Facebook's AI Created Its Own Language, so Its Creators Shut It Down
Facebook's newest artificial intelligence (AI), which is capable of communicating and negotiating with other AI, has stopped speaking English in favor of something that at first seemed completely unintelligible. But this robot isn't defective; its language capabilities aren't malfunctioning. In fact, this AI has, with the other AI, created its own language. Researchers shared this interaction as an example:
Bob (AI1): "I can i i everything else"
Alice (AI2): "balls have zero to me to me to me…"
It was reported that the rest of the conversation continued with variations of these two sentences. This "language" doesn't seem great—it comes off as badly interpreted, broken English. But it has, according to these researchers, a much more advanced explanation. The researchers think that this exchange is of two AI figuring out how many of a given item each should take. They use repetition (like with "i" and "to me") to condense information in sentences in a manner that the researchers, after analysis, described as highly logical, more so than many English phrases.
In an interview with Fast Co. Design, one of Facebook's AI researchers, Druv Batra, said that "Agents will drift off from understandable language and invent code-words for themselves...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."
Image Credit: Seanbatty / Pixabay
The AI lab founded by Elon Musk, OpenAI, has also observed this "shorthand" and is allowing its AI to continue to develop their own languages, so far successfully. But this research team at Facebook isn't as keen on continuing this type of development. In fact, they shut down the system in which these AI were functioning, afraid that it could develop out of their control.
As others make progress with AI developing their own languages, it is not overly dramatic that these researchers thought to end the program because of such development. It is a main goal of AI research to create systems that can modify and better themselves. But in creating its own language, one that will likely only get more complex and difficult for humans to interact with, other concerns are raised. By creating and using a language seemingly more logical than English ( according to current observations), these AI could theoretically become advanced enough that the researchers who created them might no longer have any control over them. This is a classic sci-fi storyline, but it is also becoming more and more of a possibility.