What happens if you order two smart systems to communicate with each other in an encrypted – secret – way and you tell a third party to try to break the encryption key? This is what has happened in Google's department, Google Brain, whose researchers Martín Abadi and David Andersen have succeeded in teaching machines to create their own encrypted message without human intervention in the process . They have also found that computers are better at encrypting than decrypting. In their study "Learning to Protect Communications with Adverse Neural Cryptography," they describe how two neural networks named Bob and Alice share a secret key. And on the other hand, a third neural network called Eve was assigned the mission of intercepting the information transmitted between the first two and trying to decipher it.
Alice and Bob passed the test successfully as they developed a very advanced encryption system even by human standards . And it is that Eve was unable to break the code created by Alice and Bob. In any case, Eve did not do so badly, as she provided numerous indications of a great improvement in the development of the decryption algorithm.
And that's despite the fact that both Alice and Bob, as Eve tried to break the encryption system, was continually improving the system. Research shows that, like humans, neural networks work better at encrypting and encrypting information than at deciphering secret codes, which makes sense, since neural networks are created in the image and likeness of our brains.
