An Overview of Courts’ Treatment of Experts’ Use of AI

AI is a promising tool with the potential to transform the legal industry. Despite publicized misuse, Courts have acknowledged that “AI, in many ways, has the potential to revolutionize legal practice for the better.”¹ Much has been written and discussed about attorneys using AI. Here we look at how it may be used by expert witnesses. Like attorneys, AI use by experts creates risks related to false citations and overreliance on AI when drafting reports or declarations. The consequences can result in damaged credibility or the expert’s exclusion by the court. Even with these dangers, AI can be a helpful research tool, and courts do not prohibit its use. This article examines recent cases involving exert witnesses’ use of AI and how that AI use has been treated by courts.

A common danger of using generative AI is its tendency to misrepresent facts and provide citations to sources that do not exist, also known as “hallucinated” citations. As some attorneys have learned the hard way, using AI to assist in writing briefs can lead to courts imposing harsh admonitions for failing to ensure that AI-generated work product is accurate and properly cite-checked. For example, in Mata v. Avianca, a court imposed sanctions on an attorney for including fake, AI-generated legal citations.² Likewise in, Park v. Kim, the court referred an attorney for potential discipline for including fake, AI-generated legal citations in a filing.³

Submitting work product that includes AI-generated citations is not unique to lawyers, and courts have treated expert witnesses’ failure to verify citations similarly. For example, in Kohls v. Ellison, an expert declaration included citations to “non-existent academic articles” that “originated from [the expert’s] use of … a generative AI tool.”⁴ Even though the expert provided a “helpful, thorough, and plausible explanation” for the hallucinated citations and the court did “not believe that [the expert] intentionally cited to fake sources,” the expert was denied the opportunity to file an amended declaration.⁵ The court explained that “[c]ourts do not, and should not, ‘make allowances for a [party] who cites to fake, nonexistent, misleading authorities.’”⁶ With the expert unable to amend his declaration, the court determined that the expert’s “citation to fake, AI-generated sources in his declaration … shatters his credibility with this Court” and then excluded the expert’s declaration in its entirety.⁷

Ledoux v. Outliers, Inc., provides another example of an expert submitting a report including “hallucinated” AI-generated citations.⁸ In this instance, the plaintiff tried to preserve the expert report by withdrawing the portions of the report containing the hallucinated citations.⁹ Even with those portions withdrawn, the court, quoting Kohls v. Ellison, nonetheless determined that the expert’s “multiple hallucinated citations ‘shatter[ed] his credibility with this Court.’”¹⁰ Specifically, the court stated that the hallucinated citations demonstrated the expert’s “lack of diligence in verifying the sources that he represented were the basis for his opinion [and] prevent[ed] Plaintiff from meeting her burden to show that [the expert’s] opinion is ‘based on sufficient facts or data,’ is ‘the product of reliable principles and methods,’ or ‘reflects a reliable application of the principles and methods to the facts of the case.’”¹¹ Under the circumstances, the court thus excluded the entire report as unreliable.¹²

Not all AI use results in the exclusion of an expert. In Ferlito v. Harbor Freight Tools USA, the defendant sought to exclude an expert report when, “after completing the [expert] report, [the expert] entered a query into ChatGPT about the best way to secure a hammer head to a handle, which produced a response consistent with his expert opinion.”¹³ Here, the court determined that there was “little risk that [the expert’s] use of ChatGPT impaired his judgment” because the expert “used ChatGPT after he had written his report to confirm his findings.”¹⁴ The court did not exclude the report, reasoning that “[t]here is no indication that [the expert] used ChatGPT to generate a report with false authority or that his use of AI would render his testimony less reliable.”¹⁵ Other courts have likewise noted that they “do[] not fault [experts] for using AI for research purposes.”¹⁶

Traditionally, drafts of reports prepared by an expert witness are not discoverable.¹⁷ Despite this protection, “expert witness’s methodology is fair game for discovery”¹⁸ and “any analyses performed by the expert underlying the opinions to which he will testify are discoverable.”¹⁹ Parties may also agree to further limit the scope of discovery related to an expert’s report, such as precluding an expert’s notes from the scope of discovery.²⁰

An expert witness using AI to assist in drafting his report and/or declaration raises the likely possibility of creating discoverable information which may become available to opposing parties. For example, in Conservation Law Found. v. Shell Oil., an expert used AI to “cull down” produced documents “into a subset to be worked with.”²¹ The court determined that the expert’s AI prompts were not protected by the parties’ Rule 29 discovery agreement, in which the parties’ had agreed “not to take discovery of each other’s ‘expert notes, drafts, or communications needed by, and made during, the report drafting process.’”²² The court held that this use was “an aspect of [the expert’s] methodology” and that the AI prompts used by the expert were within the scope of discovery under Rule 26(b).²³ Similarly, in Laake v. 3M Co., an attorney was able to demand and receive, mid-deposition, over 300 pages of AI prompts the expert used in preparing his report.²⁴

One strategy for dealing with excessive AI use by an opposing expert may simply be to examine that expert about his AI use at trial. In Laake v. 3M Co., rather than move to exclude an expert based on the expert’s use of AI, the plaintiff subpoenaed the expert for direct examination at trial despite this expert having been intentionally left off the defendant’s witness list. The trial testimony revealed that the prompts used by the expert equated to nearly ninety percent of the report having come from AI.²⁵ The expert further admitted that, if the AI generated a response unfavorable to the defendant, he would revise the prompt and try again.²⁶ One of the prompts the expert used was that he “need[ed] to create an expert witness report to defend 3M’s standard of care” and “need[ed] to show that 3M met the standard of care and why.”²⁷ The expert then uploaded hundreds of files for the AI to analyze before asking it to create an entire first draft of the report.²⁸ The expert also used prompts to “[s]how how 3M is zero percent at fault.” The AI told the expert that finding the plaintiff “zero percent” at fault would make him look biased, and the expert then removed this requirement.²⁹ This testimony likely damaged the defendants’ case, leading to an eventual $61.5 million jury award for the plaintiffs.

It is well established that to be admissible, an expert witness’s testimony must be based on the expert’s “scientific, technical, or other specialized knowledge,” the testimony must be “based on sufficient facts or data,” must be “the product of reliable principles and methods,” and the expert’s opinion must “reflect[] a reliable application of the principles and methods to the facts of the case.”³⁰ An expert using AI must still satisfy these requirements. Numerous questions remain about how an expert’s use of AI may affect these inquiries. For example, can an expert ask AI to perform an analysis on which the expert bases an opinion? Can an expert ask AI for its opinion before independently formulating his own, or would doing so introduce bias that renders the expert’s opinion unreliable? If an expert asks AI for assistance in developing an opinion, is the result the expert’s opinion or AI’s? Can an expert adopt an opinion partially or entirely formulated by AI? AI presents a new frontier for opposing counsel to challenge expert opinions when AI is involved. Further clarity will come as courts have to make rulings on these uses and experts’ AI reliance.

There are also unanswered questions about the permissible scope of discovery into an expert witness’s use of AI. For example, is all AI use by an expert discoverable as part of the expert’s methodology? If an expert provides prompts to AI with his opinions, reasoning, and supporting evidence and then only uses AI as a typist, would that use be discoverable?³¹ If an expert asks AI to edit a draft report, are the AI’s edits discoverable? These are questions that parties and courts will most likely face going forward.

With these cases in mind, attorneys should be careful to verify the content and citations of their expert witnesses’ declarations and reports. Indeed, courts have suggested that Fed. R. Civ. P. 11(b) “may now require attorneys to ask their witnesses whether they have used AI in drafting their declarations and what they have done to verify any AI-generated content.”³² Attorneys wishing to shield experts’ AI prompts from discovery should consider crafting Rule 29 agreements that specifically exclude AI prompts from the scope of discovery.

The legal treatment of AI assistance in drafting expert reports and declarations continues to evolve. Future cases will likely determine the permissible scope of AI use in forming expert opinions and preparing reports and declarations, as well as the extent to which AI use by an expert is discoverable. Attorneys should continue to monitor the latest developments in this area.

AI was not used in writing this article.