Google Brain has created two artificial intelligence (AI) neural networks that developed their own system for encrypting messages, beating a third AI system attempting to crack the code and intercept their communications.
The experiment was largely successful, although the two AIs often failed to decrypt the messages sent between them initially. But over time as their algorithms evolved, they were able to communicate with mostly positive results.
Meanwhile, the third AI was sometimes able to intercept the messages by random guessing but the other two AIs were able to make improvements as they evolved and eventually were mostly able to lock out the third-party AI.
"Neural networks can indeed learn to protect their communications, just by telling [them] to value secrecy above all else - and importantly, that secrecy can be obtained without prescribing a certain set of cryptographic algorithms," according to a report in Ars Technica.