Artificial intelligence is rapidly transforming forensic science, medicine, engineering, finance, and litigation support. As AI systems become capable of producing sophisticated reports, risk assessments, and technical analyses that closely resemble expert opinions, courts are increasingly being asked to determine how such material should be treated as evidence. The central legal question is straightforward but significant: if an AI system produces what is effectively an expert opinion, should it be subjected to the same judicial scrutiny as a human expert witness?
Under the current Federal Rules of Evidence, expert testimony is governed by Rule 702 and the principles established in Daubert v. Merrell Dow Pharmaceuticals, Inc.. Before expert evidence can be presented to a jury, the trial judge must determine that the testimony is based upon reliable methods, sufficient facts or data, and that those methods have been applied properly to the case. This judicial gatekeeping function is designed to ensure that unreliable or speculative expert opinions do not influence the outcome of litigation.
The challenge arises because AI-generated outputs do not neatly fit within the existing evidentiary framework. Since an AI system is not a human witness, it cannot be cross-examined, questioned about its methodology, or challenged in the traditional manner. In some cases, machine-generated reports could potentially be introduced through a lay witness or authenticated as computer-generated records, thereby avoiding the rigorous reliability assessment normally required for expert testimony.
To address this emerging issue, the Judicial Conference Advisory Committee on Evidence Rules has proposed Federal Rule of Evidence 707. The proposed rule would require courts to treat certain AI-generated outputs as expert evidence whenever those outputs would have required expert testimony had they been produced by a human. In practical terms, this would prevent litigants from bypassing Rule 702 simply because the opinion originated from a machine rather than an individual expert.
The Committee's work has been led by David J. Novak, who serves as Chair of the Advisory Committee on Evidence Rules. The drafting and explanatory work has been supported by Daniel J. Capra, the Committee's Reporter and one of the country's leading scholars on the Federal Rules of Evidence. Professor Capra has long been involved in the development of federal evidence doctrine and has played a central role in explaining the proposed Rule 707 framework to judges, practitioners, and academics.
If adopted, Rule 707 would require the party relying upon AI-generated evidence to establish that the underlying system is sufficiently reliable before its conclusions could be admitted at trial. Judges would be expected to evaluate factors such as the design of the AI model, the quality of its training data, validation studies, testing procedures, known error rates, and whether the technology has been properly applied in the specific case. These considerations closely mirror the reliability principles already familiar to courts under the Daubert standard.
The Advisory Committee has also recognised that authenticity presents a separate but equally important challenge. Advances in artificial intelligence have made it increasingly easy to create convincing fabricated images, videos, and audio recordings β commonly known as deepfakes. To address this concern, the Committee is also considering proposed Rule 901(c), which would strengthen the requirements for authenticating digital evidence where questions arise regarding manipulation or fabrication.
For expert witnesses, the proposed Rule 707 serves as an important reminder that technological innovation does not eliminate the need for transparency, sound methodology, and independent professional judgment. Even where sophisticated AI tools are used during an investigation or analysis, courts are likely to expect experts to understand the technology, explain its methodology, identify its limitations, and justify why its conclusions can be relied upon.
Until any new rules are formally adopted, federal courts will continue to evaluate AI-generated evidence under the existing framework of Rules 702 and 901. Nevertheless, the ongoing debate surrounding proposed Rule 707 signals a significant evolution in evidentiary law. As artificial intelligence becomes increasingly integrated into forensic investigations and litigation support, the legal system appears determined to ensure that machine-generated opinions are subjected to reliability standards comparable to those applied to human experts.
βIt cannot be that a proponent can evade the reliability requirements of Rule 702 by offering machine output directly or through a lay witness, where the output would be subject to Rule 702 if rendered as an opinion by a human expert.β β Advisory Committee Note to Proposed Rule 707
By Edward Price
https://www.uscourts.gov/committees/evidence
https://www.uscourts.gov/committees/evidence
