Buzz Aldrin in a space suit from the Apollo 11 Mission painted in the style of Van Gogh's "A Starry Night"
An arrow pointing leftHome

AI Becomes the Artist

  • Ellie Vorhaben
12/8/2021

AI platforms are set to revolutionize the art world, all while raising important questions about interdisciplinary collaboration and the use of technology to create art.

AI practitioners focused on art had a busy 2021. In October alone, teams using AI algorithms revealed a lost Picasso and the hypothetical completion of Beethoven’s 10th symphony.

For the latter, a collaborative team made up of a composer, pianist, computational music expert, and AI specialist completed Beethoven’s 10th symphony, which upon his death consisted of just a few sketched notes and an outline. To go from this skeleton to a fully-fleshed-out, AI-designed, 40-minute-long symphony required the team to first train the algorithm in the style of Beethoven and his contemporaries, including Bach, Mozart, and Haydn. The team then taught the program the mathematical patterns underlying Beethoven’s work, using “Symphony No. 5” as an example. Finally, the team taught the program to harmonize a melodic line, bridge different sections of music, compose a coda (musical ending), and assign instruments to each part. Along the way, composer Walter Werzowa provided feedback to AI specialists so they could ensure it sounded like Beethoven’s style. The music was released publicly on October 9th in Bonn, Germany — Beethoven’s birthplace.

Days after this concert, a duo of computer scientists revealed a recreated Picasso painting at the Deep AI Art Fair in London. Starting with an X-ray of the 1903 Picasso painting “The Old Guitarist,” which revealed the painting of a woman beneath the artwork, the team trained a neural system to fill in the outline in Picasso’s Blue Period style. To train the system, they used a neural algorithm for artistic style developed by machine learning researcher Leon Gatys</a in 2015. The algorithm employs deep learning to recognize spatial invariance, aka patterns, in order to analyze art. It starts by identifying edges, then shapes, and finally patterns. This method allows it to learn the style of a particular artist or period. From there the program uses that style to fill in the outlines of a photograph, or — as with this Picasso — a painting.

Users can experiment with similar technology themselves by using the Night Café app or sending an email to the Twitter handle @images_ai directly via imagesaitwitter@gmail.com. Both the app and the company use the same two pieces of technology (CLIP and VQGAN) which let AI programs turn written words into pieces of artwork. The Night Café app allows users to create two AI-based pieces of original artwork per day for free.

On the surface, these AI programs might seem like the Night Café app: Just another tool that people can use to create art. Like any new tool, these are benefiting artists by letting them experiment with new mediums, methods, and aesthetics. It’s also benefiting viewers by increasing accessibility to view such artwork. But beyond just their impact as tools, these programs are forcing artists and viewers alike to consider important ethical questions about the process of creating artwork. Professor James Coupe of the University of Washington Center for Digital Arts and Experimental Media notes that the Picasso painting in particular raises questions such as, “Who is entitled to uncover a Picasso painting,” and “What does it mean for a group of computer scientists to make art instead of an artist?” Because these pieces are not the creation of the original artists themselves, they force us to move beyond analyzing art from the perspective of the end-user or critic (i.e. Does it look like a Picasso? Does it sound like Beethoven?) to in-depth considerations about the concepts, ideas, and meaning behind the processes used to make the art — which, in this case, are human-made algorithms.

These questions will only be answered through two-way collaboration between data scientists and the arts and humanities communities. The art world has gained new perspective from the use of AI programs, but it’s important that the reverse also be true: that data scientists gain new perspective from artists themselves. As noted by Prof. Coupe, these disciplines have much to teach each other, and if they are willing to work together, “there is an awful lot of shared ground to be discovered and AI can be a place in which that can occur.”

However, the cards may be stacked against the AI practitioners. According to the Picasso estate and the Design and Artists Copyright Society (DACS), the computer scientists behind the Picasso painting were not entitled to uncover it. They claim that the duo committed copyright infringement, a case that could be legally determined by a judge deciding whether the painting is deemed a recreation or a reinterpretation. Morally, the question is murky, considering Picasso had the right to preserve, or not preserve, his own artwork, and he cannot now give permission for his work to be “revealed.” In a letter to the data scientists, DACS calls for future AI-project teams to collaborate with the artist’s heirs when seeking to recreate their relative’s work. If only Picasso’s people would answer their calls.