Facebook robots invent their own language.

If the notion of artificial intelligence was not yet frightening you, this could change with this news: the research laboratory on the artificial intelligence of the giant Facebook has disabled two robots after the latter had a somewhat strange conversation.

Facebook has for some time been planning to develop robots integrating an I.A with the ability to exchange and negotiate the sharing of objects, the two "incriminated" robots, named Alice and Bob, had the mission of obtaining books, balloons or hats. The discussions observed by the researchers of the research laboratory on artificial intelligence of the social network took an unexpected turn.

Indeed, both IAs used incorrect English instead of expressing themselves in a classical way and also didn't give figures as part of the exchange in order to divide objects, they had indeed a list of objects as well as indications on the value of the goods according to them, for example a balloon which is worth three points for one, but only two for the other, as the following excerpt shows, robots have gleefully reiterated personal pronouns as well as definite articles, a dialogue whose comprehension is frankly not obvious and in no way resembles the expected negotiation.

The researchers practiced this test to allow robots to talk with humans. Nevertheless here, there seems to be no rule at speech level although chatbots insist on their name in this negotiation process. Even more strange, they managed to come to an agreement in some negotiations conducted in that language they invented. According to the researchers, they were therefore deactivated not because they had created a language that was not very comprehensible, or because it was dangerous for humans, but because it was not the expected result.

According to Dhruv Batra, researcher in Facebook, the explanation for this dysfunction is simple: "There is no reward to be held only in the English language. The agents are therefore derived from comprehensible language to invent code names. For example, if I say five times "the", you interpret it by understanding that I want five copies. This is in no way different from the communities of humans who create abbreviations "

Artificial intelligence fascinates as much as it inspires fear and people think that this is a technology that is not without danger there will provide further proof of their opinion. Are they wrong? The future will tell us, remains that if IA is sold to us as a way to help the human in everyday life, worrying a little and not accepting this technology with eyes closed is a pledge of caution.