Google loves an innovative AI project, whether it's having fun with the Google Arts and Culture selfie or creating dancing GIFs with last year’s Move Mirror experiment.
The latest AI to come from the company is called PoemPortraits, a web app that asks you to suggest a word, take a picture, and then creates a moody selfie, complete with some AI poetry.
Google is working with London-based artist Es Devlin on the new artwork experiment. According to Devlin, the experiment creates poetry based on the word you've chosen thanks to an algorithm that has been trained on 25 million words of 19th-century poetry.
The algorithm works like predictive text, says Devlin, by not copying or reworking existing phrases but instead using its training material to generate original phrases, based on the style its been trained on.
“The resulting poems can be surprisingly poignant, and at other times nonsensical. And it’s the profoundly human way that we seek and find personal resonance in machine-generated text that’s the essence of this project. I was inspired by the writing of Shoshana Zuboff on the “information civilization”—she writes, “If the digital future is to be our home then it is we who must make it so.’,” writes Devlin.
All the lines of poetry that are created through the project will be combined into an ever-evolving collective poem.
Machine generated art is starting to pick up steam. Last year, auction house Christie’s put an AI-generated artwork under the hammer for the first time, and it eventually sold for £337,000.
Created by French artist collective Obvious, the algorithm which created the portrait was trained using a data set of 15,000 portraits painted between the 14th and 20th centuries, before eventually coming up with the final portrait of the fictional Edmond Belamy.
It beats another picture of avocado on toast on your Instagram feed any day.
