Google AI now predicting if you'll die within 95% accuracy

[rcmaehl]

Sources: 

https://www.foxbusiness.com/markets/google-ai-can-predict-when-you-will-die
https://www.vanityfair.com/news/2018/06/google-thinks-it-can-predict-when-you-will-die
https://www.bloomberg.com/news/articles/2018-06-18/google-is-training-machines-to-predict-when-a-patient-will-die

 

Quotes:

Quote

Google is training its new artificial intelligence computer to predict whether hospital patients will die 24-hours after admission. The accuracy of the initial trials is at 95%.

 

Quote

What impressed medical experts most was Google’s ability to sift through data previously out of reach: notes buried in PDFs or scribbled on old charts. The neural net gobbled up all this unruly information then spat out predictions. And it did it far faster and more accurately than existing techniques. Google’s system even showed which records led it to conclusions.

 

Quote

According to Bloomberg, Google’s ambitions span many areas of the health-care field, including radiology, ophthalmology, and cardiology. Last year, Google developed an algorithm that detected diabetic retinopathy, the fastest-growing cause of blindness, with over 90 percent accuracy. 

Media:

http://video.foxbusiness.com/v/5799198997001/?#sp=show-clips

 

My thoughts:

Jesus Christ this is scary. However, if it'll help people get treatment for issues that'll probably kill them within 24 hours. I'm all for it. Machine Learning is getting scary.

Edited June 19, 2018 by rcmaehl
Added more sources and quotes

[minibois]

I thought this was gonna be one of those early 2000's website that will say: "You will die on xx-xx date in xxxx year!" that used to be spooky for elementary school kiddos like myself.

But no, this is actually useful, but only for hospital patients..

[Princess Luna]

Click-bait title, learned with the finest; LTT.

 

Seriously, fine medics can out of their experience and knowledge make as much as accurate prediction after your entry trial at the hospital, this just automates the process.

[AskTJ]

http://www.deathclock.cc/

 

Much more accurate IMO.

[Christophe Corazza]

How can it be 95% accurate, who confirms the accuracy?

 

Just look at the expiry date on your birth certificate.

[Lurick]

Just now, Christophe Corazza said:

How can it be 95% accurate, who confirms the accuracy?

The dead people

[rcmaehl]

3 minutes ago, Christophe Corazza said:

How can it be 95% accurate, who confirms the accuracy?

 

Just look at the expiry date on your birth certificate.

It's predicting when you'll regularly die... It's predicting if you'll die within a period of time of being admitted to a hospital for an issue. Basically, would help doctors put a rush on things if someone is going to die extremely soon.

[rcmaehl]

7 minutes ago, Princess Cadence said:

Click-bait title, learned with the finest; LTT.

Updated with slightly better wording. 

[captain_to_fire]

20 minutes ago, rcmaehl said:

Machine Learning is getting scary.

I want to see a long term study on Google's AI that is corroborated with credentialed authorities especially ER doctors and pathologists if it's indeed true that it can predict time of death in a hospital. If what Google has been doing is training their ML models to determine who will die from a gunshot wound in the chest or someone having a stroke, then I can say that their ML model is shoddy because you don't need to be an expert to know that a GSW in the chest or a blood clot in the brain creating a hypoxic environment can kill a person within minutes.

 

 

[Christophe Corazza]

I think some hospital/clinic waiting rooms already use this technology.

 

while patient.isnt_quite_dead():

        let_them_wait()

[captain_to_fire]

2 minutes ago, Christophe Corazza said:

I think some hospital/clinic waiting rooms already use this technology.

 

while patient.isnt_quite_dead():

        let_them_wait()

I hope not. First of all, that could mean lawsuits against the hospital if the doctors and nurses didn't attend to the needs of the patient brought to the ER because an AI in a computer says that the person is going to die and the staff can't be bothered to push to use a defibrilator.

[Doobeedoo]

Now, let us us humans predict Google AI death. Hmmm... 

[Ildiesalot]

Well technically if someone will die will always be true! So I can predict if anyone will die with 100% accuracy.

 

Hire me Google.

[pizzaboard]

To get that 95% accuracy  it means that AI contains millions of information from studies and real life situations. To predict that the patient will die 

[Drak3]

So what? Give me a knife and I can do the same damn thing.

 

 

 

 

 

Kappa.

[LAwLz]

39 minutes ago, Princess Cadence said:

Seriously, fine medics can out of their experience and knowledge make as much as accurate prediction after your entry trial at the hospital, this just automates the process.

[Citation Needed]

 

 

7 minutes ago, huilun02 said:

But seriously wtf is the doctor doing uploading their patients' medical conditions to the internet??

This is not just Google scraping stuff they can find online on open website.

It's a collaboration between hospitals and Google directly.

[ScratchCat]

100% of patients died after being told that there was a 95% chance they would die.

2 minutes ago, LAwLz said:

[Citation Needed]

Assuming death rate is less than 5% of patients guessing "Will live" would give you a 95% accuracy so the accuracy itself is not impressive.

[Princess Luna]

9 minutes ago, LAwLz said:

[Citation Needed]

Are you serious? okay, forget people who do 5 years of college and another couple years of residency and later another many years of specialization are incapable of prediction on their area of expertise, we totally need AI or we're doomed.

 

Common Sense works fine, finding specific nonexistent research of every logic thing in life just to please someone's need to debate/argue is not practical in life.

[captain_to_fire]

15 minutes ago, LAwLz said:

[Citation Needed]

No citation needed for that. Medics, nurses and doctors are trained not only to save lives but to determine prognosis and chances or survival. For every minute that passes by, a stroke patient's survival is getting slimmer. So I don't know what kind of citation do you need? Medical records? Not gonna happen as those are strictly confidential. How about textbooks in emergency medicine?

[mynameisjuan]

14 minutes ago, huilun02 said:

Did patients give consent?

And how will this benefit them?

So they get spammed with ads and snuffed by insurance agencies?

This data is probably from bodies on the doner list. Technically is still research.

[GoldenLag]

95% accuracy is terrible in the medical sence. 95% accurate tests om a mass scale is bad. Which are some of the reasons doctors dont recommend healthy people with no family history to go get a mammogram or certain other tests that can come out with a false-positive

 

Still quite cool though

[mynameisjuan]

6 minutes ago, GoldenLag said:

95% accuracy is terrible in the medical sence. 95% accurate tests om a mass scale is bad. Which are some of the reasons doctors dont recommend healthy people with no family history to go get a mammogram or certain other tests that can come out with a false-positive

 

Still quite cool though

You mean like the accuracy doctors give their patients and family. "You have 2 months, 3 top"....."Four years later"

[Cookybiscuit]

27 minutes ago, huilun02 said:

And how will this benefit them?

Probably so socialist countries with "free" healthcare can simply let people die on the operating table because Google said they'd be unlikely to survive.

[captain_to_fire]

8 minutes ago, mynameisjuan said:

This data is probably from bodies on the doner list. Technically is still research.

Found the primary source and the data definitely didn't come from a donor list. They used actual patient records. https://www.nature.com/articles/s41746-018-0029-1

 

Now Google claims that they anonymized and the data they have collected cannot be traced to a single patient and that they submitted their proposal to an ethics committee with consent from each hospital.

Quote

We included EHR data from the University of California, San Francisco (UCSF) from 2012 to 2016, and the University of Chicago Medicine (UCM) from 2009 to 2016. We refer to each health system as Hospital A and Hospital B. All EHRs were de-identified, except that dates of service were maintained in the UCM dataset. Both datasets contained patient demographics, provider orders, diagnoses, procedures, medications, laboratory values, vital signs, and flowsheet data, which represent all other structured data elements (e.g., nursing flowsheets), from all inpatient and outpatient encounters. The UCM dataset additionally contained de-identified, free-text medical notes. Each dataset was kept in an encrypted, access-controlled, and audited sandbox.

 

Ethics review and institutional review boards approved the study with waiver of informed consent or exemption at each institution.

 

[GoldenLag]

1 minute ago, mynameisjuan said:

You mean like the accuracy doctors give their patients and family. "You have 2 months, 3 top"....."Four years later"

Which is why cancer tests arent tye greatest.

 

They cant tell tye difference between cancers.

 

Some grow so fast you cant be saved.

Some grow so slow they can kill you before you grow old.

Some stop growing for no good reason

Then you have the one that you can be saved from.

 

They treat all of them the same just to be shure. And the cancer treatments take a toll om the body.

 

Id still recommend taking the treatment, just dont go around testing yourself 24/7 unless the doctor actually recommends it