And the diagnosis is...
"It's not lupis. It's never lupis. Now give me my Vicodin!"
IBM has signed a deal with health insurance provider WellPoint to use Big Blue’s Watson question-and-answer system to help doctors decide what’s wrong with you – and offer possible remedies. The Watson machine – which famously beat humans at the popular US game show Jeopardy! earlier this year – will now be put to task …
As some-one with major health problems which went undiagnosed for over three years, I feel this can only lead to better health care, and better practice in health providers.
I'm sure there are many other applications where knowledge is advancing at a rate where it's impossible to keep up with developments without computer assistance.
A beer for Mr Blue!
If it comes up with answers that are *better* than medical experts can ever comprehend, how can we ever be certain it isn't wrong?
If it comes up with wrong answers, who would then have the confidence to assert that - in their expert opinion- it was acutally malfunctioning rather than offering new insights?
I didn't see the bit about '*better* than medical experts can ever comprehend'.
From the article, it seems the idea is that it will come to the same conclusion as a doctor would, if the doctor had all the relevant info and analysed it thoroughly. A doctor will then check the conclusion, and say "By Jove, I think it's got it".
At least initially. I imagine the long-term hope is automatic diagnosis without a human doctor at all.
that if it is basing its decisions on more information than a doctor could ever read and digest...
If it finds a previously unknown relationship between a range of symptoms and an illness - but the catch is you need to read 75m medical records to understand why it recommends a heart transplant - how do you know whether it is right or simply malfunctioning?
3 years ago. Ended up with a gigantic Bayesian tree with only a hundred or so diseases simply because you don't always present with the same symptoms, which screwed over our false negative rate for a lot of the early work. Also why we got rid of psych disorders, too many possibilities to code for without more research than a idle side project could be afforded. Great fun while it lasted though. Wonder what the false positive rate will be when compared to doctors before Med tests are administered.
Read the opening sentence again. Twice. You seeing what I'm seeing?
-- -- -- IBM has signed a deal *** with health insurance provider WellPoint *** to use Big Blue’s Watson question-and-answer system to *** help doctors decide *** what’s wrong with you...
I envision the conversation going something like this:
-- -- Insurance Adjuster: "Well, Doctor, wha'd'ya think about patient 656-5827B?"
-- -- Doctor Livingstone: "Based on symptoms, obstructed or dysfunctional gallbladder; suggest surgical removal followed by post-operative enzyme level monitoring and dietary changes"
-- -- Insurance Adjuster: "Watson, your analysis of patient 656-5827B?"
-- -- Watson: "Simple stomach ulcers and/or gastroesophageal reflux disease; suggest treatment with standard course of generic proton-pump inhibitor medication and stress-reducing lifestyle changes."
-- -- (Insurance Adjuster thinks: "Watson's diagnosis is cheaper; we'll go with its recommendation.")
-- -- Insurance Adjuster (out loud): "OK, we'll pay $75.00 per month to cover acid reflux control prescriptions. If 5827B's condition doesn't improve within 90 days, we'll revisit the case. Next..."
Now, if Watson's diagnoses were 100 percent filtered through the patients' doctors before they landed on the desks of the insurance adjusters, I wouldn't have a problem with it; I believe that Weak AI like Watson can go a long way toward helping doctors make the right call the first time around.
But that isn't what's going to happen. Instead, the insurance companies will use Watson's capabilities as a hedge against doctors recommending expensive but effective cutting-edge treatments for their patients, and steer them toward more prosaic and less costly solutions.
Similar medical software exists, see one famous example used for infectious diseases, GIDEON here:
http://en.wikipedia.org/wiki/Global_Infectious_Disease_Epidemiology_Network
Quote: "...generates a Bayesian ranked differential diagnosis based on signs, symptoms, laboratory tests, country of origin and incubation period - and can be used for diagnosis support and simulation of all infectious diseases in all countries."
So GIDEON doesn't give a single answer like on Jeopardy, but a ranked list, which the infectious disease specialist uses as an aid to diagnosis.