Google’s artificial intelligence is so smart it has developed its own method of encryption, according to a recent report from New Scientist. Google developers, Martín Abadi and David G. Andersen, detailed how three neural networks came up with a way to keep secret codes from one another in a new research paper.
In its most simple explanation, the developers observed how three neural networks named Alice, Bob and Eve reacted to a task of sending notes to one another in secret. Alice was given a worded message note to share with Bob, and Eve was tasked with trying to decipher the code without assistance.
In the beginning, Eve was easily able to decipher the message. But over 15,000 attempts, Alice developed her own encryption, taking the plain-text message to a 16-digit code comprised of 1s or 0s. Bob, in turn, was able to figure out how to decrypt the code and get to the original message.
However, with the encryption in place, Eve was eventually unable to decipher the message and decrypted only half of it by guessing.
Google is currently using its AI algorithms on products and services including the Google Allo chat app, the Google Duo video chat app, Google Home and the Google Pixel phones, among many others. The artificial intelligence allows users to communicate with their machines in a more contextual fashion, similar to communicating with another human.
The study is an attempt to demonstrate how machines learn without human interaction. The developers are supposedly unaware of what kind of encryption Alice created. This experiment is not expected to be translated into everyday use.
