Medical brain from Google. The neural network has learned to predict the death of patients: what next?

Google claims to have finally figured out how to apply artificial intelligence to medicine. The company has published a study on its new technology, which uses AI to help determine how a patient's condition will change, twice as accurately as doctors can predict. After examining a woman with cancer, doctors concluded that the probability of death is 9.3%. The Google system, after analyzing more than 150 thousand pieces of data, gave a probability of death of 19.9%. The patient died within a few days.

Why is a neural network better than a doctor?

• What surprised medical experts the most was that Google's technology was able to analyze medical data that was previously inaccessible to artificial intelligence: notes from doctors and readings from old studies. The neural network not only dug up all this messy information and made a prediction, but also did it much faster and more accurately than all the systems that currently exist, while highlighting exactly which data influenced the final conclusion.
• The main advantage of the Google system is that it does not need to spend time putting the data in order (while in other predictive models it takes 80% of the time). Most of today's systems of this kind of data analysis have to be configured manually, but the Google model does it without outside intervention, learning and developing in the process.
• The company noted that the new technology will help doctors predict not only the death of patients, but also the chances of their re-admission to the hospital and the time spent in medical facilities.

Main characters

• The development of technology at Google is carried out by a division of the company with the unofficial name Medical Brain (“Medical Brain”). When creating a product, developers closely consult with medical professionals, said Jeff Dean, head of artificial intelligence at the company.
• One of the employees of the company believes that Google has finally found a use for artificial intelligence, which looks promising from a commercial point of view. Since the company took a course in AI development in 2016, the efforts have been mainly spent on improving existing Internet services, rather than creating new ones. The success of Medical Brain could enable Google to break into an entirely new market, something founders Larry Page and Sergey Brin have repeatedly tried to do.

What's next

• Despite the optimism of Google employees, the use of artificial intelligence for medical needs is still a challenge. Similar projects from other large companies, such as IBM, had difficulty integrating the technology into insurance companies' systems. Google has already faced challenges in gaining access to patient data (for this it had to enter into special agreements with several hospitals), as well as in creating a system that can work in several hospitals at once.
• Google does not yet know how it will monetize its new technology. Representatives of the company say that it is too early to talk about the business model. To get started, a company needs access to a huge amount of data. Perhaps in the future, Google will license the system in clinics or sell it through the company's cloud division as a "diagnostic service."

Medicine, with its millions of analyzes, extracts and medical records of patients over decades, is one of the most obvious areas for the application of new Big Data technologies, but there have been no breakthroughs in this area yet. It is predictable that such a breakthrough could be provided by a company like Google, which lives on processing the data of billions of people. But at the same time, this means that we must be prepared for the fact that in a few years Google (and in our case, Yandex or Mail.Ru Group with a delay of several more years) will gain access to medical data - and a potential scandal if they leaks will make the Cambridge Analytica story look like a childish prank.

Liana Faizova