Slashdot Mirror
When AI Botches Your Medical Diagnosis, Who's To Blame? (qz.com)
Robert Hart has posed an interested question in his report on Quartz: When artificial intelligence botches your medical diagnosis, who's to blame? Do you blame the AI, designer or organization? It's just one of many questions popping up and starting to be seriously pondered by experts as artificial intelligence and automation continue to become more entwined into our daily lives. From the report: The prospect of being diagnosed by an AI might feel foreign and impersonal at first, but what if you were told that a robot physician was more likely to give you a correct diagnosis? Medical error is currently the third leading cause of death in the U.S., and as many as one in six patients in the British NHS receive incorrect diagnoses. With statistics like these, it's unsurprising that researchers at Johns Hopkins University believe diagnostic errors to be "the next frontier for patient safety." Of course, there are downsides. AI raises profound questions regarding medical responsibility. Usually when something goes wrong, it is a fairly straightforward matter to determine blame. A misdiagnosis, for instance, would likely be the responsibility of the presiding physician. A faulty machine or medical device that harms a patient would likely see the manufacturer or operator held to account. What would this mean for an AI?
So the computer produces a list of possible diagnosis. This list I understand is called a "differential diagnosis" and may have as few as 2 or 3 items or as many as several hundred.
I would expect that this diagnosis list, as well as a management plan, to be then put in the hands of a human. After a series of tests I would expect the AI to be consulted again if necessary.
In today's world and probably the near future, say the next decade, I doubt that medicine will become "autodoc" "robotic physician" the holographic "Doctor" or some "magic cryokit" There will be a human with a powerful tool to aid in diagnosis of the patient.
Now, what will happen in 50 years, that is to be seen.
Phil
Laugh, it's good for you!
They're not even a real country anyway....
But to be more serious, this is going to become a serious problem soon. Whether it's cars or medical diagnosis or some other AI application. I think promoters of AI underestimate just how outraged the public will be when someone is "killed by an evil robot." Human error we can understand and sometimes condemn. But I think there's the potential for a lot more backlash even for minor incidents with AI -- and even if they likely wouldn't have been preventable by a human doctor/driver/whatever. At that point, it won't matter that the stats say it actually saves more lives overall, if the error or the death is egregious enough.
A misdiagnosis by a human physician can only be analyzed and argued about. A misdiagnosis by an AI physician can be forensically investigated. It can even be perfectly reenacted, both with the same and different inputs. That would allow, for example, a determination of whether the fault was a design flaw or a problem with the supplied inputs.
This would allow for very precise determination of responsibility. Today, if the patient omits some medically relevant detail and a misdiagnosis occurs, the human physician can only argue that he could have possibly come to a different conclusion with the additional information. With an AI, we can feed the updated parameters to it and actually see whether the result would or would not have been different. If the result would have in fact been different and correct, then the fault lies with the patient, or possible whomever was responsible for collecting the input data. If the result would not have changed, then there is a possible design flaw for which the developer/manufacturer may be held liable.
In my mind, this can only mean an improvement from where things currently stand.
The one with the deepest pocket is the one to blame. If others have any resources the ambulance chasers will go after them too.
sed -e 's/Chuck Norris/Rajnikant/g' joke > fact
Obviously whoever hired him is to blame.
After seeing his performance in "Like a Surgeon" I'm surprised anyone would hire him to diagnose anything.
"Nine times out of ten, starting a fire is not the best way to solve the problem." - my wife
I'm sorry, I just don't believe that medical error is the third greatest cause of death.That's just stupid. Nobody in his right mind would ever go to a doctor if the odds were that high. Does anybody ever question the stats people toss around these days?
Is it too early to be disappointed in Slashdot again? Maybe someone will post a funny joke that actually gets some funny mod points? Ditto insightful, eh? Even an actually interesting or informative comment? Not holding my breath. Short summary: No such luck yet (and including keyword searches).
Ever hear the old philosophy joke: "If a tree falls in the forest and there is no one to hear it, does it still make a sound?"
The equivalent question for today's feeble article is: "If a corporation's AI botches your diagnosis and there is no one to sue, does your death still matter?"
You may safely anticipate that the EULA will protect the AI from liability much more than it protects the patient from mistakes or software glitches, no matter how egregious and flagrant. Actually, the hierarchy of protection will probably go something like (1) Corporation that created the AI, (2) Corporation that is licensing the AI, (3) The hospital corporation, (4) The doctors who use the AI, and so on. They may remember to include the patient somewhere in there, or maybe not.
Compare to Dr Mayo's motto: "The best interest of the patient is the only interest to be considered..." The current incorporated Mayo Clinic still mentions patients on the website, but I couldn't find such a strong form.
Not sure what the trigger was, but I recently realized that individuals don't count now. It's only the biggest corporations and political parties that matter. I had still been clinging to some delusions from my "respect for the individual" days, but now the individual is just a cog, and the only question is which cog can do the job most cheaply before being discarded. Trigger might have been the book Hitlerland , which has NO relation to #PresidentTweety, since it was published some years ago. Not even sure if I want to recommend it, though it's still bothering me...
Freedom = (Meaningful - Coerced) Choice != (Speech | Beer^2), and sad sock puppets' bad mods avail them naught.
It's a stupid question that illustrates a misunderstanding about what diagnosis is.
If a fortune teller fails to predict you're going to get hit by a bus tomorrow, who's to blame, her, her crystal ball or it's manufacturer?
Physicians misdiagnose patients all the time because diagnosis depends on a variety of imperfect information and very often cannot be done accurately. That is nobody's fault.
Physicians also misdiagnose patients all the time because they aren't very good at keeping up with new developments in medicine or are otherwise negligent. That's their fault.
An AI could be wrong for the first reason. If so, nobody is at fault. If the AI is wrong because of a manufacturing defect, the manufacturer is at fault, or the supervising physician is, if they insist on being in that position.
The doctor. All of these systems are marketed and sold with the proviso that these provide only advice for the physician. This is to make sure that liability is clearly allocated. And notice that it's not the software company accepting this liability.
That is all.
Nobody in his right mind would ever go to a doctor if the odds were that high.
That depends on the nature of the "medical error". I expect that the vast majority of these cases are people who have a serious condition which is misdiagnosed or incorrectly treated and because the condition is not treated it eventually kills them i.e. they die from the condition. If true then going to the doctor results in a 67% chance of proper treatment and possible survival vs. no going which will be 100% fatal.
To be worse off going to the doctor there has to be a serious risk that the treatment of a minor condition is so seriously botched that it kills you when the condition itself would not. This seems far less likely.
If only there was some organization that provides a safety net in case something like this happens. In exchange for a small fee (whether through taxes or however else it's implemented), patients get economically protected should a misdiagnosis causes problems.
Maybe the safety net could be called "insurance". Perhaps it may be possible to do a for-profit organization under this concept.
If the AI turns out to be more accurate, then that would make insurance payouts less frequent.
If you want to minimize problems, you can have the AI provide the most likely issues, and a competent human doctor make sure that the diagnosis is sane.
Christ, you watch porn in 360p? Is this 1994?
The IT guy's malpractice insurance naturally.
This type of question is a non-existant dilemma.
The concept of blame should disappear if you are diagnosed by a system that is orders of magnitude more precise than any human could ever be. AFAIK that is exactly the point of medical AIs like Watson. If maintained well, a system like Watson can "know" things a human or an entire army of human medical experts could never know, can process cross-reference cases and drug interference and genetic information at a speed, scale and accuracy that will make the last 500 years of advancement in medicine look like a pre-school exercise in comparsion. Miss- or non-diagnosis by human medical experts is high, and we wouldn't be happier about it if we have someone to blame. Doctors can only operate because there are catch-alls in place that keep a doctor who screwed up from going to jail. Given the 80% chance of dying in the next 5 years or the 80% chance of being cured with an 20% chance of an operation done by a human still failing and killing me really fast I probably would still take my chances. It's always a trade-off and capable AIs driving for us or doing 95% of all medical diagnose work will tip the odds so far in favour of humans, playing the blame game if something at some point does go wrong would be nothing short of stupid and/or silly.
The same goes for "Whos the AI driven car going to kill? The the young handicapped kid on life support or the old grandma 5 years away from the grave but with 4 grandchildren who love her?".
This type of question entirely misses the point. AI will be let on to the streets when they drive way, way better than a human ever could, always and everywhere. Deaths in traffic will plummet by orders of magnitude and the occasional situation where an AI can't prevent someone from dying will be so rare society will shrug it off. Experts even expect an extreme organ donor shortage once AI hits the streets. Less idiots killing themselves and others. ... On second though, maybe we should keep a subset of roads for those who insist on racing around.
We suffer more in our imagination than in reality. - Seneca