Nobody knows when they’ll die. But artificial intelligence (AI) may.
At least, that’s the idea behind the latest MedTech development from the ‘Medical Brain’ team at Google.
Deep Learning & Death
Deep learning is a subfield of machine learning which centers around learning data representations instead of just task-specific algorithms. In a new paper published in Nature, this category of AI demonstrates that not only can it predict the length of a patient’s stay, their projected time of discharge, and their chances of readmission, but it can also estimate the time of death.
Data in the paper came from trials spanning two U.S. hospitals. In one incident, a woman afflicted with late-stage breast cancer saw two doctors and received a radiology scan. After the hospital computers read her vital signs, it came back with a 9.3% probability that she would die during this visit. After assessing 175,639 data points, a deep learning neural network from Google estimated the probability of death to be at 19.9%. The patient passed away days later.
This new neural network comes from the Medical Brain team at Google. Acting as the San Francisco development giant’s AI health research unit, Medical Brain is led by Jeff Dean, Google’s AI chief. They’re currently working on a plethora of AI tools centered around medical applications.
The Vast Potential of Neural Networks
Medical experts were thoroughly impressed with the neural network’s ability to seamlessly sift through immense amounts of data. Information previously out of reach, like scribbles on old records or chicken scratch on PDFs, were all ingested and utilized. An algorithm would then align the patient’s health history into a timeline, which the deep learning model then used to predict future outcomes.
Learning how to harness the mountains of medical data at their disposal has proven to be an impossible obstacle for the healthcare industry in recent years. Not only was the AI faster and more accurate than computer system currently employed, but it even showed which data points led to its conclusion, like the patient’s predicted time of death.
This technology is giving healthcare workers a different perspective on maladies, which in turn could allow them to adjust treatment, prioritize patient care steps, and even detect medical emergencies before they happen.
This technology could also disrupt healthcare workflow on a profound level. Nigam Shah is an associate professor at Stanford University and a co-author of the recently published research. He says that almost 80% of the time and effort spent on current predictive models goes towards the process of formatting and feeding the data properly. Removing this barrier could free up more than just a few hours for each worker.
Can This AI Predict Its Own Future?
Innovations like this neural network could help Google break into the healthcare business, something it’s been trying to do for quite a while now. AI chief Jeff Dean said the company’s next move is to implement this system in clinics. He foresees a future where this AI is helping doctors optimize their diagnoses and treatment plans.
But before that happens, Google’s AI does have its own fair share of obstacles in its path. The obvious one is the question of whether people feel comfortable giving one of the world’s largest tech giants even more personal data. Many skeptics think that allowing this could let Google exploit the industry and even grow into a monopoly in healthcare.
Recently, DeepMind Health, owned by Alphabet, Google’s parent company, faced heavy scrutiny from the U.K. government over the possibility it could “exert excessive monopoly power.” This is coming off of previous allegations that DeepMind Health collected patient data without gaining proper consent last year.
There isn’t any established regulatory framework in the U.S. that holds companies accountable by promoting transparency. This essentially leaves private companies open to interpreting whether their products are actually impacting patients in a positive way or not.
What do you think the future of Google’s latest AI innovation is? Do you think it could deduce its outcome?