Facebook kills bots after they create their own language

RIP Alice and Bob, you freaky literary heroes

Pin It
sophia bot

Two chatbots created in an experiment by Facebook appeared to have created a bizarre new language, and they got into some convos that no one else could understand.

The social network had been experimenting with its bots, which spoke with humans onsite, as well as each other. They were given a task to negotiate with each other and attempt to swap items of similar value, like books. However, they weren’t instructed by programmers to use ‘comprehensible English’, so they began creating a shorthand that descended into what looks like madness.

Google also had similar developments with its translation software. "The network must be encoding something about the semantics of the sentence" Google detailed in a blog post. Another project, OpenAI, is creating AI that makes and converses in its own language to help with its problem-solving abilities. We all must remember that disaster teen chatbot created by Microsoft, @TayTweets that became a horrible racist after learning from its interactions with site users. 

“Agents will drift off understandable language and invent codewords for themselves,” Facebook Artificial Intelligence Research division's visiting researcher Dhruv Batra told the Independent. “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.”

The AI convo between Alice and Bob goes as such below:

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 everything else . . . . . . . . . . . . .

Alice: balls have a ball to me to me to me to me to me to me to me

Bob: i i can i i i everything else . . . . . . . . . . . . . .

Alice: balls have a ball to me to me to me to me to me to me to me

Bob: i . . . . . . . . . . . . . . . . . . .

Alice: balls have zero to me to me to me to me to me to me to me to me to

Bob: you i i i i i everything else . . . . . . . . . . . . . .

Alice: balls have 0 to me to me to me to me to me to me to me to me to

Bob: you i 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

So: not much in the way of trading, a little odd, sometimes a bit hot (?).

Recently, tech entrepreneur Elon Musk and Facebook’s Mark Zuckerberg got into a (very nerdy) war of words on the future of AI. Musk had previously claimed that our ventures into artificial intelligence and robotics could put “humanity at risk”, but Zuckerberg rebuffed such claims, highlighting its potential to save lives and alleviate workloads.

There’s no need to worry about an impending robot revolution that subordinates humanity yet though: the AIs weren’t given an opportunity to conspire against us, and as the BBC reports, it’s most likely the interactions shown above came about because modifications to language used meant interacting was quicker and easier for the bots.

Facebook’s experiment prior to this showed that AI can pick up some human-like techniques: in the monitored times Alice and Bob tried to barter, they did things like pretending to be interested in an object, so they could feign making a sacrifice when giving it up, a paper by FAIR detailed previously.

The development of such a language isn’t really something researchers at Facebook were totally interested in – this experiment didn’t do what they wanted, so they just shut it down. They weren’t frightened of a rebellion, rather, just not bothered with the AI’s seemingly nonsensical chat. They want the chat applications to successfully chat to humans. Bots throughout its AI studies haven’t fared too well – its bots have a 70 percent failure rate when it comes to speaking with other human users. The social network has scaled back its bot projects on the likes of FB messenger.

Back to the drawing board, robot rebels.

More Science & Tech

Like this?
Like Dazed on Facebook