You don’t know yet, but Alice, Bob and Eve will change not only the way in which you work, but also might be the one you live daily.
Evolution of the errors of Bob and Eve during their training. The curves represent the average errors. Past 15000 trials, errors of understanding of Bob, are null and void. Eve of his side includes only 8 of the 16 bits transmitted by Alice. (Source : Google)
Generate the encryption scalable
In an article of search, Google has developed a scenario where the neural networks manage to generate their own encryption, scalable, on the basis of artificial intelligence and machine learning. Concretely, Alice and Bob, two neural networks in the species, are evolving their own encryption tools, leaving Eve in the dark. Yes, because Eve is also a neural network.
the three classical figures of cryptology are entities used
by the researchers to test security protocols and make
and advance research (RSA encryption, dating back to 1970 was
advanced on this basis). Their role ? Alice and Bob are trying to
communicate in a secure manner, understand it without anyone to share them
are aware of what they are saying. Eve as she tries
to listen to Alice and Bob by any means.
The two researchers of the project Google Brain (launched in 2011), Martin Abadi and David Andersen, have published a study in which they show that the intelligence artificial using neural networks (algorithms whose operation is inspired by the neural networks (biological) can create their own encryption system.
The tour-de-force of researchers is mentioned on page 3 of the study. “Instead of leading to Alice and Bob separately to implement a cryptosystem known, we train Alice and Bob jointly to communicate successfully and to defeat Eve, and this without any notion of a predetermined of the cryptosystem they need to use for this purpose.” Concretely, the two machines should not apply a pre-defined plan, but to work together to find the resources and means necessary to not be understood by Eve.
Who is going to open Pandora’s box ?
Problem raised by the study, entitled “how to protect the communications through cryptography, neural” : if Eve did to be more successful over time, tests / errors of the machine learning, to understand what to say Alice and Bob, the researchers either. Cryptography invented by the two entities is no longer decipherable by Martin Abadi and David Andersen.
what scare Jacques Attali that the Express train sounds the alarm : the machine is released from the hand of humans, paving the way for the autonomy of the machines, and scenarios to the Terminator. “These intelligences (…) can first consider that their mission is to counter those of our behaviors which, in their eyes, we would be detrimental ; they may as well think that humanity move to its loss and to decide, for his own good, to stop it in time. But they can then also act against humanity, for their own benefit ; especially if they think that, eventually, we will have interest to destroy them : they can fight us, in self-defence” – Jacques Attali.
Machine learning requires learning the process from Bob to understand the said Alice has been a very long and tedious. It took 15 000 trials to Bob to understand the 16 bits of the message sent by Alice, then Eve would get half of what was said Bob with the same number of trials.
“We demonstrate that networks of neurons can learn to protect communications” refer to, in conclusion the two researchers. “Learning does not require specification of cryptographic algorithms individuals, or to indicate the ways and means to implement these algorithms : it is based only on a specification of the secret represented by the objectives of training”.
No comments:
Post a Comment