Rules for robots, and why medical AI breaks them.

"There is a strong case against extending medical privacy laws to non-medical AI systems that process or produce health-related data. Medical privacy laws allow and, in the case of individual access to data and inferences about oneself require, various disclosures. Those same norms might burden people’s privacy unjustifiably, or deny data processors the fruits of their own inferences, in other data processing contexts where the ends, values, data quality, and social relationships are different. 68: See, eg, Letter from William W. Stead, Chair, Nat’l Comm. Vital & Health Statistics, to Honorable Sylvia M. Burwell, Secretary, U.S. Dep’t of Health & Human Servs. app.A at 15–19 (Nov. 9, 2016), https://www.ncvhs.hhs.gov/wp-content/uploads/2013/12/2016-Ltr-Privacy-Minimum-Necessary-formatted-on-ltrhead-Nov-9-FINAL-w-sig.pdf [https://perma.cc/ J7DF-X9VP] (listing Privacy Rule provisions allowing unconsented disclosure and use of data) (Accessed Jan. 22, 2023). 164: Inst. of Med., Health Data in the Information Age: Use, Disclosure, and Privacy 191 (Molla S. Donaldson & Kathless N. Lohr, eds., 1994) (calling for legislation that protects the confidentiality of identifiable data based on ‘the data elements themselves, regardless of who holds the data.’)."

"FUNDING This work received support under the National Institutes of Health Common Fund’s Bridge2AI ‘Patient-Focused Collaborative Hospital Repository Uniting Standards (CHoRUS) for Equitable AI’ project (OT2OD0327-01, Eric S. Rosenthal, PI), but views expressed are the author’s own and do not necessarily reflect positions of her institution, research collaborators, or funders."