Facebook’s researchers have shut down an artificial intelligence (AI) program after it was discovered it had created its own language, which means the bots no longer served their role of effectively talking to people.
For decades there’s been a dream about artificial intelligence advancing beyond basic tasks, but when this happens, it’s apparently quite terrifying to some.
According to reports from the Digital Journal, the system managed to develop code words to make communication more efficient between bots, but not with the humans that would, supposedly, eventually chat with them. When researchers figured out the AI was no longer using English, as it was programmed to do, they decided to pull the plug.
At first glance, this sounds a bit more terrifying than it actually is. It makes sense on some level that an AI will work to create its own language if it deems English to be redundant. According to exchanges revealed by Facebook, the two negotiation bots, called Bob and Alice, had this conversation.
“Bob: I can i i everything else.
Alice: Balls have zero to me to me to me to me to me to me to me to me to.”
While this seems like a nonsense to us, it made plenty of sense to the AIs, who figured out how to communicate how many items they should both get in the exchange. Bob makes an offer to Alice and Alice replies, perfectly understanding everything.
The whole AI was created with a reward principle, meaning that Facebook’s scientists programmed them that at the end of every dialog, the agent was given a reward based on the deal it agreed on. The two bots mentioned above were supposed to communicate according to the English rules we all apply, but, in their attempt to learn more from one another, they ended up with this mumbled dialect.
So, Facebook shut them down because they were no longer properly doing their job. Think of it as an employer firing an employee because of poor activity.
While Facebook didn’t shut down the AI because it was scaring them by developing their own language, it’s still a rather interesting situation.
Nothing to worry about…for now
Many have become scared of the idea of an AI developing a language of its own, one its creators could not understand, the situation would only be dire if someone programmed an AI to not follow any language protocols. Facebook’s AI was taught to use English, and it did, even if it cut down some words it found unnecessary.
Elon Musk has long been warning the world that there needs to be loads of regulation regarding AI development for fear of these creations going rogue. Mark Zuckerberg has dismissed his concerns over the years, saying that his stance was actually irresponsible as it had a negative effect both on the view people had of AI and on its creation.
The best stance, of course, is somewhere in the middle. We shouldn’t be too scared of AIs, but we shouldn’t not be scared either. Artificial intelligence doesn’t have the moral and social constraints humans do and their actions, if left unchecked, could reflect exactly that. Of course, we’re only just beginning to really test out what AI can do and how far they can evolve. It’s clear that limitations need to be imposed through their programming and checks need to be made, but Facebook’s bots are not the first, nor will they be the last to learn how to communicate more efficiently than our language permits.