Google's AI Taught Itself To Encrypt Messages

In the wake of the massive DDoS attack that brought down huge swaths of the internet last week, strengthening cybersecurity is on everyone's minds. It's tough to devise systems that are truly secure, because as security evolves, so do the people trying to hack those systems. But a new experiment from Google showed a possible way forward using artificial intelligence. Google created an encryption game using three distinct entities--Alice, Eve, and Bob--created by deep learning neural networks. Alice sent thousands of 16 character strings of zeros and ones to Bob, encrypting them with a shared key. Eve sat in the middle, trying to decrypt the messages.