Google engineers believe they have inadvertently created a language for Artificial Intelligence.
The unknown is scary, and for example, that’s why we don’t like two people speaking in their own language when we are part of the conversation; apart from being rude, of course.
Now imagine that instead of two people speaking their own language, there were two AIs; It doesn’t improve, does it? What could those computerized systems be talking about?
Machine learning to translate languages naturally
Google has come across this issue with the new neural network it uses to translate text from one language to another. This system is not based on a word database; instead of that, use machine learning to find out which translations are more natural and accurate.
While learning, AI has to check the meaning of a phrase with each supported language; for example, in an English, Japanese, and Korean translator, you have to translate a phrase from English to Japanese and Korean, and then vice versa.
As much machine learning as I have, they are still no less than 103 languages that the machine has to learn and translate; Even for Google, it is too costly a process in computing terms.
So the engineers had an interesting idea. If the AI knows how to translate from English to Korean, and from English to Japanese; Could only with these data translate from Korean to Japanese, even if you have never translated between both languages?
Makes sense. If you know the meaning of car (car in English), you translate it into Japanese () and Korean (), it stands to reason that you and they mean the same thing.
A language for Artificial Intelligence created by itself
And indeed, it was discovered that this was possible; than AI was able to translate between two languages that it had never translated between them. Only based on his knowledge of a third language.
However, this discovery soon revealed something even more exciting. Because for AI to translate between two languages without using English as an intermediary, it was necessary to have an internal concept with a similar meaning to that of the words in both languages.
In other words, to associate with is necessary to have an intermediate concept of what that word means, car. You may not know what a car is, but you know it’s a unique concept that has its own meaning regardless of the language you use.
This gif published by Google better explains this:
Based on the way words and phrases are related to each other in a neural network, Google believes that we can define this as a language for Artificial Intelligence; created by the AI itself and understandable only by it.
Will we soon see robots talking to each other in our presence, like rude?