Trust me, I'm a computer: Watson takes on health care challenge IBM’s game show-topping Watson system is starting its first real job early next year. Fresh from its triumphant drubbing of the human race on US quiz show Jeopardy, Watson is now being outfitted to help us slack-jawed dim-witted humans live better and longer. WellPoint, a large US health insurer, will deploy a shiny new Watson …
... the time spent considering possible ailments and possible treatments can be slashed even further.
At least Watson's "professional judgement" won't be swayed by the latest shiny tchotchkes the pharmaceutical sales flaks toss across the desk ... right?
"...."research" underwrit by Big Pharma ..."
So those punting product plough a load of cash back into R&D to generate better product.
This is a bad thing how exactly? What did you have in mind? The pharmaceutical industry gets to keep all the cash and taxpayers (under)fund all research? Yeah....right.....
Big tax on pharmaceutical products spent on research^H^H^H^H^H^H^H^Hwhatever's a vote-winner this week? Yeah......right again.....
Nice shiny hat. Suits you sir......
... is that they can all be wrong, even when they contradict.
Certainly there are problems with the influence of profit-seeking corporations on medical research, and much of that influence comes directly through funding and directing research.
However, the funding environment for research in medicine and cognate fields (various areas in biology, chemistry, etc) is enormously complicated. Eliminating corporate funding, or ignoring its results, is not the answer.
There are about a zillion studies done from all sorts of perspectives (economic, sociological, etc) on how research funding happens and its effects on that research and its reception. None of the reputable ones I've seen have concluded that "all research by pharmaceutical companies is junk", or conversely "is trustworthy". They all - mirabile dictu - paint a more complex picture.
i cant help but think that watson will end up being the doctor's equivalent of 1st line tech support...
The assumption is that the doc is competent, and not just blindly following what the computer says to do.
The flip side of that is if watson *isnt* followed because the doc thinks he knows better, and then turns out to be wrong...
You cant exactly send the engineer round to to a re-install and swap the mobo...
Watson still got many answers wrong.
What will happen when those are not answers in a silly quiz show but very real treatment courses for people?
I can see many overworked (or just lazy) doctors just going by "Watson" without the right kind of critical thinking.. Not to mention that for an insurer those answers will probably be optimised just to make them more money, not exactly better care.
Having said that it may well turn out better than a single doctor of little experience, however if it goes wrong what can you actually hold responsible?
"Watson still got many answers wrong."
True, but so do humans. The correct way of using it would be to use it as a source of advice rather than a way to get a definitive answer. Having said that, the way a lot of people blindly follow satellite navigation is probably an indication that your concern is at least partially valid.
I'm more concerned about the bedside manner: "What is ... you have 2 months, 3 weeks, 5 days and 7 hours to live?"
So, what's the probability that a given doctor will apply "the right kind of critical thinking" when relying on a fairly random history of medical school, anecdotal experience, exposure to a small proportion of the available research, ad hoc conversations with colleagues, intuition, prejudice, etc - but not when they add some Watson queries to that mix?
Essentially you're claiming that there's a good chance doctors will treat Watson as an oracle. I suspect that most doctors don't finish their training and get their license without acquiring a bit of habitual skepticism.
Diagnosticians already search medical databases. They also routinely reexamine, question, and ignore results from diagnostic equipment, when those results strike them as implausible. (And they should, just as the etymology of "diagnostic" suggests.) Watson is essentially just a better query interface for those medical databases. It doesn't magically change anything, and the only doctors who are likely to believe it does are the same bad apples that end up advocating snake-oil treatments or publishing bad research. That latter group is already a problem; Watson doesn't make it a worse one.
“There is an 80 per cent chance it’s this, a 40 per cent chance it’s that, but also a ten per cent chance it’s this one.”
And a 130% chance I won't trust a system that was so poorly implemented it don't get basic probability correct.
----
Seriously, this *could* be a good way to point a doctor in a set of directions to research. Just don't rely upon it, or your treatment could end up like an episode of House. At least Watson don't do Vicodin.
Those percentages should not add up to 100% - it would be far more suspicious if they did.
We aren't throwing dice here with a specific and known *complete* set of results that are dependent on each other.
These are independent possibilities, thus independent probabilities.
Also, a patient could have more than one ailment - some of the symptoms could be irrelevant.
You're right though - this isn't intended as a catch-all "diagnosis", it's intended to point out useful avenues of enquiry.
It's very helpful to have the list of most likely possibilities - then you can tailor the tests to distinguish between X, Y, Z and "something else" rather than running every test you can think of.
For example, when it says "80% chance it's X", you can do a quick and easy test for X and then maybe re-run Watson with "Given that it's not X, what is it?".
There's an interesting take on a `Medical diagnostic computer in the Michael Crichton film, `The Andromeda Strain'. This was a movie that placed scientific concepts right at the center of the plot and didn't insult the audience. It's incredulus that Crichton is also responciple for Jurasic Park, unless Spielberg dumbed it down for the younger audience. All we have for a scientific concept is Jeff Goldbulm waffling on about 'strange atractors'. Yes I know the effects were good, but that's why we go and watch movies featuring dinosaurs, to see them eating people .. :)
http://www.imdb.com/title/tt0066769/
http://www.imdb.com/title/tt0107290/
This is definitely a very, very good thing. This will almost undoubtedly save lives. For now, we have to have an extraordinarily expensive device to perform this function - combine this idea with those portable medical scanners a few companies are working on, and assuming Moore's Law continues to chug along for a while yet, we will have tricorders by 2030.
IBM, you uptight, formal bastard - thank you.
Well, this is a fairly important point. So far it's a tool. But no matter how useful the tool, the crux is in how we use it. Wouldn't be the first time someone mis-uses, mis-interprets, or otherwise wrongly employs the tool. And especially with big large spendy shiny magic boxes like this, it's all too easy for some dimwit pen-pusher to make use of the tool mandatory and then make its advice authoritative.
Sometimes people really do know better and it's impossible to say when that'll be the case. It's also a question of who's responsible. If it's the doctor, then it's his call and it doesn't do to take away from him the very things that he needs to properly do his job. Those are, perhaps contrarily to intuition, not necessarily the best information tools available, like this one. It's the ability to make the call. However much informed he is just gives us better quality decisions, hopefully. But even high-quality algorithms using high-quality data are only just that. It's the human that shoulders the responsibility, and with it must come the tools to shoulder it. The tools cannot be allowed to overrule whoever is responsible.
As long as that is adhered to, perhaps religiously, I'm reasonably fine with it, on this angle at least. Another thing to watch out for is becoming dependent on the tool, killing people if it's somehow somewhen not available. I do expect my doctor to be competent without it, even if I do appreciate better quality decisions when available. That, too, is becoming a harder sell as more information is becoming available for the search-engine-using patient. Perhaps somewhat paradoxically so.
Meaning that now that we have better tools we have more to watch out for, too. This is not a bad thing unless we fail to do so. I'd say it's worth looking into putting this sort of thing into modern medico classes, for example.
If the insurers want this, it'll happen in a few years, at least in the States. If they don't, it'll never happen in a million. I guess the key will be the quality -- for a court -- of the "explanation" that the system can give for its choices. That has been the great failing of blackboard expert systems like Watson in the past.
There will still be plenty of room for medical incompetence though, and Watsonesque systems will increase the percentage of incompetence that has its arse well covered: "I followed the 80% recommendation."
Patient: No its not appendicitis doctor
Doctor: That's what the machine reckons and having physically examined you I agree
Patient: No that's not right. I was on wikipedia last night and that Jimmy Wales died of a mutating hysterectomy and I've got the same symptoms
Doctor: No such thing as a mutuating hysterectomy
Patient: There is its on the web, your computer will know about it 'cos it searches the web to find the matching symptoms
Doctor: That's not how it works
Patient: Yes it is, I was reading Answers on yahoo and they said that's how it works
Doctor: Do you want a second opinion
Patient: Yes please can you ask the guy who does Supersize v Super Skinny please, 'cos I trust him
I remember when I studied AI as part of my CS course at university, we had a few lectures discussing medical expert systems. The statistics were interesting both in terms of how much accuracy was increased by the use of expert systems and how bad doctors did in general. These were very much area-specific applications but if a system like Watson can offer a more general solution and becomes widely used, that can only be good news for anyone who gets ill in future.
Next episode of House;
- person comes in with unusual symtoms...
- a junior Dr asks computer what could it be...
- they cure patient!
- Meanwhile House (aka Hugh Laurie) stands in the background ranting and shouting I was going to say that....(eventually).
Everybody’s physiology is slightly different, case in point, perfume. While, from the bottle it generally smells the same to everybody, once it’s applied to 10 women, it’s subtly different on each one.
Watson may help narrow things down slightly, but it’s not responsible for the ultimate welfare of the patient, the doctor is.
However, I am curious, what’s say a patient comes in complaining of flu like symptoms, and the Watson device misdiagnoses the person’s malady, who’s held responsible, IBM and Watson’s programmers, the insurance company or the Doctor and the medical institution he’s affiliated with?
I don’t know, I’ve got really mixed emotions about this, but I’m leaning towards the “I don’t think this is the smartest idea” side.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
