Just like reading bedtime stories to our children helps them learn and grow, so Google wants to feed its artificial intelligence engine with stories to make it a better conversationalist. And it looks like their bedtime story of choice comes in the form of steamy romance novels.
According to Andrew Dai, the Google engineer running the project, romance novels are better than, say, children’s books for this purpose because they tend to follow fairly similar story arcs but use a broad range of vocabulary to express those themes.
Yeah, we all know that story. Nice girl has a nice life but doesn’t know how to seek out more excitement. So she hides all her raw sensual beauty under layers of guilt and frumpy clothing until one day, handsome and sexually experienced guy comes along and takes off her glasses. And then her shirt.
Wait, that’s not the plotline of every romance novel ever written? You’re saying I’m making a blanket generalization here?
In any case, you might wonder why the Google AI engine was given a reading assignment of 2,865 romance novels these past few months. Well, it’s not because the company is trying to teach the engine to hit on you. The Google App currently has a problem with sounding too staid and factual. It can answer basic questions but the company wants to humanize it so users can start liking it a little more.
“It would be much more satisfying to ask Google questions if it really understood the nuances of what you were asking for, and could reply in a more natural and familiar way,” Dai told The Verge.
The more natural sounding language would also benefit other apps in Google’s ecosystem. For instance, the ‘Smart Reply’ feature in Inbox could show more conversational style reply options.
And to verify what the AI engine learned from this lesson in romantic intrigue, Google took it on a trial run by presenting two sentences. One would begin a story and the other would end it. The engine was tasked with thinking up plausible intermediary sentences that could connect the two. The process was repeated over and over again and the engine continually compared its own results to the source material in order to improve its tone and style. Early attempts were pretty ludicrous. But by the end, it was churning out some fairly decent constructions. So who knows, maybe we have a new novelist in the making on our hands.