An AI Wrote A Poem That Fooled Experts And It’s Really, Really Bad

Poetry is often subjective, with one written piece potentially being good or bad depending on who is reading it. Recently, however, an artificial intelligence took a stab at writing poems and they are objectively bad. Then again, the point of the exercise was to see if a machine could generate sentences that might be indistinguishable from real people and it seems that it’s getting there.

A joint venture involving researchers from Microsoft and Kyoto University, the AI that was developed was apparently good enough at making poems to fool judges most of the time. Of course, this was simply based on criteria such as whether or not the pieces were written by a person or a machine. Not whether or not they were any good. Here’s a sample of the poems that were generated:

the sun is a beautiful thing

in silence is drawn

between the trees

only the beginning of light

Perhaps it says more about how bad actual people are at writing poems that this verse managed to fool actual experts, but this is exactly what happened. Barring the terrible writing, however, it’s clear based on the results that machines are getting really good at passing for humans.

The researchers took their AI on a trial run by having Amazon Mechanical Turk judges take a look at some of its work and see if they can tell which are written by people are which were written by the AI. As Futurism notes, the AI managed to fool the experts more often than not.

Now, it’s worth pointing out that no company is likely to produce a book of poems written by an AI any time soon. This exercise was simply a benchmark for machine communication capabilities. It’s more probable that the AI is going to be used for more commercially viable enterprises such as customer service or as the next generation of chatbots.