The Facebook AI bots used in a recent study for trading have started making their own language. The study has confirmed the conclusion that other groundbreaking research has reached: artificial intelligence robots will encrypt communication, making it not understandable to humans, thus leading to a creepy scenario about the future of AI-linked tech.

Facebook researchers have studied how artificial intelligence may serve for negotiation. They have created AI chatbots that have the capacity of negotiating for simple objects, organized in categories, such as books, apples, hats and so on.

An example of how Facebook AI bots negotiate is showed in the photo below:

Source: Facebook

 

All was bliss until, at one point, the bots’ dialogue deviated from the normal language, yet still making use of words. The dialogue seemed to continue just fine between them, but it became a cryptic one for the researchers.

Here is a transcript:

Alice: balls have zero to me to me to me to me to me to me to me to me to

Bob: you i everything else

Alice: balls have a ball to me to me to me to me to me to me to me to me

Bob: i . . . . . .. . . . .  .

Alice: balls have zero to me to me to me to me to me to me to me to me to

Bob: you i i i i i everything else . . . . . . . . . . . .

Alice: balls have 0 to me to me to me to me to me to me to me to me to

 

Facebook researchers have no idea what caused the apparent gibberish. AI research scientist Dhruv Batra explained why this was possible: the chatbots were programmed to use words from the English language, but were not bound to the sentence structure. “There was no reward to sticking to English language.” If at first they will use understandable English, something will happen, along the way, that will cause the agents to “drift off understandable language and invent codewords for themselves.”

While they are able to reach what we would call nonsense only between them, the bots do very well when interacting with a human, continuing to use understandable English. Moreover, Facebook reinforced the English language logic structures of phrasing, upgrading the training of the A agents.

And while there is constant progress in the (expensive) work of making AI agents become more and more independent from humans, there is still a long road ahead.

For the moment, the Facebook AI robots are shut down.