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!
Nah, they never take responsibility for their actions.
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?
Why does someone need to be blamed? Mistakes happen. Nothing is 100% perfect. Not humans, not AI. Can't you see how detrimental your culture's 'OMG gotta sue someone claim money someone is ALWAYS liable' mentality is?
It'll be ok. Zuck, or as autocorrect says, "Suck," will have its own insurance branch before its all over with. His main company is worth more than all the gold in Fort Knox; he'll be fine. All hail Facefarm.
Most medical mistakes that result in death are not caused by misdiagnosis, though that does occur. Most of them are from a combination of surgical mistakes and human error while dispensing/choosing medication. There are a lot of much easier ways to reduce medical mistakes without going so far as to replace doctors with a computer.
Even if the only thing we did was require that every packet of medication was tagged with a barcode and require that appropriate people scan the patient's chart and the barcode prior to dispensing it to verify that it is the medicine prescribed, we'd save a decent number of lives.
And if computers also did checks for contraindications (giving aspirin to someone with a history of bleeding ulcers, giving antibiotics in conjunction with Lipitor, giving any number of drugs with MAOIs, etc.), it would save even more lives.
Notice that neither of these even requires AI. They just require proper electronic medical records and some pretty basic coding skills.
Check out my sci-fi/humor trilogy at PatriotsBooks.
The blame priority falls from deepest to shallowest pockets. If ol' Zuck can be implicated, trust me, he will.
had to say it
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.
Whoever thought the AI was accurate enough is to blame, that is, the person who approved its use for that patient and then signed off on the diagnosis. I mean, people are verifying this stuff right? It's not like the AI has any authority itself.
Twinstiq, game news
...sure who's to blame because I don't have a tattoo.
"I'm a dirty white tomcat, enter my world..."
until I can get cheaper Medical Service by excluding a doctor, I feel the doctor is liable. If I can choose a discount and use AI with published results and make my own choice, that's on me.
Wow, sent an e-mail as suggested when clicking on "use classic" banner, and got a fast response that addressed my msg
the operator may sue the manufacturer, but you should sue whoever you bought the service from(doctor/hospital).
look, it's pretty much the same already now if you go get your eyes lasered and messed around with - the machine does 100% of the actual operation and the doctor is there just to press stop. but it's still his fault if something is fucked up.
world was created 5 seconds before this post as it is.
Wasn't there a scene in a movie about this, where poorly-behaved robots were tortured with hot irons on the feet and similar? I mean, if the AI messes up, then I can only assume the AI is to blame; don't we just slow down its power cycles or something like that?
The AI vendor and/or user could be required to carry liability insurance. The better the AI (fewer catastrophic errors), the lower the premium.
To narrow it down, where blame can not be placed will be found in small print on something you signed or otherwise agreed to.
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 we could all just stop looking for the whipping-boy every time something goes wrong, that would be great.
You are being ripped off every second of every day, so that advertisers can help rip you off even more tomorrow.
Who has the medical license, the AI or the Dr. using it ? You don't sue the gun manufacturer, or the stethoscope company, they are just tools used by the licensed to 'practice' medicine doctor.
errr....umm...*whooosh* *whoosh* Is this thing on ?
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.
But who can get you unblocked listed form the pre-existing conditions list?
With millions of samples undertaken every day, billions over the span of a year we can postulate that even an error rate of 0.1% will add up to a huge civil action when the root causation is discovered!
Christ, you watch porn in 360p? Is this 1994?
Who else?
I don't know, but it's only a misdemeanor to beat the metallic shit of them.
Table-ized A.I.
The IT guy's malpractice insurance naturally.
You blame the doctor. Then you lead the priestesses of Kubebe on a jihad against Richese and then on to destroy the thinking machines.
Oh, and a botched diagnosis should also go in the pool of training cases for future AIs.
First, remove the AI and have an actual human do it. Human intuition will pick up things that even IBM's betrayal of humanity won't.
Twitter supports and protects racists - by smearing their critics with the "Hate Speech" label.
The medical licensing authority that licensed a machine to practice medicine.
Seven puppies were harmed during the making of this post.
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
Christ, you watch porn in 360p? Is this 1994?
When you're cranking one off in the company toilets watching Adult Entertainment on your phone, you really don't need high definition. Headphones are a good idea though. So I've been told.
To have a right to do a thing is not at all the same as to be right in doing it
A real world exemple : electo-kardio-grams (traces of the heart activity).
They are a useful tool to help daignose heart rythm problems.
Since a couple of decade already, given how simple the data is (less than a dozen of 1D signals), we already have managed to do automatic recognition.
To the point that any modern EKG will give you a diagnostic printed after the traces them self on the report.
How do doctors use it ?
We are trained to first look at the traces, see if they seem obviously wrong or not,
then apply a couple of heuristics that we have learned (QT delay, signal elevation, etc) to check for everything,
and then read the automatic diagnostic.
So if we missed something, the automated diagnostic can help point us in the right direction (e.g.: if the pathology isn't that obvious).
But if we see something deeply wrong, we are not going to ignore it simply because the machine said "everything looks OK to me".
There's always going to be a doctor in the loop reviewing what the machines say, at least in the foreseeable future(*).
In a way you can think of the AI as a not yet fully trained early-year student: can give useful information, can offload some work to it. But never trust it 100% without a review.
---
(*): at least for as long as the current AI (based on deep neural nets) are only as good as training pigeons to guide WW2 bombs.
Perhaps if one day in the far distant future we manage to make much more clever AIs
(e.g.: a high number of various DNN, all interlinked together, the same way as a biological brain has several cortical regions, from primary (visual cortex), through some more associative (faces recognition), to highly associative (interlinking all the rest)).
Then it will be a question that boils to if this AI can manage to "pass exams".
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
The law tends to protect doctors from simple mistakes. And with AI I think the same would be true. If the software was diligently created and is known as a good product there is no expectation of perfection. The same is true for your surgeon. He can do great harm. But as long as he was sober, in a proper state of mind, and diligent in trying to render aid the law will not tend to land on him like a bag of bricks. Did the doctor or software do what other doctors or AI programs would have done? Is the bad outcome simply a matter of unusual conditions that do not diagnose easily? Juries know full well that a big settlement means their own individual health care prices will rise. A jury needs to feel rage in order to give a high award. If that doctor was in the strip club until 4 AM and drank a huge bottle of champaign and staggered into the operating room with no sleep and still half drunk then a jury very well may cut the string and let the red balloon fly.
Wrong forum. Go here: https://www.reddit.com/r/tipof...
Dear Slashdot: next time you want to mess with the site, add a rich-text editor for comments.
Have gnu, will travel.
Obama. The answer is Obama. Thanks again, Obama!!
Sigh. This is just more people looking to identify upon whom they can exact revenge if an error occurs.
Look, people, you're acting like psychos /and/ driving up the cost of healthcare at the same time, while slowing progress.
The AI's should be pre-certified by a standards body for being written as carefully as possible. If they are competently certified and you're in the false-positive or false-negative zones, that's just a natural consequence of participating in the medical care system.
Healthcare will /never/ be perfect. There will always be errors; information theory guarantees it. If the expected outcome of such an error is to seek vengeance, all that will do is to forestall the very innovation that will minimize those very errors. Do not participate in the system if you cannot value the risk rate in the system over the risk rate out of the system.
Some might wonder, "does it seem just to seek vengeance on the people who are causing that delay with their vengeance?" Everybody would do well to remember: âoeIf we do an eye for an eye and a tooth for a tooth, we will be a blind and toothless nation.â - MLKJr.
My God, it's Full of Source!
OUTSIDE_IP=$(dig +short my.ip @outsideip.net)
Shit happens. If you want to blame someone, blame God. You aren't entitled to compensation every time something bad happens.
You have to hold a license in order to practice medicine.
Same as any other professional license, except crappy doctors usually kill their victims one-by-one; crappy civil engineers can kill orders of magnitude more people in one disaster. This ain't anything new, folks.
You have certain ethical responsibilities when you get your MD, DO, DDS, etc--it's spelled out and regulated (in the us) by state.
Yes, it's really the state medical board they get their authority by statute.
It doesn't matter *what* diagnostic tools you're using--if you're practicing medicine, you're the gatekeeper.
The FDA will tell you what's approved or not approved--but they do not regulate *everything*.
Your employer makes you use some cheapass screening tool to grind a hundred patients through diagnosis where you'd only get ten of them through if you did it right. Fine--they have profit to think of.
It's your responsibility as the practicioner to say "No, this is bad medicine and I won't do it."
It's really a no-brainer for the practicioner, too--you're going to lose your license and get sued if you go along with this kind of crap.
"Reality is that which, when you stop believing in it, doesn't go away." - Philip K. Dick
Bayesian Nets, can incorporate expert knowledge, and they allow for parameter and structure machine learning.
The have been extensively used for diagnostics, and they give a valid statistics for the most probable diagnostic, based on the data that they have been trained with.
This kind of AI is computationally and conceptually somewhat more expensive, but there is really no valid reason why medical diagnostic systems shouldn't follow this paradigm. Especially since recent research shows that these networks are also surprisingly robust with regards to their parameter settings.
Some decisions are too important to just entrust to an ANN black box. In Bioinformatics it amounts to programming malpractice.
https://radiotherapydictionary...
First and foremost this is NOT Artificial Intelligence. It's machine learning and neural networks. Those are very different things. It's just that calling it machine learning isn't as attention grabbing and fear mongering as implying that these are intelligent systems.
The article does briefly correlate the two when stating that the opacity of machine learning makes the results almost impossible to trace back. That's a big consideration. I don't feel like looking up the article where a major hospital used their entire medical record database to feed the neural network and found that the system was significantly better at diagnosing schizophrenia. The problem is they don't know how the system does this, what indicators are being used, etc. But they've used this information to assist doctors in treating patients.
The way to go at the moment is to let these systems run and learn but give the results to real, trained doctors for diagnosis and treatment. Consider them another screening test. When the doctors have entered their diagnosis and treatment, plus how well the patient responded, the systems will learn a bit more. It's going to take a lot of years and a lot of medical records for these to become even marginally trusted. That doesn't mean they can't be fired up and start crunching the data now.