When I was a young attorney, researching briefs for a court was a laborious and time- consuming process. For each topic, or point, in a brief, the lawyer often began with a legal encyclopedia to find a principle of law that fit the facts of his or her case. These were found in American Jurisprudence (Am. Jur.) or Corpus Juris Secundum (C.J.S.). Next, to find relevant cases by subject, the attorney consulted a digest using West’s “Key Number” system which listed summaries of on-point cases and directed him or her to case reporters for each jurisdiction where the cases were printed chronologically.
The lawyer then read the cases to see how closely they matched the point the brief was supposed to make. The most persuasive cases were written by the highest court in the jurisdiction where the court action was, although cases in other jurisdictions, particularly federal cases, could be persuasive, too. Because cases could be overturned at any time, it was necessary to “Shepardize” each cited case (i.e., check whether it was referenced in future cases) to see if a more recent case followed, questioned, overturned or modified it. Paper “pocket parts” supplemented the Shepard’s bound volumes to provide the most up-to-date precedents.
With the advent of computers in the late 1970s and the 1980s, research tools developed by Lexis and Westlaw aided this process, particularly in large corporate law firms. Heavy use of these computer tools was sometimes discouraged because of their high cost, which could not always be passed on to clients, and because the increased associate efficiency meant fewer billable hours to charge the client.
Nowadays, Artificial Intelligence tools make it possible for the machines to conduct much of the research, to quote relevant case law -- and to draft parts of the briefs themselves. Because AI agents are not infallible, this has led to a phenomenon known as “AI hallucinations.” In the case of Mata v. Avianca, a lawyer with thirty years of experience used ChatGPT to do legal research in a personal injury case. The AI found cases with applicable legal precedent, including citations and quotations. The problem was that none of those cases existed in real life. The AI “hallucinated” them; that is, the machine made them up out of whole cloth. When opposing counsel told the judge they could not find some of the cases cited by the lawyer, the judge ordered the lawyer to check his cited authorities. But instead of looking up the cases himself -- or even using Lexis or Westlaw to confirm they existed -- he asked ChatGPT whether the cases were genuine. ChatGPT told him the cases were real and could be found in computer databases.
The judge held that the filing lawyer, not AI and not another lawyer or any other source, stands behind court filings. This 2023 ruling has been expanded upon by the federal Court of Appeals for the Seventh Circuit in Dec v. Mullin. There, the court found several cited decisions that did not exist and other “tell-tale signs of AI hallucinations.” The problem is not getting better. According to one source, approximately 1800 cases (1200 in the United States) have had AI hallucinations.
Technology has changed, but good lawyering has not. Lawyers should read every case cited in their briefs and check every quotation because they -- and only they -- are responsible for the content of what they sign off to the court.