Google AI pens dark poems after being force-fed 11k books

How do humans make an artificial intelligence system better at conversations? One method-in-testing is a project in which a Google AI was force-fed 2,865 romance books, about 1,500 fantasy books, and more. The work was done by researchers with Google Brain, and involved feeding a total of 11,000 unpublished books to the neural network, then testing whether it could take a couple sentences from the book and create its own corresponding phrases. The results sound like cryptic, dark poems.

The AI's task wasn't terribly simple – the two sentences were the start and end of a progression the neural network had to make — a sort of machine-generated poem that had to form a sensible transition between the two sentences. The 'poems' also had to feature proper grammar and syntax. Here's one of the resulting gems:

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.

It's strange, and a bit nonsensical, but impressive enough for an AI (keep in mind, everything between the first and last sentence was generated). This one, though, is something you'd probably guess is the work of an angsty serial killer rather than a neural network:

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.

The entire process behind the results you see above is detailed in this paper. This is just a small part of Google's work with artificial intelligence and the goal of bestowing it with human language capabilities. One of part of its research, the highly accurate natural language parser Parsey McParseface, was released today as an open source offering.

VIA: Quartz