This week in The Atlantic, Ezekiel Emanuel and Vinod Khosla argue that medicine should let AI practice autonomously: not assist, practice. The AMA's position is that a physician must always have the final say. Both sides are arguing about outcomes. I read the fight as a lawyer, so I hear a different question underneath it. When the clinician is a model, who carries the standard of care?
Malpractice doctrine is built on two humans. The physician-patient relationship creates the duty, the breach is measured against the reasonably prudent physician in the same circumstances, and customary practice is the evidence everyone fights over. Every piece of that assumes a licensee whose judgment is the thing being judged. The model has no license. So the doctrine reaches for the people around it.
The assisted era fits the old frame cleanly, which is why it hasn't produced interesting law yet. The doctor consults the model, the doctor decides, the doctor answers. The model is an instrumentality, no different from the lab assay or the imaging software, and "the AI suggested it" is no more a defense than "the calculator suggested it." If anything, the exposure runs the other way, and this is the part the medical societies are missing.
In 1932 a barge sank in a storm that the tugs towing it never heard coming, because they carried no radio receivers. The industry hadn't generally adopted radios yet, and the tugs said so. Judge Learned Hand was unimpressed:
Indeed in most cases reasonable prudence is in fact common prudence; but strictly it is never its measure; a whole calling may have unduly lagged in the adoption of new and available devices. It never may set its own tests, however persuasive be its usages. Courts must in the end say what is required; there are precautions so imperative that even their universal disregard will not excuse their omission.
The T.J. Hooper, 60 F.2d 737, 740 (2d Cir. 1932)
Custom is the floor of the evidence, not the ceiling of the duty. Now hold that against the strongest fact in the Atlantic piece: a 52-study review in which hybrid doctor-plus-AI arrangements beat neither AI alone nor the best clinicians. If autonomous models keep outperforming, the day arrives when the reasonably prudent physician consults the model, and the physician who overrides it on instinct becomes the negligent one. The standard of care does not wait for the AMA's permission. It never has. That is the quiet bomb inside the "doctor always decides" position: it may already be the substandard one.
The autonomous era strains the frame in a different place, because somebody still has to be the defendant. Three candidates. The manufacturer, in product liability, except the learned intermediary doctrine was built around a physician in the middle, and autonomy removes the intermediary. The configurer, the people who set the parameters and protocols the model executes, because the judgment was theirs even if the execution wasn't. And the deployer, the hospital that puts the model in front of patients.
Courts will find the deployer first, because they already found it for humans. In Sword v. NKC Hospitals, Indiana's supreme court held a hospital vicariously liable for the negligence of an independent-contractor anesthesiologist under apparent agency: the patient doesn't audit the employment structure, the patient sees the hospital's name and relies on it. An autonomous model wearing the hospital's logo is the cleanest apparent agent ever pleaded. Respondeat superior was built for exactly this arrangement: the principal who puts the actor in front of the world answers for the act.
The statute books already drew the assisted-versus-autonomous line once, for regulation rather than liability. The 21st Century Cures Act carved clinical decision support out of the medical device definition only so long as a clinician can independently review the basis for the recommendation:
...so that it is not the intent that the HCP rely primarily on any of such recommendations to make a clinical diagnosis or treatment decision regarding an individual patient.
FD&C Act § 520(o)(1)(E), as added by the 21st Century Cures Act (2016)
Remove the clinician and the same software is a medical device: FDA-regulated, product-liability territory. Congress saw the line clearly enough to legislate it. The courts will draw the liability version of the same line, case by case, starting the first time an autonomous pilot goes wrong in front of a jury.
Insurers will draw it before the courts do. Malpractice carriers are already pricing the question: who was supervising, what could they see, what were they allowed to override, and where is the log. If you're a hospital standing up an autonomous pilot, the standard of care you'll be judged against is being written right now, in your protocols and your audit trails. Write it down like you'll be asked for it.
Emanuel and Khosla say let the robots try being doctors. Fine. The law's habit, centuries old, is to hand the license's weight to whoever put the robot in the room.