There’s an old saying that goes “when you make plans, the Universe laughs”.

When it comes to life, the notion that we can predict anything with any real air of certainty is an exercise in futility. And when it comes to people, well - forget it. We are a wily species, prone to fits of surprise behaviors and capable at all times of reacting in the manner that was least expected. So imagine for a moment that it is your job to teach robots the impossible - how to predict human behavior. What tools or techniques, what amount of schooling or education, could possibly prepare artificial intelligence to accurately predict how a human will respond to the millions of situations and stimuli that they encounter on a daily basis?

A team of researchers at Stanford University think they’ve found the answer, and have taken a decidedly novel approach to robot learning. Their new A.I., which has been dubbed ‘Augur’, provides the vector machines (basically, learning algorithms) with access to the hundreds of thousands of stories available in the online writing community Wattpad. According to the researchers, “Over many millions of worlds, these mundane patterns [of people’s reactions] are far more common than their dramatic counterparts. Characters in modern fiction turn on the lights after entering rooms; they react to compliments by blushing; they do not answer their phones when they are in meeting.” The theory is that while humans may react in any number of ways to a limitless number of situations, there are a certain set of expected responses to common situations that can be learned over time and through repetition. Works of modern Fiction especially offer a broad depiction of human observance and behavior; most especially the boring, mundane and routine aspects of daily life which encompass the majority of the learned behaviors which scientists are seeking to teach the AI models.

Virtual assistants like Apple’s Siri and Alexa, the virtual voice of Amazon’s Echo, are prime examples of technology that rely on learning the behaviors and associated responses of the humans they serve. Initial tests of Augur show that the system already has a 71% success rate for unsupervised prediction of what a user will do next. What’s more, 96% of the time it can accurately recall or identify human events. Considering the endless possibilities that life offers, those numbers are pretty stellar for a technology in its relative infancy. A few other major players are likewise using books to teach their computers, including one of the leading research entities into the world of AI, Facebook. Just last week the social media platform and AI champion announced that they had released more than 1.6gb of children’s’ stories to their research community for the purpose of teaching human behaviors.

It’s by no means a perfect system yet, but it’s getting there. As one Stanford researcher pointed out, “If fiction were truly representative of our lives, we might be constantly drawing swords and kissing in the rain”. Speaking personally, if the biggest downside to fiction as a human study for AI is that my robot one day whips out a light saber in the rain and swears to defend my honor, then sign me up.