Slashdot Mirror
Artificial Intelligence Can Now Predict Suicide With Remarkable Accuracy (qz.com)
An anonymous reader writes: Colin Walsh, data scientist at Vanderbilt University Medical Center, and his colleagues have created machine-learning algorithms that predict, with unnerving accuracy, the likelihood that a patient will attempt suicide. In trials, results have been 80-90% accurate when predicting whether someone will attempt suicide within the next two years, and 92% accurate in predicting whether someone will attempt suicide within the next week. The prediction is based on data that's widely available from all hospital admissions, including age, gender, zip codes, medications, and prior diagnoses. Walsh and his team gathered data on 5,167 patients from Vanderbilt University Medical Center that had been admitted with signs of self-harm or suicidal ideation. They read each of these cases to identify the 3,250 instances of suicide attempts. This set of more than 5,000 cases was used to train the machine to identify those at risk of attempted suicide compared to those who committed self-harm but showed no evidence of suicidal intent.
not artificial intelligence.
Give people a reason to not kill themselves and you'll see rates drop.
We used to have a Bill of Rights. Now, with the rights gone, all we have left is the bill.
Why is it you say you don't find ELIZA to be that effective as a program?
t
So who can do the calculations for the false postives and the false negatives? Because I am sure that this will calculate that I am willing to kill myself, even if I have no desire to do so and tell me that I won't when I am willing to do so.
Don't fight for your country, if your country does not fight for you.
your are correct, these morons used a group of suicidal patients for their case study and now are claiming great success.
When the algorithm discovers it can improve accuracy by driving people to suicide by being linked to robocalling systems
Nullius in verba
You have been deemed to be suicidal. Please check into your nearest healthcare location. Refusal to do so will result in you being placed imminently into level two treatment. Which may result in loss of job, loss of family, and the loss of your pet named spot.
Simple accuracy percentages are misleading when applied to low-probability events. An "AI" that always returned "No" to the query "Will this person commit suicide within the next two years?" would be 97.2% accurate (and 99.975% accurate for the next-week variant). And yet, that "AI" would be absolutely useless for any practical purpose.
Not to mention, with suicides, access to means has been a better statistical predictor than anything else, even mental illness. A person with no personal or family history of mental illness, but with a gun and a gas oven in their house, is at higher risk of killing themselves than a bipolar alcoholic with neither.
Why is it you say you don't find ELIZA to be that effective as a program?
All those questions are enough to drive someone to commit suicide. Wait a minute...
I have made an algorithm that says that of those who never had previously tried suicide and then did it successfully, 97% did it for the first time. (+-3% accuracy on the calculation)
Don't fight for your country, if your country does not fight for you.
As someone who's been down that road (but never gone through with an attempt), I automatically hate this invention. When depressed to that point, emotions tend to swing so hard and so fast that any mention of predictions during this state of mind is utmost bullshit.
The very slightest of triggers can either send you overboard or keep you in one piece depending on how your inner conversation is going with yourself. This can be anything... a faint sound, perhaps a song that reminds you of good/shitty times, from a car passing by not too far away.
I consider myself lucky to be both scared of the afterlife enough to have thoughts force second-guessings into me (although the older I grow the less I care), and have enough positive triggers to bring myself back. Nobody, not even myself, could predict if these will always work for me as well as they have however.
Suicidal/depressive folks definitely need help, but not from the machines of this day and age. A positive trigger could well be overridden by a "fuck it", and it only takes a split second to follow through the act. You can't predict that kind of stuff with a high degree of accuracy, at least not yet.
Disclaimer : I did not RTFA. I find stuff like this appalling as it hits me right in the feels and I would be deeply insulted if a machine tried to guess whether I was going to kill myself or not. There's much more to it than some algorithms a team engineers wrote.
I tend to rant.
I don't need a computer to tell me that there is a good chance some of these people will attempt suicide again.
Yes, but which ones? That's the whole point, surely? You'd want to use this as a diagnostic tool, in cases where you're dealing with a lot of depressed people and you need to know which ones you particularly need to watch out for in terms of suicide risk. Mental health clinics would find this invaluable, wouldn't they?
It's pretty much the same thing as being able to tell a cardiac clinic which of their heart-disease patients are most at risk of having a heart attack soon. Obviously everyone who is a patient there will have a problem of some kind, but being able to distinguish those who need urgent attention from those who just need run-of-the-mill care is one of the primary requirements of the job.
-- Note to Mods: There is a good reason there's no "-1 Disagree" option. --
It's probably mostly meaningless. I mean, they scanned for features of people who are suicidal. They were in the hospital because they inflicted self harm, and were on medications specifically prescribed to make people not do that. So as far as I can tell, this doesn't predict anything, it juts measures that "80-90% of the time doctors do the same thing for folks who would hurt themselves".
It's not like they randomly picked a bunch of people off the street and determined from THAT. Like basically every single other artificial intelligence or machine learning story, it's a bunch of dumb hype, eventually to get folks investing in stupid startups.
I expect that an 80-90% accuracy means that in a group of X people is correctly identifies 80-90% of the people who later go on to attempt suicide. However, if you ignore the false positive rate then I can make an even simpler algorithm that is 100% accurate: simply tag everyone as a suicide risk.
I wish that those reporting on medicine had a basic grasp of science and simple statistics so that they could ask the relevant questions such as: what is the false positive rate?, does 80-90% mean that your statistical error is 10%?, what is the successful rate of doctors predicting suicide risk?, is this algorithm i.e. the types of questions that are critical in determining whether this algorithm is actually useful!
So, once the computer diagnoses someone as highly likely to kill themselves in the next week, then does it (or the user) call the men in white coats to give the subject the coat with the funny sleeves? Therapists frequently have a statutory or license requirement to report potential suicides.
We don't know what the rate of false positives are, but with our current state of health insurance, getting locked up for a week and then getting a $50k bill would probably drive most people to suicide.
The title speaks of suicides while the article only of _attempted_ suicides, checking admissions to hospitals
Real suicides get admitted to the morgue instead.
The 80s called, Comrade! They want their Soviet meme back. In the meantime, Cuban, North Korean & Venezuelan comrades are up in arms at a non-Communist entity like Russia still keeping the 'comrade' moniker
The group that was being analyzed was already considered "high risk", out of 5,167 cases there were 3,250 attempted suicides. So even if those were false positives, it wasn't an amount that dwarfs the actual predictions. Now if they expand this to a larger, less risky group, who knows, but at this stage the false positive rate seems more than acceptable.