A poetic Turing test was held at Dartmouth College in New Hampshire this week to pit artificial intelligence against human poets, AP reported. Good news: the computers lost.
Judges could easily tell the difference between robot-written sonnets. One of the judges, Harvard English professor and New Yorker contributor Louis Menand, cited syntactical oddities and “uses of language that were just a little off”, as reasons why the AI-written sonnets stuck out as so obviously fraudulent.
Machines are getting better at natural language. They can caption our photos. They know when we’re slurring our tweets. They can even write well enough to fool us into thinking they’re human, at least sometimes. But poetry — arguably the most lenient written form in terms of grammar, punctuation, or anything else an AI might use incorrectly — is counterintuitively the task the robots still suck at.