July 10, 2020

Automated poetry is the future of literature

Poets, in their uniquely human skill for finding value in the ugly and unrelated, are applauded for using language to organize emotion and unlikely connections into readable refrains. But what if mournful, powerful and thought-provoking poetry is created by artificial intelligence (AI) programs? 

Many argue that these are not poems, as they are not the product of human ideas and emotions. I argue that they are and that we should embrace these developments in language despite the literal lack of humanity in them.

The Infinite Monkey Theorem states that a monkey infinitely hitting a keyboard at random, will almost surely type any given text,  such as the complete works of William Shakespeare. I sometimes consider this when scrolling through poetry bot accounts on Twitter. Especially @poems.exe. 

 

Look how pretty this brief robot poem is! It contains assonance in the words “dragonfly and “glide,” alliteration in the words “slept-in sheets” and creates an interesting narrative as the toddler walks through where a dragonfly has flown or is about to fly. The speaker of the poem may be watching from an unmade bed. 

Yes, I just analyzed a poem written by a robot. It’s a nice little poem.

It’s true that a program designed to put phrases together might randomly create something of value, like the monkeys infinitely smashing on keyboards. At first, poetry automation may seem this way. But the monkeys in this equation (the AI behind the bots) are advancing in a fascinating and complex direction.

Humans shape these programs to follow the rules of language, but the programs are independently making choices on word usage. Unlike the monkeys, these programs may pump out something as valuable as the works of Shakespeare, given enough time — but the actual language would be crafted, not random.

According to Wired, Google fed their AI with 11,000 unpublished books to advance the neural network in its ability to communicate in May 2016. The team asked the program to generate poetry using the information, and the result was pretty mournful:

amazing, isn’t it?

so, what is it?

it hurts, isn’t it?

why would you do that?

“you can do it.

“I can do it.

I can’t do it.

“I can do it.

“don’t do it.

“I can do it.

I couldn’t do it.

Alex Lee from the Wired article humorously notes that 3,000 of the titles were romance novels, which may have contributed to the heartbreaking tone of this poem. The research paper released by Google’s team working on the AI system notes that there were several types of sentence autoencoders used in the generation of the poem. These encoders use different techniques to copy the structure, semantics and grammar of the content fed into the program.

It’s copied, but there are decisions independently made by the program to create the poem. Isn’t that what we do when we study literature? Don’t we read from the greats in hopes of imitating their greatness, but in our own way? T.S. Eliot, the great St. Louis-born poet once said, “good writers borrow, great writers steal.” 

Although these programs are copying the structure of written human emotion and thought, they’re starting to look similar to the human learning process. The art that results from these program’s learnings has just as much value as human art, and the poetry created by AI will expand and strengthen this genre of writing in the near future.

Share this post

Facebooktwitterredditpinterestlinkedinmail