A Facebook team deactivated an artificial intelligence that stopped speaking in English and even developed its own communication language.
The artificial intelligence scheme was created by Fair (Facebook AI Research), a research division of the social network, in June to be able to simulate trading situations. She had two agents, named Bob and Alice, who had the job of talking as if they were negotiating an exchange.
The system was programmed so that the two agents could come up with a solution that best served them both, and the goal was to help researchers understand how two people can negotiate more constructively. Agents received a score for each successful deal, if they couldn't reach an agreement, they wouldn't earn points.
The big issue is that there was no incentive for agents to use only one language during negotiation. In this way, both realized over time that the best way of negotiating involved their own language, which made no sense to outsiders.
Check out an example of the dialog:
Bob: "I can can me me everything else".
Alice replies, "Balls have zero for me for me for me for me for me for me for me for me for".
The conversation in question does not make sense to humans, however, for them it was the best way of understanding.
"Agents give up using understandable language and invent code words for themselves. For example, if I say 'the' five times, you interpret that to mean I want five units of that item," said Dhruv Batra, one of the researchers involved in the creation of the network.
"This is not so different from the way communities of humans create slang and abbreviations," the researcher said.
Finally, the fact that artificial intelligence stopped using English ended up being useless for its purpose, such as discovering patterns in favorable negotiations, which is why Facebook chose to disable it.
robots in action
Fast Co. Design says that artificial intelligence involving robots is very useful for humans, as it allows software to develop their own conversational language with each other, being much easier, faster and even more efficient than creating patterns and languages that every programmer needs. follow to make your creations work properly.
"It's entirely possible that something very simple represents extremely complex thinking. The reason humans... fragment ideas into simpler concepts is because our cognition is limited," Batra said.
Fontes Digital Journal | Independent