In the News: Expert witness used Copilot to cross-check calculations, irking judge¹
Background:
In late 2024, a New York judge criticized an expert witness for using Microsoft Copilot to cross-check a damages estimate in a trust dispute involving a rental property in the Bahamas. The issue at hand revolved around whether the executrix/trustee breached her fiduciary duties to the trust beneficiary by delaying the sale of the property and admittedly using it for personal vacations.
The expert witness was tasked with preparing an analysis of how much would potentially be owed to the beneficiary had the property been sold in 2008 compared to the actual sales price in 2022. The judge found that certain aspects of the expert's testimony were "entirely speculative", and "found his testimony and opinion not credible", specifically calling out the expert for using Microsoft Copilot in cross-checking his calculations.
The expert insisted that his use of Copilot in drafting his report is generally accepted in the field, however, he could not point to any publications or sources to confirm that it was a generally accepted methodology. The expert had used Copilot to cross-check numerical computations and, when the judge attempted to replicate these computations using various queries into Copilot, he received slightly different answers each time, which called into question the reliability and accuracy of Copilot in generating evidence to be relied upon in a court proceeding.
Somewhat amusingly, the judge further experimented with Copilot asking it whether Copilots responses were accurate enough to be cited in Court, to which copilot replied, "my accuracy is only as good as my sources, so for critical matters, it’s always wise to verify." Sage advice from the rapidly evolving LLM.
Main Issues:
AI Use for Expert Witnesses: As the use of AI continues to become more ubiquitous, it is inevitable that an increasing number of experts have or will use it in some form in the process of developing their expert opinions. Right now, there is no bright line rule on the acceptability of the use of AI in drafting expert opinions and reports. At this time, it is incumbent on the expert, and the court which governs the admissibility of the expert's testimony, to make an assessment on whether and to what extent it is acceptable to us AI in forming expert opinions.
Should Expert Witnesses Disclose the Use of AI Prior to Testimony? The judge in this case suggested that "due to the nature of the rapid evolution of artificial intelligence and its inherent reliability issues," any use of AI should be disclosed before testimony or evidence is admitted in court. It raises the question of whether, and to what extent, an expert should consider disclosing the use of AI in developing his or her opinions.
Expert Witness Testimony Admissibility: In Federal Court, Federal Rules of Evidence Rule 702 is still the law of the land. In order for expert witness testimony to be admissible, the proponent must demonstrate that it is more likely than not that:
the expert's scientific, technical, or other specialized knowledge will help the trier of fact to understand the evidence or determine a fact in issue;
the testimony is based on sufficient facts or data;
the testimony is the product of reliable principles and methods, and
the expert's opinion reflects a reliable application of the principles and methods to the facts of the case.
Similar rules apply to state courts and other jurisdictions. How does the use of AI in preparing an expert opinion fit into the framework of these rules?
Practitioner Best Practices:
For forensic accountants providing expert witness services, it is important to understand that AI is an evolving technology that has the potential to assist them in their role as an expert but also comes with inherent limitations at this time. Some of those limitations, including accuracy, reliability, transparency of methodology, and repeatability, are key considerations to evaluate in assessing whether expert testimony is admissible in court. Experts must also realize that due to these limitations, courts will vary in their tolerance and acceptance of the use of AI in forming expert opinions. Until clear rules are laid out, or the technology improves to a point where current limitations are no longer a concern, experts should consider the following:
Recognize AI's usefulness but always verify sources and results: I am not of the opinion that the use of AI is off-limits to an expert witness.
AI can be an incredibly useful research tool, it can assist in combing through mountains of evidence, and it can even be of use in drafting report language. However, an expert should never simply trust the output from AI research without verifying its sources and conducting additional research to confirm its accurate results. Experts can utilize AI in drafting language, but ultimately, the expert report must accurately and completely reflect the expert's opinion, not the opinion of a LLM.
Often, forensic accounting experts rely on the work of other individuals in their firm working at their direction and under their supervision. In that circumstance, it is incumbent on the testifying expert to verify that the work performed by others is accurate and complete, and that the testifying expert fully understands what work was performed and how that work factors into calculations that form their opinion. After all, it is the testifying expert that will need to explain how they formed their opinion during deposition and trial.
I view an expert's use of AI in much the same way. An expert should never testify to something they do not fully understand and hold as their own opinion, whether the development of that opinion was assisted by technology applications or an associate working at their direction.Disclosure Considerations: As discussed above, courts remain appropriately skeptical of the accuracy and reliability of AI. This presents an expert with a dilemma, if the expert chooses to use AI in some way during the expert analysis, should the expert disclose and describe their use of AI?
In my view, if an expert relies substantially on the results of an AI query in forming their opinion, the expert should disclose that query, ensuring that the platform used, the query prompt, and the date of the query are documented in detail and cited in their report. Such disclosure is comparable to citing any other source or reference in an expert report, as required by Federal Rule of Civil Procedure 26.
Having said that, I would caution that experts should not rely on the results of an AI query in forming their opinion in the first place—at least not at this time. Doing so exposes them to the very issues this case cautions against. Rather, AI should be viewed as an initial search tool or a guiding resource. It is incumbent upon the expert to verify any query results against reputable sources and source documents, and to perform their own independent analysis.
If used in this fashion, the expert may feel comfortable not disclosing the use of AI whatsoever. If you can cite a valid source that you found using AI, the valid source is what needs to be disclosed, rather than how you got there. Similarly, experts do not disclose their use of other technology applications, such as Microsoft Excel, Word, and even search engines, which are used during the course of forming their opinions but do not form the opinion itself.Expert Admissibility: Experts should always bear in mind the admissibility standards of the jurisdiction in which they are asked to testify. Using any technology applications, including AI, with a clear understanding of the standards the court will apply can assist in forming a well-reasoned opinion.
Experts are permitted to testify because, as experts, they possess specialized knowledge that will assist the trier of fact in understanding the evidence in the case. AI has never been designated as an expert witness. Experts must base their opinions on sufficient relevant facts and data, understand the sources of that information, and make an informed judgment about its reliability—something that cannot be achieved by relying solely on AI query results. It is the expert's responsibility, not AI's, to prepare an analysis or calculation using reliable principles and methods, and to apply those principles and methods reliably to the case.
Takeaways:
While the use of AI to assist experts in a courtroom setting is still a novel issue, over time, AI is expected to become a more commonplace tool used by experts and attorneys alike. However, no matter how widespread the adoption of AI becomes, it is incumbent upon experts participating in a court proceeding to fully understand the sources, data, principles, and methods used in forming their opinions in order to avoid damaging their credibility or risking exclusion.
Learning Resources:
The Role of Artificial Intelligence in Forensic Accounting and Litigation: An overview of AI's impact on forensic accounting.
¹ Ars Technica, Expert witness used Copilot to cross-check calculations, irking judge, October 2024. Expert witness uses Copilot to cross-check calculations, irking judge
About
The Forensic Accounting Technology & Analytics (FATA) task force serves as a strategic leadership group for the integrated areas of accounting information systems, data analytics, and digital financial forensics. This group facilitates uniquely cross-trained experts into an innovative and strategic thought leadership team, whose primary roles include: stimulating forward-thinking solutions to emerging FATA issues in our global economies; promulgating best practices that are unique to FATA practitioners and environments; identifying and developing FATA resources to enhance learning and professional development opportunities; and educating members and stakeholders to promote FATA’s mission through strategic global relationships.