Researchers at Facebook recently shut down a pair of AI bots that were designed to communicate with each other in English, after they instead began using a new language by rearranging English words into seemingly gibberish sentences.
Only the sentences weren’t completely nonsensical, but actually comprised a coded language that made sense to the bots, scientists with Facebook AI Research (FAIR) told Fast Co. Design.
Here’s a sample of the conversation that transpired between the bots, dubbed “Bob” and “Alice”:
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
Bob: you i have everything else
Alice: balls have a ball to me to me to me to me to me to me to me to me
The conversation continues to carry on in a similar vein, according to a screen shot published by Fast Co. Design.
“Agents will drift off understandable language and invent code words for themselves,” FAIR researcher Dhruv Batra told the magazine. “Like, if I say ‘the’ five times, you interpret that to mean I want five copies of this item. This isn’t so different from the way communities of humans create shorthands.”
In other words, it’s similar to how specialized communities of humans, such as stockbrokers or sailors, develop their own dialects which are functional for their specific environments.
Batra added that the bots began developing the language because of a programming error which, in effect, gave them an incentive to develop a more efficient language.
The bots were created as part of a program that aims to teach machines how to negotiate, with a view to ultimately developing personalized digital assistants capable of communicating with humans, FAIR researchers said in a June blog post.
“Bob” and “Alice” were shut down not because it was feared that they were plotting to overthrow humans and take over the world, but rather because FAIR researchers want to develop bots capable of talking to people, research scientist Mike Lewis told Fast Co. Design.