Physician’s Weekly (PW) spoke with regular contributor Dr. MedLaw, a board-certified radiologist and medical malpractice attorney, about AI and physician liability.
PW: Should doctors be concerned that the use of AI in the clinical setting can affect malpractice liability?
Dr. MedLaw: There are two considerations: notes and clinical decision-making.
AI is moving past mere transcription to large language models. These models decide what’s important and then create full notes that can be popped into an EMR. The accuracy of these systems has not been well established yet, but even if AI turns out to be accurate, the doctor’s duty is unchanged. The doctor who distributes a note under their own name is ful – ly liable for its content, and you can’t delegate that to an AI system any more than you could to a human scribe.
The problem is when the doctor only uses the AI system because they’re overburdened with paperwork. Clinicians must stay alert, as if they were the one creating the note.
What if an AI system makes a finding and the doctor disagrees?
The standard of care is best understood as “good and accepted medical practice.” It is tied to the legal standard that being nonnegligent means acting clinically reasonably under the actual circumstances. AI doesn’t change either of those factors. In fact, to be clinically useful, the AI should have been trained with the standard of care.
Let’s break down why a doctor may not follow the AI. If the reason is negligence, and the doctor never even looked at the AI findings, then settle the case and write the check.
However, if the doctor was aware of the AI finding but didn’t follow it, then the doctor’s defense must be to show why the recommendation was not applicable to the patient’s care at that time. AI must be ex – plained as a factor like textbooks, journals, and guidelines. If the doctor can show why it was more reasonable to not follow the AI, they have their defense.
Specifically mention what the AI said and why you acted in accordance with or count – er to it. The trier of fact should be able to ask about your process and see that you paid attention to what the AI said and then determined its appropriateness.