Google Brain researchers Martín Abadi and David Andersen proved computers powered by artificial intelligence (AI) could keep secrets. In the past, Google’s AI composed a song, and now it can allow computers to protect messages from prying eyes.
In the experiment, the team of Google’s deep learning research project showed neural networks, also known as neural nets, could work without the need of humans. Computers made up an encryption without the developers’ intervention.
Compared to the systems currently designed by humans, the resulting encryption technique is simple. More impressive is the fact that neural nets are not meant to be great at cryptography. In fact, researchers didn’t teach computers cryptographic algorithms.
The Google Brain unit merely perfected the role of the three neural nets taking part in the communication experiment. They also gave them names: Alice, Bob and Eve. Their task was simple; Alice had to send Bob a message and keep the content secret from Eve.
Bob and Alice learned how to handle encrypted data on their own
Once Alice had sent the message, Bob’s job was to decode it while Eve tried to eavesdrop. To keep the message’s content a secret, Alice converted the plain-text to gobbledygook or cipher text. In the end, Even intercepted the message, but only Bob could read it.
Initially, Alice and Bob fared poorly in sending secret messages for Eve’s delight. However, they got amazingly better in time. Alice developed an encryption strategy, and Bob improved his decryption skills.
After 15,000 tries, Eve could decrypt just 8 of the 16 bits forming Alice’s message, but Bob was able to get all the original text.
YouTube uses Google Brain’s AI technology
The study, published on arXiv, is the first step to match the sophistication of encryption methods made by humans. Computing with neural nets is Google Brain’s new deep learning approach.
Google Senior Fellow Jeff Dean and Stanford University professor Andrew Ng founded Google Brain in 2011. They aimed to crack the central problems of AI by using deep learning techniques. Ng left Google Brain in 2014 to lead the AI group at Baidu.
Google Brain’s technology currently powers Android’s speech recognition system, Google+’s photo search, and video recommendations in YouTube. The Google Brain team is also building a deep learning software system called DistBelief and funding Google X project.
Source: Google Brain