Want to Find Out the Moment of Your Death? Google's Deep-Learning AI Can Tell You
If it was possible, would you want to learn the moment when you will die? If Google's new artificial intelligence is as sophisticated as its scientists say it is, you may one day have the chance to answer that question for real.
Hearing a doctor say "You have a week to live" is probably the last thing any person wants when they walk into a hospital, but it turns out that these diagnoses aren't always foolproof—there are many stories of patients living months or even years after their supposed death date, while other patients' deaths come suddenly and without apparent cause...until someone realizes that the medical staff missed a key detail.
The key difficulty of modern medicine, it seems, is making sure healthcare providers have all the information they need to make the right decisions, but with so many factors to sort through, it can be impossible to spot the relevant ones. So researchers have proposed bringing Google's deep-learning algorithms to the table.
Their goal is to be able to predict the course of a patient's treatment by using artificial intelligence to analyze their medical records, which, according to the new study published in Nature, turns out is much more efficient and accurate than having humans do it:
"Deep learning models achieved high accuracy for tasks such as predicting: in-hospital mortality, 30-day unplanned readmission, prolonged length of stay, and all of a patient's final discharge diagnoses. These models outperformed traditional, clinically-used predictive models in all cases. We believe that this approach can be used to create accurate and scalable predictions for a variety of clinical scenarios."
Sharp-eyed readers will notice that "in-hospital mortality"—the death of a patient—is one of the key events that the AI program aims to predict.
That might sound like a step too far for science, but when you consider that researchers have already created AI that bests expert human fighter pilots in simulated dogfights, accurately predicts what you'll do in the future, can read your emotions, and even actually tell us if you're lying, we clearly passed into Skynet territory a long time ago. And I even made that point without having to mention the world's first "psychopathic" AI yet—so, yeah, you get the idea.
The problem, researchers admit, is that all the relevant data to make these predictions is scattered across multiple databases, hand-written notes made by doctors, and even different institutions.
To help organize it all, the research proposes collecting all the raw data and putting it in a format called FHIR (which stands for "Fast Healthcare Interoperability Resources").
Still, hospitals may be unwilling—or unable—to share their data, even if it means the difference between life and death.