Posted on 07/29/2018 1:13:12 PM PDT by Steely Tom
Last year, studies presented at the American Society of Clinical Oncology's annual meeting showed that IBM Watson was pretty darn good at creating treatment plans for cancer patients. Turns out, however, that the AI is still far from perfect: according to internal documents reviewed by health-oriented news publication Stat, some medical experts working with IBM on its Watson for Oncology system found "multiple examples of unsafe and incorrect treatment recommendations." In one particular case, a 65-year-old man was diagnosed a drug that could lead to "severe or fatal hemorrhage" even though he was already suffering from severe bleeding.
The report puts the blame on the IBM engineers and the Memorial Sloan Kettering (MSK) Cancer Center doctors who helped train the AI. They reportedly fed Watson hypothetical patients' data and treatment recommendations by MSK doctors instead of real patients' information. The approach apparently didn't work as well as they'd hoped, with one Florida Jupiter Hospital doctor telling IBM upon testing the system that the product is "a piece of shit." It's worth noting, however, that MSK believes the example involving the 65-year-old patient was merely part of a system testing and not an actual recommendation.
Despite that Jupiter doctor's less-than-stellar review, a spokesperson told Gizmodo that the hospital still uses Watson's recommendations. Its doctors don't completely rely on the plans it cooks up, though, and see them as an extra opinion when they can't agree on a treatment. As for IBM, it knows that Watson for Oncology still needs work and has taken feedback from clients into consideration to roll out multiple software updates with updated features over the past year.
The company told the publication:
"...we have learned and improved Watson Health based on continuous feedback from clients, new scientific evidence, and new cancers and treatment alternatives. This includes 11 software releases for even better functionality during the past year, including national guidelines for cancers ranging from colon to liver cancer."
The report puts the blame on the IBM engineers and the Memorial Sloan Kettering (MSK) Cancer Center doctors who helped train the AI. They reportedly fed Watson hypothetical patients’ data and treatment recommendations by MSK doctors instead of real patients’ information.
...
HAL: Well, I don’t think there is any question about it. It can only be attributable to human error. This sort of thing has cropped up before, and it has always been due to human error.
Frank: Listen HAL. There has never been any instance at all of a computer error occurring in the 9000 series, has there?
HAL: None whatsoever, Frank. The 9000 series has a perfect operational record.
Frank: Well of course I know all the wonderful achievements of the 9000 series, but, uh, are you certain there has never been any case of even the most insignificant computer error?
HAL: None whatsoever, Frank. Quite honestly, I wouldn’t worry myself about that.
“...a 65-year-old man was diagnosed a drug...”
I stopped reading right there.
My next book may be a biting critique of “modern medicine” (to state nothing of writing skills), assuming someone else hasn’t already done so. Anyone who supports a machine’s decision to “diagnose me a drug” which might kill me better be living behind a 50ft wall.
Insofar as this bs “Watson” goes, something else must’ve prompted them to develop it (government?). If it walks like a duck...
I am totally baffled by a lot of the responses. The third leading cause of death in the United States is Medical Malpractice (doctors making mistakes or simply not caring anymore) at over 300,000 per year.
All Watson is a tool, a computer that hundreds of doctors, universities, etc... are feeding the latest drugs, test results, statistics, case studies, and adjusting the data (correcting errors) based on the results. A vast knowledge base of current medicine without bias, or paid off by drug companies.
I use WebMD to look up medical stuff all the time. It is informative, factual, and helps me make decisions on first aid treatments and my medicine. Imagine this a thousandfold.
Computer aided doctoring, I could see. But I’d never ever take the doctor out of the loop.
Doctors don’t even understand human biology and you would choose to trust a machine programmed by humans because “it’s easy”?
Good luck with that. Seriously.
Then again, you’re questioning critics of a system who are responding to an article which definitively-underscores the danger...I highly-recommend investing more time in your health. WebMD is a tool, not panacea. IBM just learned a very expensive lesson there. I wish I could’ve gotten a 1% commission for the savings had they consulted me, as I would have advised that Watson was a fool’s errand. I state that from a position of having performed thousands of hours researching & writing on the topic.
To-wit, from the OP:
“This story is the technological equivalent of what accountants call a “going concern” letter. It means that IBM and MSK are intellectually bankrupt, without credibility. The language in the “internal documents” is absolutely damning: “multiple examples of unsafe and incorrect treatment recommendations.” Katie, bar the door, there’s going to be a stampede.”
And over 300,000 malpractice death a year is not katie, bar the door? We need a smart, programmable supercomputer to keep up with advancing medical science and to assist doctors. The technology is in it’s infancy, let it grow. If and when they get it right, it will have an amazing affect on humanity.
You don’t get it, but that’s Ok. That’s my experience with nearly 100% of those I speak to about health.
Let me put it this way: What you are describing is akin to everyone consulting with doctors who don’t speak English, doctors who are trained by others who don’t speak their language.
If you don’t grasp that reality, I can’t help you.
Insofar as your repeated malpractice deaths citation, I applaud you for trying to figure it out on your own, but I repeat: Since EVERYONE is different, no tool is panacea.
I’ll give you the preeminent example from my book:
No ‘diet’ is a solution for everyone.
Thus, no computer program can possibly get it right without literally being plugged into your biology.
“Doctors” are not being trained properly because our healthcare system is antiquated. Western medicine will accomplish nothing but pain & suffering for most Americans in conjunction with the popular culture & lifestyle.
Fit & healthy people drop dead every day and that’s not going to change in my lifetime, but hopefully someday I can convince others that more is required of them than just going to a doctor or “logging on” to query a solution...
lol...It was a different time.
I was waiting for someone to clear up the “programming” nonsense as I read down the thread, reading comments from people that know nothing of AI, neural networks, reinforced learning, AlphaZero, LelaChess, etc.
Bottom line is that the progress in Non-Watson AI has been astounding. However, there can be failures, just as there are for humans. It’s just that when AI fails, stories come out that play on fears.
For once I can see the need for some actual freaking Congressional oversight.
AI w/o human input is just as dangerous as humans who ignore/assume when making decisions...have a friend who was being treated for COPD for years and his doctor retired - the new one conducted a bunch of tests and it looks like he was misdiagnosed and the drugs they had him on were making things worse instead of better....the last 6 months have seen marked improvements...all because one human didn’t assume the last one was infallible.
Yeah, when human dermatologists pre-select the images to present to the algorithm, omitting those tumor classes that it's "not trained for."
Not sure I understand your Comment.
Why would human dermos try to falsely improve the accuracy of diagnostic software that threatens human dermo employment?
That's like a professional poker player stacking the deck so he will LOSE money.
In any event, the number one malpractice lawsuit filed against dermos and pathologists is the massive OVER-DIAGNOSIS of skin cancer.
Why would human dermos try to falsely improve the accuracy of diagnostic software that threatens human dermo employment?
Digital pathology companies hire MDs to help them develop their algorithms; they work closely with programmers to improve the accuracy and broaden the range of disease states that can be recognized.
It is to these dermatologists and pathologists I was referring.
However, dermo diagnosis will become the first medical specialty to be universally automated, with a couple hundred human monitors, of course.
There are literally millions of publicly available photographs of skin lesions and slides.
It is only a question of time before a lap top computer program does to dermatology what “Stockfish” and “Alpha Zero” have done to chess.
“Computer aided doctoring, I could see.”
Brings to mind that thread awhile ago about the sex robots.
If it’s okay to play doctor with the robot, why can’t the robots be doctors!?
I agree though - the AI can be just another tool for the doctor to use. This AI stuff is above my pay grade, but I bet 20 years ago they had similar controversies about looking up medical stuff on-line. “Well if it’s not in the XYZ Journal of Medicine I don’t believe it!”
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.