Artificial Intelligence Still Sucks At Poetry

Artificial intelligence has bested humanity in almost every intellectual pursuit going. They’ve trounced us at chess, murdered us on the TV show Jeopardy! and been crowned world champion at the ancient board game Go. But when it comes to poetry, computers are worse than that beret-wearing kid from high school who always rhymed “love” with “above”. A poetic Turing test was recently held at Dartmouth College and the results were pretty messy.

Poetry robot picture from Shutterstock

In an attempt to test the artistic skills of robots, researchers from Dartmouth College pitted an artificial intelligence program designed to produce sonnets against human poets in a genteel “rap” face-off. To make things fair, both teams were given the same list of nouns to work with. The judges — which included Pulitzer Prize-winning author Louis Menandwere — were then tasked with finding the AI algorithms from among the human compositions.

In every instance, the judges were able to identify the sonnet produced by the computer program. Menand said the AI poems suffered from “idiosyncrasies of syntax and diction… uses of language that were just a little off.”

As IBM’s Watson has shown, AI programs are getting much better at mimicking and understating everyday language. However, it seems they still have a long way to go when it comes to arty-farty stuff.

Interestingly, the AI competitor did better when it came to musical compositions. At the same event, a group of dance music fans were asked to distinguish between tunes produced by human DJs and computers. Almost half of the dancers in attendance failed to correctly identify the AI ditties. (To be honest, this may have more to do with the quality of modern dance music than the artistic skill of AI.)

[Via AP]