Skip to navigationSkip to content

“I want to talk to you”: See the creepy, romantic poetry that came out of a Google AI system

AP Photo/Marcio Jose Sanchez
FILE – This Jan. 3, 2013, file photo shows a Google sign at the company’s headquarters in Mountain View, Calif. When it comes to the…
By Thu-Huong Ha
Published Last updated This article is more than 2 years old.

“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.

What reads like bad teen poetry is actually supposed to help Google sound more human.

Researchers from the company’s deep learning arm, Google Brain, is hoping to find ways to make its search and apps understand and adapt to the way people actually speak—in part by feeding 2,865 romance novels to an AI system.

The unpublished paper was presented at the International Conference on Learning Representations on May 3. It shows how the team of linguists and computer scientists poured 11,000 yet unpublished books—including nearly 3,000 romance and 1,500 fantasy novels—into a neural network model, which is meant to mimic how the human brain works. Next, the researchers presented the system with two sentences from the books and asked it to generate sentences that could create a meaningful progression between the two.

So the researchers might say to the system: Create 13 sentences that morph from “I’m fine” to “But you need to talk to me now.” Each sentence in the series should be similar in meaning to the one before and after it to create a smooth transition, and they have to make sense grammatically and syntactically.

The point is to mimic the variations in human speech. So instead of trying to teach the system the perfect way to say “I want Thai food,” the researchers want it to learn as many ways to say that as possible, to anticipate how people really speak and ask questions. And the florid, repetitive language of romance fiction, it seems, makes good fodder for this kind of learning.

The ultimate goal is to create a system that can generate new and varied sentences—but the byproduct of the endeavor is some creepy, strangely tense snippets. One reads like an early draft of As I Lay Dying.

Here’s a selection of actual sentences created by the system, with the two original sentences in bold.

he said.
“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.

it made me want to cry.
no one had seen him since.
it made me feel uneasy.
no one had seen him.
the thought made me smile.
the pain was unbearable.
the crowd was silent.
the man called out.
the old man said.
the man asked.

he was silent for a long moment.
he was silent for a moment.
it was quiet for a moment.
it was dark and cold.
there was a pause.
it was my turn.

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.

📬 Kick off each morning with coffee and the Daily Brief (BYO coffee).

By providing your email, you agree to the Quartz Privacy Policy.