The ethics and governance of large language models in dentistry: A framework for research and clinical implementation

As large language models (LLMs) are increasingly utilized in dentistry, they introduce various ethical challenges, including patient privacy, data security, and liability concerns. This study aims to establish a framework for their ethical evaluation, facilitating responsible development and research of LLMs and their implementation in dental practice.

Based on a review of existing ethical guidance on LLMs, including World Health Organization (WHO) resources and publications on ethics in healthcare, an expert panel developed a checklist containing nine relevant topics. The Global Initiative on AI for Health engaged 105 participants in an anonymous online consensus process (e-Delphi), rating the importance of each item on a 10-point Likert scale.

The topics included: 1) Data Protection and Privacy, 2) Data Provenance and Copyright, 3) Gender Bias, Diversity and Discrimination, 4) Fairness and Equity, 5) Transparency, 6) Explainability, 7) Autonomy, Responsibility and Consent, 8) Prudence and Sustainable Development, and 9) Generative Empathy. Over 70% of participants rated all items between 7-10, highlighting their importance. Data Provenance and Copyright received the highest rating (10 points) from 57 participants, followed by Transparency and Gender Bias, Diversity and Discrimination (with the highest rating from 52 participants each). Generative Empathy showed divergent opinions, with 38 ratings below 7 and 31 scoring it a 10.

The e-Delphi study achieved consensus across all nine topics, which were included in the proposed ethical framework for LLMs in dentistry. Strong agreement on most items demonstrated the framework's theoretical soundness, while greater variability in Generative Empathy responses indicated this principle requires further considerations.

This framework provides systematic guidance for responsible LLM implementation in dentistry, addressing critical concerns such as data privacy, intellectual property, bias, and transparency. The framework offers specific strategies for AI developers and researchers during development phases, while providing governance guidelines for practitioners, administrators, and regulators during clinical implementation.