According to New Scientist, Abadi and Andersen Dali Dali his tasks for each neural network: Alice had to send a secret message that only Bob could read, while Eve had to try to decipher this message and find out its content. The experiment began with a simple text message that Alice converted into an unreadable nonsense and which Bob could decipher using a special key. At first, Alice and Bob did not manage to hide her secrets, however, after 15 thousand attempts, Alice developed her own encryption strategy, and at the same time Bob developed his own strategy for deciphering. The length of the message was only 16 bits (each bit was either a unit or zero), so the fact that Eve managed to solve only half the message means that she either threw a coin or simply guessed.
Of course, the personification of these three neural networks simplifies everything a little. Due to the specifics of AI, even researchers do not know what kind of encryption method Alice used, so the experiment has not yet carried a lot of practical benefits. However, this is still a very interesting exercise that could be developed, for example, with the help of the recently released second version of the Freim of Deep Learning Cognitive Toolkit from Microsoft.