We've seen how machines got smarter. From self-driving cars to drones delivering your packages right to your doorsteps, the evolution of technology is now paving a new and digitized way towards the future. At the forefront in this quest is Google, and their latest breakthrough is nothing short of science fiction.
Recently, a team from Google Brain is now at the helm in developing "human-independent" AI, or artificial intelligence with the ability to form encryptions and at the same time, protect it from intruders. According to New Scientist, Google Brain researchers Martin Abadi and David Andersen are parents to "neural nets" Alice, Bob and Eve.
They are described to be information gathering increments modeled to act like a human brain, especially the way it solves problems and process information. As per their experiment, the neural nets were able to formulate "their own form of encryption" without Google researchers teaching them a certain cryptographic algorithm. What Alice, Bob and Eve did was to learn everything from "machine learning."
Tech Times on the other hand, reported that Google's Abadi and Andersen assigned specific tasks for each neural net. Alice was tasked to send a specific message that only Bob could decipher. Eve meanwhile needed to decode that message. As an added measure, Alice converted the message into a "cypher text" to refrain Eve from interpreting, but it had to be decipherable to Bob. It took time for the neural nets to relay the message, but after 150,000 tries, the results were astonishing.
Bob was able to decode Alice's "cypher text" and even converted it to plain text. As for Eve, she was only able to decode 8 out of the 16 bits that formed the message, with each bit represented by a 1 or a 0. Here's where the tricky part comes in: the Google Brain researchers have no idea how the encryption works. Machine learning only provides them with the solution, but how it was achieved is something that is way above their pay grade.
There's also a fact about security should this practice would be observed this early in real life. AI is a feat on its own, but it is still not ready to create its own encryption. But speaking from the vantage point of Alice, Bob and Eve, anything's possible when things are in place. Google can and will continue to find new ways to improve neural nets and other forms of AI. It's only a matter of time that these Google-clad programs are as common as the Search site itself.
