Facebook researchers taught bots to negotiate (and lie) like humans

Facebook just took a big step forward in its quest to make its automated bots more like humans.

The company's AI researchers have taught the bots how to negotiate. In research published today, Facebook's Artificial Intelligence Researchers (FAIR -- yes, that's really what they call themselves) detail the "breakthrough" and how the bots ended up acting more human than they expected.

The researchers trained the bots with examples of actual human negotiations where two people were tasked with dividing up a series of objects, which had different values to each side. The scientists then used this data to teach the bots how to negotiate. (You can see an example in the GIF below.)

One thing that was unique about FAIR's approach was that their goal was to get the bots to actually act like humans, training them to anticipate what a human would do in any given situation, rather than simply teaching them to imitate their behavior. And they were ultimately successful. When they put the traivned bots up against humans, the people were unable to detect that they were interacting with a bot.

"The best bots overall had comparable abilities to humans," says FAIR research scientist Mike Lewis, though he notes that "at least some of their advantage is in their infinite patience, and determination."

Moreover, the bots were also to learn what the researchers call "intelligent maneuvers. "There were cases where agents initially feigned interest in a valueless item, only to later 'compromise' by conceding it — an effective negotiating tactic that people use regularly," FAIR wrote in a blog post. In other words: the bots learned to lie in order to get a more favorable result.

Lewis notes that this wasn't a negotiating tactic the researchers taught them, but one the bots picked up by learning from human negotiators.

The researchers have open-sourced the code behind the bots so any bot developer may take advantage of the new skills. Ultimately, though, they say the advancement will help improve digital assistants so they're better able to complete tasks for us.

"You can imagine a world where this accelerates and eases interactions -- for instance, future scenarios where people might use chat bots for retail customer support or scheduling – where bots can engage in seamless conversations and back-and-forth, human-like negotiations with other bots or people on our behalf," Lewis says.