Natalie Reed

Helios Scholar

Artificial intelligence (AI), the ability for machines to perform tasks by simulating human intelligence, holds promise for advancing translational research. However, the application of this technology is hindered by regulations that fail to address pressing concerns over appropriate data usage. Therefore, there is an unmet need for TGen to identify and mitigate these risks so that a productive approach to AI research can be established. To address this need, a review of pertinent U.S. regulations and scientific literature was conducted to determine the legal, ethical, and social implications of training AI models with research data. The Common Rule, which establishes the regulatory framework for the use of human subject data in research, inadequately addresses AI-specific risks as it was not designed with AI in mind. Such risks include diminishing the autonomy of data donors, as training AI with secondary, de-identified data is exempt from informed consent and may diverge from initial consent expectations. Additionally, the standard for ensuring privacy during secondary research relies on an outdated concept of de-identification that neither adequately protects data privacy nor accounts for vulnerabilities in AI model security. Another concern is that biased AI might exacerbate health disparities in translational research by compromising accuracy and treatment efficacy in populations that are underrepresented in the training data. Moving forward, TGen must address these challenges through proactive policy development that considers the transparency of data use during consent, the limitations and applicability of privacy-protecting techniques, and the need for rigorous bias mitigation strategies throughout AI development.