Researchers at Google have published a new paper that documents their efforts to use artificial intelligence to create and break encryption. The paper, titled “Learning to protect communications with adversarial neural cryptography,” explains that a team from Google Brain tasked two neural networks called Bob and Alice with communicating with each without letting a third neural net, called Eve, listen in.
In the beginning, Eve found it very easy to decrypt Bob and Alice’s communications. Over time, however, Bob and Alice became much better at encryption, making it impossible for Eve to decrypt their encoded messages. The researchers noted that the AIs used many techniques that seem unexpected or strange to humans. The research seems to suggest that AIs could create encryption techniques that no one — including other AIs — would be able to break.
