Last week, a research team showed off a new project that aims to teach a Google AI system to be more conversational by learning from thousands of romance and fantasy novels. A fine sample to replicate from, you say? Wait until you read the AI’s results.
Quartz found the paper which details the AI’s response when presented with two sentences, and asked to create additional sentences to link the two sentiments, i.e. “I’m fine” to “But you need to talk to me now.”
The research says the goal of this exercise is to let the AI figure out the many different ways to convey the above expressions, and say them in a logical order that make the most sense. Here’s what the Google Brain came up with:
there is no one else in the world.
there is no one else in sight.
they were the only ones who mattered.
they were the only ones left.
he had to be with me.
she had to be with him.
i had to do this.
i wanted to kill him.
i started to cry.
i turned to him.
Whoa, Google. We’re just trying to have a normal conversation, no one asked you to help write a new Simple Plan song. Let’s try this again:
“i want to talk to you.”
“i want to be with you.”
“i don’t want to be with you.”
i don’t want to be with you.
she didn’t want to be with him.
Maybe romance novels were a bad choice to sample from.
“no,” he said.
“no,” i said.
“i know,” she said.
“thank you,” she said.
“come with me,” she said.
“talk to me,” she said.
“don’t worry about it,” she said.
I’ve had easier times parsing E.E. Cummings’ poems than figuring out what the heck is happening in the above scenario.
The project’s lead Andrew Dai said the team chose romance novels because most books in this genre follow similar plot lines, making it easy for the AI to understand how language and speech may adapt over a course of time.
While the team admits there’s more work to be done, it’s happy with the AI’s success in at least creating coherent sentences… even if the passages came off more like murder mystery novels involving a psychopathic killer.
For more details and examples, check out the full research paper here.