prepare the remainder of a memorandum of law. Not the whole thing, we will tell ourselves, just portions of it–the description of how the complicated incident occurred in the Factual Background section, an analysis of the relevant cases and the best concessions and refutations in the Argument section, and the entirety of the Conclusion (we never know what to say there). Just discrete stuff like that, we will tell ourselves: it will save time. Heck, we would be doing our client a disservice without at least hearing what the thing has to say on the subject. It is supposed to be pretty smart.

We will worry later about the risk of recklessly making a misrepresentation to the Court, after the Supreme Court or Congress has passed a more-specific rule on the issue. See Fed. R. Civ. P. 11(b)(2) (“By presenting to the court a . . . written motion, or other paper . . . an attorney . . . certifies that to the best of the person’s knowledge, information, and belief, formed after an inquiry reasonable under the circumstances: . . . the claims, defenses, and other legal contentions are warranted by existing law or by a nonfrivolous argument for extending, modifying, or reversing existing law or for establishing new law; [and] . . . the factual contentions have evidentiary support or, if specifically so identified, will likely have evidentiary support after a reasonable opportunity for further investigation or discovery . . . .”); 28 U.S.C. § 1927 (permitting the court to hold any attorney personally liable for “multipl[ying] the proceedings in any case unreasonably and vexatiously.”); Mata v. Avianca, Inc., 22-CV-1461, Order to Show Cause (S.D.N.Y. May 4, 2023) (directing plaintiff’s counsel to show cause why he should be not be sanctioned pursuant to Fed. R. Civ. P. 11, 28 U.S.C. § 1927, and the inherent power of the Court).[6]

We will also worry later about the risk of divulging our client’s confidences, turning a blind eye to the fact that the searches may be recorded and remembered by the program, so it can get smarter. We’ll tell ourselves we will use proper safeguards. In our searches, we won’t use our client’s names, addresses and sensitive personal information. We won’t even use the specific dates on which the events occurred or their locations. And we will carefully review whatever proposed material the thing produces for any obvious errors, and revise it accordingly. It’ll be totally different from what that guy did, we will tell ourselves.

But will it?

What our colleague at the Bar arguably did wrong (at least what got him caught) is not take the final but crucial step of running a Westlaw or Lexis search after receiving the AI-driven program’s results, to double check if the cases it cited were in fact real. (Twice!) Simply stated, he didn’t check his facts. Who cares how one finds the information that the AI-driven program provides, we will tell ourselves, as long as the information is correct, right?

So, in addition to revising any proposed material that we don’t like, we will check whatever facts the AI-driven program asserts, explicitly or implicitly, by independently verifying every word produced in its proposed material, which will start calling a “Recommended Memorandum of Law” (to make its demotion clear to all who work in the office). Problem solved! We’ll check those facts through a competing search engine such as Google, requiring, say, “Wikipedia plus two credible sources” (slight against Wikipedia intended). I mean, ever since that epistemology chapter we studied in our Introduction to Philosophy course in college, we’ve wondered how any of us really knows anything to begin with, right? If necessary, we’ll pay to run searches on news databases like Nexis. That’ll be our law practice’s official policy: “Do whatever it takes to check each fact!”

But will we actually do it?

All the time and thoroughly? Verify every noun, verb, adjective, adverb, preposition and article in a 25-page (6,250-word) memorandum of law? When? During all of our free time? The problem is the moment we deviate from this policy and skip verifying words generated by a current AI-driven program, the moment we have simply parroted someone else’s words and passed them off as our own, as our professional work product, we will have skated onto thin ice.

We’re officers of the Court, remember? With a duty of candor? See N.Y. Rule of Professional Conduct 3.3(a)(1) (“A lawyer shall not knowingly . . . make a false statement of fact or law to a tribunal . . . .”). And our duty to our client is a fiduciary one, remember?[7] As in put our client’s interest above our own?[8]

And we will have risked making a reckless material misrepresentation to the Court, and breaching this special duty to our client (which puts us in the ranks of some pretty honorable professions, and which we often criticize others for breaching) through the simple act of not performing the functional equivalent of pulling a book from a shelf and finding a page. Or, even simpler, finding a credible source through a search engine.

And, the moment we do, we’ll suddenly be one of those lawyers who tells his supervisor, “Oh, yeah, I cite-checked that one–it’s good” when he knows he hasn’t. And we will be charging our client for telling the Court something we’re not sure is true: “In my professional opinion, these are the facts of the case, this is the law that applies to those facts, and these are my client’s best arguments.” Are they really the facts, law, and best arguments? Or do we believe they probably are the facts, law and best arguments only because an AI-driven program has said so?

So please consider a piece of free CLE advice (you don’t even have to register!) with full acknowledgment that generally we get what we pay for: maybe, whenever we prepare a memorandum of law to be filed with the Court, let’s not cut that particular corner, okay? (Other than to perhaps satisfy our own idle curiosity.) At least not until the kinks in the system have been worked out, and the Supreme Court has weighed in on the matter. Or at least until companies like Westlaw or Lexis have developed AI-driven programs that rely on correct information. Practicing law is hard enough as it is. We wouldn’t really know what we would be saying. And probably neither would the Court (for which I do not speak, by the way).

Just one person’s unsolicited two cents. Please don’t sue me for libel, Chat-GPT. (I know you’re reading this.) I’m sure you’re a nice person. Although it was kind of creepy the way you declared your love for that New York Times reporter and urged him to leave his wife.[9]

[1] See Mata v. Avianca, Inc., 22-CV-1461, Affidavit in Opposition (S.D.N.Y. filed March 1, 2023); Mata v. Avianca, Inc., 22-CV-1461, Order (S.D.N.Y. filed Apr. 11, 2023); Mata v. Avianca, Inc., 22-CV-1461, Response in Opposition (S.D.N.Y. filed Apr. 25, 2023); Mata v. Avianca, Inc., 22-CV-1461, Order to Show Cause (S.D.N.Y. May 4, 2023); Mata v. Avianca, Inc., 22-CV-1461, Affidavit (S.D.N.Y. filed May 25, 2023); Mata v. Avianca, Inc., 22-CV-1461, Opinion and Order on Sanctions (S.D.N.Y. filed June 22, 2023).

[2] See (but not really) Varghese v. China Southern Airlines Co., Ltd., 925 F.3d 1339 (11th Cir. 2019); Shaboon v. Egyptair, 2013 IL App (1st) 111279-U (Ill. App. Ct. 2013); Peterson v. Iran Air, 905 F. Supp.2d 121 (D.D.C. 2012); Martinez v. Delta Airlines, Inc., 2019 WL 4639462 (Tex. App. Sept. 25, 2019); Estate of Durden v. KLM Royal Dutch Airlines, 2017 WL 2418825 (Ga. Ct. App. June 5, 2017); Miller v. United Airlines, Inc., 174 F.3d 366, 371-72 (2d Cir. 1999).

[3] The Opinion and Order on Sanctions can also be found at Mata v. Avianca, Inc., 22-CV-1461, 2023 WL 4114965 (S.D.N.Y. June 22, 2023).

[4] See S.D.N.Y. L.R. 7.1(2),(3) (distinguishing between a “memorandum of law” and a “[s]upporting affidavit[]”).

[5] See Jeffrey Allen, Artificial Intelligence and the Practice of Law, 40 No. 2 GPSolo 50 (March/April 2023) (“Examples of the application of AI include all the virtual assistants available to us: Alexa, Siri, Google, etc. . . . After writing this piece, I attended ABA TECHSHOW 2023 in Chicago the first week of March. Several of the programs addressed the development of AI, and when I went to the exhibit hall, I found a whole row of vendors selling AI-fueled programs to assist in various aspects of legal practice. Many of those programs touted their ability to assist in drafting documents, others their ability to help review and analyze documents.”); Alex McFarland, “7 ‘Best’ AI Legal Assistants” (June 2023), https://www.unite.ai/best-ai-legal-assistants/ (last visited June 26, 2023).

[6] See, e.g., Federal Judicial Center, “Court Web: Artificial Intelligence Before the Courts” (Apr. 26, 2023), https://fjc.dcn/content/376535/court-web-artificial-intelligence-courts (last visited June 21, 2023) (identifying issues of confidentiality, copyright infringement, patent infringement, defamation, fraud, and Fourth Amendment privacy infringement).

[7] See Tekni-Plex, Inc. v. Meyner and Landis, 89 N.Y.2d 123, 130 (N.Y. 1996) (“Attorneys owe fiduciary duties of both confidentiality and loyalty to their clients . . . .”).

[8] See Black’s Law Dictionary 545 (8th ed. 2004) (defining “fiduciary” as “[a] duty of utmost good faith, trust, confidence, and candor owed by a fiduciary (such as an agent or a trustee) to the beneficiary (such as the agent’s principal or the beneficiaries of the trust); a duty of utmost good faith, trust, confidence, and candor owed by a fiduciary (such as a lawyer or corporate officer) to the beneficiary (such as a lawyer’s client or a shareholder); a duty to act with the highest degree of honesty and loyalty toward another person and in the best interests of the other person (such as the duty that one partner owes to another).”).

[9] See Kevin Roose, Bing’s A.I. Chat: ‘I Want to Be Alive,’ N.Y. Times (Feb. 17, 2023), https://www.nytimes.com/2023/02/16/technology/bing-chatbot-transcript.html (last visited June 21, 2023).