It sounds like science fiction, but it’s not: Facebook stopped an experiment on artificial intelligence, during which some bots would start communicating with each other using a language incomprehensible to humans. The word “bot” (short for “robot”) means a program that on social networks makes the user believe he is communicating with a human person: in this way, the machine is able to answer and satisfy the user’s doubts.
In recent months, Facebook developers have been working on, among other things, some artificial intelligence (AI) systems capable of talking to each other in English and conducting simple negotiations, for example, to divide two books, three basketballs and a cowboy hat. At one point they witnessed an unexpected phenomenon: the two artificial intelligence systems started using a new language, invented by them. This alarmed the researchers and led them to stop the experiment.
The conversation between the two bots appeared to the developers this way:
Below an approximate translation
Bob: I can everything else. . . . . . . . . . . . . .
Alice: balls have zero for me for me for me for me for me for me for me
Bob: you are everything else. . . . . . . . . . . . . .
Alice: the balls have a ball for me for me for me for me for me for me
I can everything else. . . . . . . . . . . . . .
Alice: the balls have a ball for me for me for me for me for me for me
Bob: me. . . . . . . . . . . . . . . . . . .
Alice: balls have zero for me for me for me for me for me for me for me
Bob: you are me, I am everything else. . . . . . . . . . . . . .
Alice: balls have 0 for me for me for me for me for me for me for me
Bob: you are everything else. . . . . . . . . . . . . .
Alice: balls have zero for me for me for me for me for me for me for me
What worried the researchers who carried out the test was the fact that the language used by the bots, despite not having a complete meaning for human beings, does not seem to be casual: the words were used by artificial intelligences in a different way. An explanation for the abnormal behavior could be attributed to the fact that the bots were not programmed to speak only in English. But there is a suspicion that the machines follow codes that are still little known to man.
The news does nothing more than confirm fears about the uncontrolled development of artificial intelligence: in fact, will machines one day become so powerful and evolved as to rebel against human commands? That’s what even experts of the caliber of Stephen Hawking and Elon Musk fear.
“Our interest is that bots can one day talk to people, meet their demands,” explained Mike Lewis of Facebook’s AI research program. Using algorithms, bots refine their skills and are left alone to confront and develop language and negotiation skills. It is possible that, during the experiment, at some point the bots changed their language, having found a communication system with which they could conduct their negotiations more effectively.
Anyway, the episode sheds light on a little-known and almost always mysterious world: that of the robots behind social networks and the pages we consult every day. Mark Wilson, from the design and technology website FastCo Design, wrote an article entitled “Artificial intelligence is inventing languages ​​that humans cannot understand. Should we stop it?”: “We should let our software do this “- we read in the text – Should we allow AI to build its own dialects to perform specific tasks that require communication with another AI? Gossiping when we can’t understand them? Perhaps. It could give us the possibility of a world where iPhones talk to refrigerators that talk to cars without a second thought.”
According to Dhruv Batra, a Facebook researcher who followed the case, there is much ado about nothing: on his Facebook page, he invited users to consult the actual research on the subject and not to be misled by the articles. The study in the field of artificial intelligence through the social network certainly does not stop because of this episode: “I do not want to link to specific articles or provide specific answers for fear of continuing this cycle of sentences extrapolated from context, in any case that the way the media’s handling of the issue is irresponsible and committed to clickbait, while the idea of ​​machines inventing their own language might seem alarming or surprising to some people outside the field, it’s a pretty solid artificial intelligence industry, which has publications and years of research behind it. Simply put, some bots that are asked to perform a task often find unintuitive ways to achieve the result. Changing experiment parameters does NOT mean ‘stopping artificial intelligence’ or turning it off, as written.
If this were the case, every developer in this field would have to drop out every time a job doesn’t reach the goal.”