One step closer to Skynet? Google’s AI built your own encryption – IDG.see

Alice, Bob and Eve. Sounds the nicer types than Terminatorfilmernas Skynet? But don’t be fooled by their maximum ordinary name. Of the three is actually the neural network that is entirely at your own hand succeeded in developing an encryption method.

There is a researcher from Google’s djupinlärningsprojekt, Google Brain, which allowed the three ai systems to beat her wise, albeit digital, heads together. We already know that the researchers "trained" artificial intelligence to the artists. But this is something else, and that can seem much more daunting.

also Read: Google’s robotkompositör has released his first song and it is not so good

The three ai members were allocated different tasks. Alice could send a secret message to Bob, which only he could take part of. Eve’s mission was to try to eavesdrop and decode the message. The whole experiment can be compared a bit with the three students who share their notes before an exam.

And the results shows that computers are really good at keeping a secret. The three have learned how they can protect messages from eavesdroppers who want to snoop. Without having to learn any specific encryption algorithm, could the computers develop their own encryption, with the help of machine learning, writes New Scientist.

Alice began by making a plain text to a perfect gibberish, which Bob was able to read by using a chiffernyckel. First it went a little si like that of the two, Eve can be listed quickly right out what they said. But after about 15 000 attempts succeeded Alice develop an encryption strategy that Bob could lock up without Eve picked up the information. Although the message is not more than 16 bits long, 1 and 0, so basically could Eve random guess to each part and still get a decent result. But it is still the first time, the neural network succeeds completely on their own.

also Read: Google’s new AI can teach itself up – without the help of mankind

Scientists do not know exactly how the encryption method works. Machine learning provides a solution, however, presents not an explanation. Which also means that they can not guarantee the security of an encryption created by artificial intelligence. Thus, also limits the uses.

Hence, may possibly troubled souls feel calm, for so long as the teach the machines do not start talking with each other in secret.