The intersection of advanced technologies and medicine has been one of the most exciting areas of development lately. CRISPR, for example, is set to revolutionize the way people are modified on a genetic level, and 3D printing skin with working blood vessels is already totally doable (at least when it comes to mice). And now, we can add to that list of technological breakthroughs in medicine an AI program that can determine whether or not you’re going to die within the next year. Charlie Brooker, we hope you’re paying attention.
Geisinger studies show AI deep learning model helping cardiologists detect AFib: A team of scientists trained a neural network to evaluate electrocardiograms to predict which patients were likely to develop an irregular heartbeat. https://t.co/SERLSeGxjZ pic.twitter.com/GMgKxyjevG
— Evan Kirstel at #AvayaEngage #Dubai (@evankirstel) November 11, 2019
While we’ve seen other examples of medical technologies aimed at predicting people’s remaining lifespans in the past, including this “death panel” blood test, this new method for figuring out when people are going to die is particularly fascinating, and frightening, because it utilizes an artificial intelligence that comes to its conclusions in a mysterious way that is currently unintelligible by its creators.
New Scientist reported on the mysterious AI, which is being developed by researchers at Geisinger, a healthcare provider located in Pennsylvania. According to the New Scientist report, which comes via Futurism, the team at Geisinger, led by the company’s Cardiac Imaging Technology Laboratory Director, Brandon Fornwalt, fed an AI—specifically a deep neural network—1.77 million electrocardiogram (ECG) results taken from almost 400,000 people. Electrocardiograms are recordings of how much voltage a given heart produces over time as it beats.
— American Heart News (@HeartNews) November 11, 2019
After training the AI with the plethora of ECGs, as well the stats for who did and did not die within a year from that specific group of patients (within a year of having their ECGs recorded, that is), the researchers put it to the test with cases where it didn’t know the outcome for the patients being analyzed. Using an Area Under the Curve (AUC) metric, the researchers were able to determine that the AI was able to figure out who was likely to die within a year with a probability far higher than simple chance guessing. The AI earned a score of 0.85 on a scale from 0.5 to 1 with, 0.5 meaning no ability to distinguish who was going to die within a year and 1 being a perfect ability to distinguish who was going to die within a year.
Looking toward the future, even though it seems like this would be a technology that should be deployed ASAP to help determine who’s at risk of dying within a single trip around the Sun, it seems physicians are reticent when it comes to using it because they don’t understand how the AI came to its conclusions. In fact, New Scientists says that “The AI accurately predicted risk of death even in people deemed by cardiologists to have a normal ECG.”
— Red Health (@RedHealth) May 21, 2019
Fornwalt told New Scientist that the fact that physicians don’t understand how the AI came up with its results ” suggests that the model is seeing things that humans probably can’t see, or at least that we just ignore and think are normal.” He added that “AI can potentially teach us things that we’ve been maybe misinterpreting for decades.” This possible trend seems a lot more disconcerting than an effective, yet mysterious, AI—although knowing why exactly somebody is going to die within a year seems like the most important piece of information to have in regards to actually saving lives.
What do you think of this AI that can accurately predict whether or not somebody is going to die in a year? Would you ever let it get a peek at your ECG, or would you rather never know? Give us your heartfelt opinions in the comments!
Feature image: speedoglyn1