Creepy Humanoid Robo-Artist Gives Public Performance Of Its Own AI-Generated Poetry

 

When people worry about robots coming to take their jobs, I don’t think “poet” is what they had in mind. Enter Ai-Da, a highly realistic, AI-driven robot firmly rooted in the uncanny valley that can paint, draw, sculpt, and, yes, write its own poetry.

In a first for robot-kind, Ai-Da gave a public performance of poetry “she” created in commemoration of famed Italian poet Dante Alighieri. The event took place Friday at the University of Oxford’s Ashmolean Museum as part of an exhibit honoring the 700th anniversary of Dante’s death.

For Ai-Da, writing poetry isn’t as simple as putting pen to paper: She was given all 14,233 lines of Dante’s three-part epic, the “Divine Comedy,” to digest and then, by drawing on her data bank of words and speech pattern analysis programs, used algorithms to draft a reactive work.

The results are pretty nonsensical, but to be fair, so is a lot of poetry. Maybe it’s just too high-brow for me. Or it could be that I’m too distracted by her soulless eyes to feel anything but creeped out.

“We looked up from our verses like blindfolded captives, / Sent out to seek the light; but it never came / A needle and thread would be necessary / For the completion of the picture. / To view the poor creatures, who were in misery, / That of a hawk, eyes sewn shut.”

Friday was the latest in a series of AI-driven artistic performances since the robot’s first solo exhibition in 2019. Gallerist Aidan Meller created Ai-Da in collaboration with Engineered Arts, a U.K.-based robotics company, and scientists at the universities of Oxford and Leeds.

Speaking to the Guardian, Meller said Ai-Da’s language model is so advanced that she can produce as many as 20,000 words in 10 seconds. While her human handlers do engage in some “restrictive editing” of her content, overwhelmingly the words and sentence structure in her poetry are entirely AI-generated.

“People are very suspicious that the robots aren’t doing much, but the reality is language models are very advanced, and in 95% of cases of editing, it’s just that she’s done too much,” he told the outlet. He posited that, given the rapid advancement of language models in recent years, soon “they will be completely indistinguishable from human text.”