Georgia Tech study reframes accountability for medical AI harms

Two researchers at Georgia Institute of Technology read 31 US lawsuits over harms tied to AI tools used in medical care, and their conclusion is that the standard story about who is answerable needs redrawing.

In a paper published in Nature Health, Gennie Mansi and Mark Riedl write that medical, legal and technology discussions "often centre physicians' role in preventing and mitigating harms from medical artificial intelligence (AI) tools." Their reading of the cases is that this focus is too narrow. From the abstract: patients' care sits inside a "complex web of stakeholders" that includes state health departments, health insurers and care facilities, "and many AI tools deployed across their healthcare delivery negatively impact their care."

The lawsuits themselves are the evidence base. The authors state that "patients have had no option but to seek legal recourse for harms." Two of the systems the paper draws on for context are the nH Predict system and an Oregon child abuse screening tool the state discontinued, logged in the incident record as "Oregon drops 'unfair' child abuse safety at screening tool."

The design implication is unusual for the medical AI literature. Alongside restructuring liability to reflect who is really in the decision loop, the authors ask technologists to build tools that support "patients' lawyers, provide critical legal expertise and practically support recourse for patients," and to move from a "physician-centred to patient-centred accountability" stance. The specific case-level breakdowns sit in the full paper; the abstract does not publish counts by defendant type or outcome.