Slashdot Mirror
Watson Goes To Medical School
First time accepted submitter Kwyj1b0 writes "I.B.M's Watson is headed to the Cleavland Clinic Lerner College of Medicine of Case Western Reserve University for training. Clinicians and students will answer and correct Watson's questions, in an attempt to crowdsource its education. From the article: '“Hopefully, we can contribute to the training of this technology,” said Dr. James K. Stoller, chairman of the Education Institute at Cleveland Clinic. The goal, he added, was for Watson to become a “very smart assistant.” Part of Watson’s training will be to feed it test questions from the United States Medical Licensing Exam, which every human student must pass to become a practicing physician. The benefit for Watson should be to have a difficult but measurable set of questions on which to measure the progress of its machine-learning technology.'"
oops, looks like I couldn't bother verb there. I should get a /. editor job too.
I recently worked at IBM and this is one of the things they really love to showcase - I think primarily because it's really cool, but also has really useful end results - exactly the kind of thing you want to be working on.
-- Lattyware (www.lattyware.co.uk)
Design another Watson designed as a database for crimes which could analyse crime scenes, point out potential minute details and give data on similar crimes. Then call it... Holmes.
I'd like that.
Reminds me of a short story from the Stories of Ibis collection. A salesman impressed a group of colleagues in a nursing home that a new robot had passed the nursing exam.
Do you see what I did there?
just wait for it to make a error like that in a Medical setting.
Watson actually does some really cool stuff with respect to not just being 'do this' - it tells you what it 'thinks', but also tells you why it thinks those things, and how sure it is of that. So it will say 'I think it's likely the patient has Y because of family history, environmental factors, this symptom and these studies', etc... It's more giving the doctor all the (relevant) information possible than telling the doctor what to do.
-- Lattyware (www.lattyware.co.uk)
Particularly since the way it work is by probabilities. So a physician goes and inputs all the symptoms a patient reports, perhaps along with a confidence of how likely it is to be real. Watson could then spit out the likely causes, and the probability of each, as well as how to narrow it down. Then with additional tests, they can exclude things and get a re-factored list.
It won't remove the need for a medical professional with good judgement, but it could be a boon for searching through things and presenting possibilities. What's more each new case can be logged, improving its database.
So when someone presents with a rare disease, it would be much easier for a physician to diagnose it, even if they've never heard of it.
If implemented right, it could cut down on misdiagnosis a ton.
The American Medical Schools select for automatons who can memorize and regurgitate vast amounts of data. Talk to any physician who graduated from med school in the past 20 years about what they took as undergrads and they'll most likely tell you they don't remember the courses specifically because they memorized them to pass (and then promptly forgot them).
The real question is whether or not this is the best way to train our future health care professionals. While indeed there are some really good physicians coming out of our med schools - and even some of those who memorized their way through undergrad will be great physicians - we have also excluded from selection many who would have been excellent caregivers based on their inability to memorize quite as quickly as their classmates.
Damn_registrars has no butt-hole. Damn_registrars has no use for a butt-hole.
I know it's just a joke, but just to clarify: the big difference here is the scope of what Watson does. Websites generally take one symptom and list off all possible causes (so naturally people focus on the worst ones). Watson takes (in natural language) patient history, symptoms, environmental factors, current medication, etc... - everything a doctor has access to (and a doctor would have to perform the exam to make this worthwhile), and then cross references with a huge amount of knowledge from papers, classes like this, 'experience', etc... this is then used to give an idea of what the patient has, how to treat, etc... It's really cool.
-- Lattyware (www.lattyware.co.uk)
Errors like that can get you on the pre-existing condition black list. Where after that your only doctor may be Watson ER
lots of colleges put to much on cramming for tests and not much on being able to do stuff in a real setting. Closed book / closed notes / no Google tests only test memorizing.
What professionals one who is real good at test cramming or on who knows most of the day to day stuff and how to lookup the other stuff that they need at the time they need it. is better?
No humans involved. Only computers and the occasional tech support call to Bangalore.
When I saw the twitter feed headline, I was hoping Lucy Liu was going to attend UW Medical School.
Sigh.
Dr. Watson, you're needed here!
-- Tigger warning: This post may contain tiggers! --
In the medical profession we call those things Lists. A number of studies, based on original research at the UW Medical Center and VA hospitals, have found that check lists do wonders.
But ... we already do that. We even have iPads in VA Hospitals now. Watson is just doing checklists.
-- Tigger warning: This post may contain tiggers! --
No one. Duh.
"I.B.M's Watson is headed to the Cleavland Clinic Lerner College of Medicine of Case Western Reserve University...
/. can't afford its own watson, how about a spell checker??? They are much cheaper and much, much less likely to make mistakes.
And if
Circle the wagons and fire inward. Entropy increases without bounds.
and they wonder why they cant find anybody to place in positions. between this and the ads to 10 years exp in such and such and if you 9 you are out BS.
NEVER NEVER NEVER NEVER NEVER NEVER NEVER NEVER GIVE UP! "No limitations, no boundaries, there is no reason for them."
...and call it Professor Moriarty
It's always Lupus..
It`s never Lupus.
GENERATION 24: The first time you see this, copy it into your sig on any forum and add 1 to the generation. Social exper
"I.B.M's Watson is headed to the Cleavland Clinic" Really? It's the _Cleveland_ Clinic. Seriously "samzenpus". That's pretty bad. Even worse for the fact that in the next sentence it's spelled correctly. A bit too much Halloween bubbly perhaps?
You are wrong, broadly speaking. This is the whole point of machine learning: given a very complicated task that it would take a human a tremendous amount of effort to program correctly, you can instead get the machine to figure out how to perform the task itself, rather than explicitly programming it to do one thing. Some types of learning are supervised, particularly classifiers: I tell the computer which items belong to which class, and given a new, previously unseen item, the computer attempts to determine its class based on the training. Others are unsupervised: set the robot free in the environment with some goal function and let it learn through trial-and-error how to optimize its behavior toward the goal. Watson is a combination of first-order logic (prolog and a huge kb) and a variety of such learning algorithms. Some of this is stuff that was considered an industry failure in the 80s but, paired with modern machine learning techniques, is quite powerful. Indeed we may be seeing the first instances of computers that have some form of this "intelligence" of which you speak, though I think we are still a long way from "strong AI".
Virtual Doc: You've got: leprosy.
What I would love to see is Watson's training interface on the Internet, as a service, with anyone able to pick a domain and contribute expert knowledge, whether in the form of questions Watson should ask, answers to those questions, or even just links to sources of relevant information. Through a crowd-sourced approach Watson's capacities could be so much more quickly developed. By keeping each user or group in sandboxes and maintaining knowledge in each domain more or less separately, there would be no problem for those who just input nonsense, or wrong information, because Watson can build up a reliability profile and consider that in later recombination of its knowledge.
Of course first there has to be a nice easy way of making Watson nodes that can be widely deployed. Frankly this is one area of research that deserves all the money we can afford to throw at it. In the future Watson will be able to derive new hypotheses and new knowledge obtained from inference, and it will certainly accelerate our research capacity.
-- thinkyhead software and media
"Cleavland Clinic Lerner College of Medicine of Case Western Reserve University" Anyone else a CCLCMCWRU fan? :)
In my engineering classes they are pretty much allowing us to use books, notes, homework, previous exams, and pretty advanced calculators like the nspire but the also give FAR more realistic problems.
The exams tend to be very difficult and sometimes have unrealistic time constraints (ie nobody in a class of 100 people or so finish the exam).
Pretty much the only thing we can't use is a laptop/cellphone etc or anything with an internet connection.
It would certainly be nice if I could use matlab, python or excel on exams.
Computer modeling for biotech drug manufacturing is HARD!
Soon we will have affordable, quality healthcare for all.
Next goal world peace!
--whacky
Nurse: Doctor patient Parkison, has numbness in the groin area.
Web MD : Did the numbness come on suddenly?
Nurse: No.
Web MD: Does it affect both sides?
Nurse: Yes.
Web MD: Thoracic Spinal Stenosis
Watson: Parkinson's disease. (Patent Pending)
Really they've put the text through a crude parser and stuck it into a database similar to Web MD and now they'll patent the hell out of this. But they're using stock language parsing algorithms and stock medical look databases, and the only purpose for this is to generate fake cover for a lot of patents in the medical field.
You won't see any product from IBM in this field other than the patent.
Thought this was going to be about Emma Watson! /disappointed
"You can justify anything by putting it in quotes, adding a famous name and making it a sig" - Albert Einstein
He might discover the complexities of designing a health department pay system, since IBM screwed up the Queensland system so badly.
Go well
Older systems used expert systems (hierarchical knowledge systems). The new models use machine learning (neural network classifiers, ...).
The machine learning creates a generalized model based on small amount of data. Expert systems search the existing data and the data should be very accurate.
Sure, and I could get a team to spend a week performing analysis to determine appropriate staffing levels - dealing with tens of thousands of data points, with data at all sites varying on an hourly basis. Alternatively I could have a learning system that keeps far better track of these things than humans would, and presents its working for review before implementation.
-- Using the preview button since 2005
Stone/earth is what computers are made of and they run images of human/beast thought processes...... So what is coming is ......
There is more to life than just abstraction, which is what language is. There is more to reality than what our abstraction identify...
Some things worth considering.
"Watson is just doing checklists" - You miss the point, in a year or two, it will probably be more capable of writing and correcting those lists than any single human being.
And did you exchange a walk on part in the war for a lead role in a cage? - Pink Floyd.
Is a computer psychiatrist .... let's call him Marvin!
load "$",8,1
Cleavageland would have been a better typo!
This is a UDP joke, I don't care if you get it or not...
Watson better learn to read crabby handwriting and minds. The fundamental issue in applying AI to medical data is the low information dataset. We are working on a couple of smaller AI projects in my hospital and finding that even when electronic, most entries in the record are narrative not discrete.
Applying ranges and logic to data pulled out of narrative records is tricky and leads to unusual responses. Even when discrete, the data set may be difficult to use. My favorite example seen nationally is blood transfusion data. It can be in units of 250ML, 500ML or just ML. Users don't look at the unit label for a field. If we ask for the most precise measurement in ML, we get amounts of 1 or 2. We know nobody got a thimble of blood.
AI interventions have to be quick. It's no good telling the provider later he made a mistake. The goal is to steer them in the right direction before they act without presenting them with so much info they throw in the towel. Not easy and in infancy.
Oops wasn't logged in but that was me, in case. You were wondering.