Cracking the code of the Enigma machine with Artificial Intelligence

Thanks to the 2000 servers DigitalOcean and the programs that use Artificial Intelligence, engineers of the company Enigma Pattern, cracked in 13 minutes what took him years to Turing and his team at Bletchley Park in the Second Great War. And in addition, the cost of this was only 7 dollars.

Hollywood used the AI to predict the success or failure of his productions

On the 23rd of June was the anniversary of the birth of Turing, exactly, would have turned 106 years. And due to the importance of the work of the british scientist, much of what is done is done in his honor. For example, the fact that a replica of the Bombe, one of the first machines that were used to help decrypt the German messages from the Enigma machine, he moved to the National Museum of British Computing., located in Bletchley Park, and in addition, is located to the side of the first computer rebuilt, the Colossus.

For those who ignore the issue of cryptography and its relationship with Turing, to say that he and a team of collaborators in poland, they worked intensively to try and find out what they were saying the encrypted messages from the Germany of Hitler. For this work, which involved mathematicians of the highest order, were 10 Colossus and 210 Bombes. But now the Enigma Pattern used AI and the computational capacity current to find the solution that I longed to Turing in record time.

Enigma Pattern began using AI and Machine Learning to help companies that collect a lot of data, what is called Big Data. Considering the “youth” of the AI as a science, the company decided to use its resources, for example, 20 percent of the time available, in projects of any way to use the AI of less conventional ways.

Lukasz Kuncewicz, one of the co-founders of Enigma Pattern, he decided to return to walk where Turing walked, and decided to apply the IA modern, undertook to break the impossible to break, the encryption of the German Enigma machine, used by the nazis to encode their messages in one of the darkest moments of humanity, the Second World War.

According to the co-founder of Enigma Pattern, Rafal Janczyk, the project began with the question: “What would you have been able to make Alan Turing if you had the computational power of current and the development that we have now of the IA?”

It was then decided to recreate the version of a naval Enigma Machine, which was more sophisticated and gave him something like 15 billion possible passwords. Enigma Pattern began to recreate the machine, the rotors and connectors, all through Python. Initially she tried to teach the AI to decode the code nazi by itself, but it didn’t work this approach. Neither Lambda functions Amazon worked due to the amount of computation required and to the limitations of the servers of Amazon (AWS) in execution times. But that all changed when DigitalOcean agreed to support the project.

The code of all this work is on GitHub, including the simulator of the Enigma machine, which was used to train the machine of deep learning to classify the German language. This was done using Keras on TensorFlow with Grimmy’s Fairy tales as the source of a simple vocabulary. The researchers used a neural network of two layers and trained then the network using 10-letter German against 10 random letters.

In September 2017 the researchers commented: “Remember when we ran the whole project for the first time… After a couple of minutes things were made apparent. The good news is that the project worked well. The simulator of the Enigma was testing the combinations and the AI I was sorting out the messages decrypted. The bad news is that it took about 2 weeks to find the passwords.”

The solution then was to distribute the work in a multitude of virtual servers, and using the resources that gave them DigitalOcean, the time was reduced to 19 minutes, with 13 million combinations tested per second. Later, in the same 2017, the number of resources of virtual servers to be doubled to 2000, which could prove 41 million passwords per second. The result was to be able to decrypt a message at 13 minutes. The best thing is that according to DigitalOcean, the real cost of all this was only 7 dollars of machine time.

According to Janczyk: “With AI and Machine Learning, there are an unlimited number of possibilities, which makes everything is now so exciting…”.

Related posts

Leave a Reply

Be the First to Comment!

Notify of