News and Updates

Advanced Artificial Intelligence as a Tool for Reducing Bias in Expert Witness Opinions in Personal Injury Litigation

Background. In personal injury litigation, parties are required to substantiate medical claims through expert opinions. Despite the expectation of professional integrity and objectivity, bias is sometimes observed among experts drafting opinions, including in cases where the court itself appoints the expert. In light of the rapid growth of medical knowledge and ongoing technological developments, the question arises whether emerging artificial intelligence tools can reduce biases of experts when reviewing the medical literature.

Methods. We conducted a proof-of-concept study examining the potential contribution of an advanced AI platform, OpenEvidence — a physician-only system based on a large language model trained on peer-reviewed literature, combined with technology that reduces the risk of “hallucinations” (using a method known as Retrieval-Augmented Generation). The study included two arms: a retrospective arm — examining the tool’s performance on medical information that had been submitted to the court, across 12 court rulings selected as a convenience sample; and a prospective arm — examining the tool’s performance on questions in 13 clinical domains that arose in real time during the consideration of expert opinion preparation.

Results. The tool provided rapid and accurate summaries of the medical literature with references to peer-reviewed academic sources: responses were obtained immediately and more efficiently than manual literature search, and references were verified to be valid and free of hallucinations. In the retrospective arm, in most cases the tool’s conclusion aligned with the factual findings determined by the court based on expert medical opinions, and we identified sensitivity of the tool’s responses to question phrasing. In the prospective arm, dialogue with the tool enabled in-depth exploration through sequential queries, functioning as a kind of simulator for clinical reasoning — for examining questions such as the extent to which the practitioner’s conduct was reasonable (not based on the outcome but rather on literature-based clinical guidelines), and the extent to which, according to current scientific knowledge, a causal link exists between the team’s conduct and an adverse outcome.

Limitations. Given the substantial variability between case facts, causes of action, and court rulings analyzing expert opinions across different judicial instances, it is difficult to draw evidence-based conclusions, but only directions for further inquiry. This is not a systematic empirical study, but rather a preliminary demonstration based on a convenience sample of cases. Therefore, we offer no conclusive findings regarding the tool’s effectiveness or its appropriate role in medical malpractice litigation, but rather describe the potential and limitations emerging from this initial experience and point to directions for future research and implementation.

Conclusions. These preliminary findings indicate the potential of advanced AI tools to assist in rapid and transparent review of the medical literature. Several limitations warrant attention: sensitivity of the response to question phrasing, possible biases in the literature itself, and the question of algorithmic transparency of AI tools (the “black box problem”). We propose extending this observation to a larger and more diverse sample of cases, while examining the response to varied question phrasing and verifying cited literature.
In the meantime, the tool may serve experts in preparing opinions as a decision-support aid in information retrieval — not as a substitute for professional judgment — preferably in a transparent manner, with reference within the opinion to an open link to the tool’s report and accurate citation of question phrasing. Additional proposals include using the tool in inter-party discussions with transparent publication of the dialogue, a controlled pilot for implementing the tool in mediation proceedings, and gradual integration of the tool into legal proceedings subject to the emerging regulatory framework.

Keywords: artificial intelligence, personal injury litigation, medical malpractice, expert witness opinion, evidence-based medicine, bias mitigation, regulation.