醫生使用人工智慧時的醫療事故責任

Medical malpractice claims in the US usually require four elements: a duty to the patient, a breach of the standard of care, causation linking the breach to the injury, and damages. The standard of care is not written in one place. It is typically established through expert testimony, professional guidelines and common practice, and it changes as medicine changes. In a widely cited 2019 JAMA analysis, W. Nicholson Price, Sara Gerke and I. Glenn Cohen laid out how current law treats AI advice. When an AI recommends standard care and the doctor follows it, the doctor is generally protected. The riskiest case is following an AI recommendation that departs from standard care when the patient is then harmed. Rejecting a correct but nonstandard AI suggestion is usually defensible today, because the doctor stayed within accepted practice. The authors noted that this pushes clinicians to use AI mainly to confirm what they would do anyway. As AI tools become part of accepted practice, ignoring or failing to use them could itself start to look like a breach. Other parties can be liable too. Hospitals can be vicariously responsible for employees and can face corporate negligence claims over how they select, validate and train staff on tools. Developers may face product liability claims, but those are harder to win. Courts have often treated software as a service rather than a product, and the learned intermediary idea holds that the clinician stands between the tool and the patient. Two misconceptions are common. The first is that FDA clearance protects the doctor. Clearance through the 510(k) pathway generally does not block state negligence claims, and it doesn't define the standard of care. The second is that the vendor absorbs the risk. Contracts often shift liability back to the health system. Few court decisions address clinical AI directly, and outcomes vary by state and country.