What happened with the robots at Facebook?

Did two artificial intelligences really invent their own language?
08 August 2017

humans and robots.jpg

humans and robots



What happened with the robots at Facebook?


Peter Clarke, founder of Resurgo Genetics, fills us in on the recent news story...

Peter - They were training these AIs to negotiate with each other and what they were trying to do was to model the ways humans negotiate with each other. So they were playing them lots, and lots, and lots of human interactions and trying to get them to learn the ways humans decide on what they're going to say. What happened was that they then set them free and got them to negotiate with each other and they ended up taking the human language and developing their own language on top of that apparently, according to the stories. But really, it was a very sort of simplistic thing, which has been seen many times before in computer science research where you get two competing artificial intelligences that are negotiating with each other that they do develop these ways of simple interaction.

But it really wasn’t anything very scary and I can read out some of their…

Chris - Can you understand it?

Peter - I’ll just read you out one snippet of conversation:

Bob says to Alice -  I can I I everything else. And Alice says - balls have zero to me, to me, to me, to me, to me, to me, to me, too.

So that was one of their exchanges. Actually, what was happening is that they were transmitting numbers with little english phrases. So the to me, to me, to me, to me..

Chris - Ah. So those are numbers?

Peter - So that we them working out just an almost random way of taking the words that humans…

Chris - So it isn’t just gibberish, there is meaning in there?

Peter - No. There is meaning. They basically have evolved a way of taking words and just happening to use those words as a way of communicating relatively simple things to each other.

Simon - Where they also trying to work out the relative value of what they were negotiating for?

Peter - Yes, I think so. They were actually in a competitive situation.

Simon - Because, obviously, a lot of human negotiation is about mostly misdirecting people if you’re trying to get the better part of a negotiation.

Peter - Yeah.

Simon - If you’re trying to sell your car you always want to get the best price.

Chris - Do you think we could get them involved in the Brexit negotiations - would that work?

Simon - With language like that - maybe!

Chris - It seems it’d make more progress wouldn’t it. Elon Musk I think went on the record. He’s the guy who said Paypal and other ventures isn’t he. He said I think Mark Zuckerberg, Chief Executive of Facebook’s understanding of AI is “limited.”

Peter - I think this was something slightly different in that Elon Musk had really been coming out and saying, as he has for a while, really, highlighting some of the very real and significant dangers of this technology, and really calling for regulation. This is, I guess, where the serious point comes was that I think, to some extent, maybe Mark Zuckerberg was pushing back against that. It’s this debate about the level at which you’re regulating these technologies given that they can be, potentially, so powerful in the future, and I think that was the more interesting debate.


Add a comment