Google Feeds Its AI Engine Formulaic Romance Novels To Improve Emotional Intelligence
LYNN NEARY, HOST:
Five years ago, playing against past human champions Ken Jennings and Brad Rutter, IBM's Watson won "Jeopardy!" in a landslide. Now, supercomputers are trying to win your heart. In the last few months, Google's Artificial Intelligence Engine has begun poring over romance novels, nearly 3,000 of them so far. The aim is to help the Artificial Intelligence, or AI as it's known, develop a more varied emotive tone in its interactions with humans. The books - well, imagine titles like "Ignited" and "Unconditional Love." Here's a sample from "Jacked Up."
ROBERT SIEGEL, BYLINE: (Reading) You weren't eating it. He stared down at her, trying to read her thoughts. Was it that she didn't want to be with him or was it just too spontaneous? OK, he told her. He wasn't going to beg or pressure her or let her think that he was being a jerk about it. Like you said, I understand. Work is hard. Maybe this isn't the time to mix it with pleasure.
NEARY: Google's researchers say that because romance plots are often formulaic, the engine can compare sentences across novels in order to learn several different ways of expressing the same idea. "Fatal Desire" is another work that Google's researchers are using to make their Artificial Intelligence more intelligent.
SIEGEL: (Reading) Walking into the room, Amber locked eyes with a tall, dark-headed stranger. She felt her breath leave her body, as if a fist had slammed into her rib cage. Quickly looking away, Amber walked to the bar and started filling her plate. Looking up, Amber found herself looking into blue eyes that she wanted to just get lost in. The stranger had also stepped up to the bar and was filling another plate with eggs and bacon.
NEARY: Google says the project isn't supposed to introduce anything lurid to Google's products. All the same, from now on, I'm going to keep a wary eye on my computer screen and all those auto-complete suggestions. Transcript provided by NPR, Copyright NPR.