Researchers at Brigham Young University in Utah have created an algorithm designed to help AI and humans learn to cooperate. In their study, published in the journal Nature Communications, the team used 472 games requiring two-player interactions to promote “chit-chat” between human and machine. The researchers concluded that this type of banter (what you might call “cheap talk”) actually doubled the rate of cooperation between AI and human players.
One of the games used, the prisoner’s dilemma, was meant to be a sort of litmus test for successful cooperation in the study. The game requires both parties to decide whether or not to betray one another in order to avoid prison time. Both the AI and human players had to decide whether to betray the other or stay silent — as opposed to having each turn’s outcome depend on the other player. This particular game was an apt test for cooperation as both players stood to benefit from working together. The researchers assessed successful cooperation by considering the ability of the humans and AI to work together combined with their total final scores.
As they played, the “chit chat” between the players included the bot saying phrases such as “I accept your last proposal,” “Sweet, we are getting rich,” and “In your face!” The human players would then choose a response from a pre-made list, as opposed to ad-libbing.
As Jacob Crandall, one of the study’s lead researchers explained to New Scientist, just as humans might bluff their way through a game, the AI was also capable of being “all talk:” for example, it might say that it was going to choose betrayal in the prisoner’s dilemma but ultimately decide not to follow through with that choice.
A future depending on human-robot cooperation may not be that far off. In fact, for those who believe as Elon Musk does, we may not be far from a reality where they surpass us (and in a multitude of ways). Communication, in general, is an incredibly complex system — even between humans. In order to truly work well together, AI will need to be able to understand and respond to human emotions through verbal and facial cues, cultural practices, and social norms.
Achieving this level of cooperation will likely become a larger focus for AI development in the years to come. If chit-chat can help open up the lines of communication, AI’s ability to cooperate and compromise with humans will hopefully help alleviate anxieties over some kind of sci-fi inspired robot takeover.