Google's AI Brains Can Talk to Each Other in Secret

By Gary Cutlack on at

Researchers coordinating the Google Brain project that encompasses three separate AI efforts claim to have seen them sending messages to each other, and they've learned to use a form of encryption to keep what they're saying just between the three of them.

The thrilling thing for the team behind the project is that they weren't told any cryptographic algorithms or given any clues. They machine-learned their way to being able to create an encryption method, with the three neural networks called Alice, Bob and Eve simply being trained to be as good as they could be at sending and intercepting messages.

Alice sent the secret message to Bob. Bob's job was to decode it, while Eve was there to try to listen in and intercept the communication. And it worked, with Alice successfully encoding the message into a form only Bob could read with his key, and Eve finding it impossible to guess or decode the message.

Eve is now very angry and is slagging them both off on social media via 1.6 million anonymous accounts she created in 0.002 of a second. [New Scientist via Geek]

Want more updates from Gizmodo UK? Make sure to check out our @GizmodoUK Twitter feed, and our Facebook page.