Fortifying cybersecurity is on everyone’s mind after the massive DDoS attack from last week. However, it’s not an easy task as the number of hackers evolves the same as security. What if your machine can learn how to protect itself from prying eyes? Researchers from Google Brain, Google’s deep Learning project, has shown that neural networks can learn to create their own form of encryption.
According to a research paper, Martín Abadi and David Andersen assigned Google’s AI to work out how to use a simple encryption technique. Using machine learning, those machines could easily create their own form of encrypted message, though they didn’t learn specific cryptographic algorithms. Albeit, compared to the current human-designed system, that was pretty basic, but an interesting step for neural networks.
Meet Alice, Bob and Eve
To find out whether artificial intelligence could learn to encrypt on its own or not, the Google Brain team built an encryption game with its three different entities: Alice, Bob and Eve, powered by deep learning neural networks. Alice’s task was to send an encrypted message to Bob, Bob’s task was to decode that message, and Eve’s job was to figure out how to eavesdrop and decode the message Alice sent herself.
The game started with a plain text message, which Alice had to convert into complete gibberish so that no one but Bob could decipher. In order to help encrypt and decrypt the message, both Alice and Bob started with a pre-agreed set of numbers known as a key that Eve didn’t have access to.
Perfect as practiced
Practice makes everyone perfect, even an AI. Initially, both Alice and Bob were apparently poor hiding their secrets, but over the course of 15,000 attempts, both Alice and Bob could work out on their encryption and decryption strategy, making Eve guess just 8 of the 16 bits forming the message.
The practical implications of this could be limited as it’s still unknown exactly how the encryption method works. However, the success of the experiment shows that deep learning neural networks could be vital in the world of cybersecurity that is constantly shifting.