Keeping you current

The Titles of These AI-Generated Christmas Carols Are Pure Cinnamon Hollybells

🎶 We wish you a Merry Jinglelog 🎶

These new songs will not be performed by a children's Christmas choir. (Wikimedia Commons)
smithsonian.com

When it comes to the visual arts, artificial intelligence has become quite good at producing strange, interesting and aesthetically pleasing images and videos. In fact, one AI-generated piece sold for $430,000 earlier this year. AI-generated music, however, hasn’t evolved quite as quickly. It’s certainly got the strange element down, but not much else. That’s evident in a batch of AI-generated Christmas tunes released by Swedish artificial intelligence company Made by AI.

As Amanda Kooser at CNET reports, the company fed 100 Christmas songs into a neural network, then waited for the bells to start ringing. While the resulting tunes are kind of a jingly mess, the titles are genius. The five songs-slash-fairy names are: “Syllabub Chocolatebells,” “Cinnamon Hollybells,” “Peaches Twinkleleaves,” “Cocoa Jollyfluff” and “Merry Jinglelog.” The music probably won’t be on heavy rotation. The songs are curiously fast-paced, relying heavily on synth piano and bell tones that start to make the tracks sound like music from The Exorcist. A few of the compositions like “Cocoa Jollyfluff,” which has a long section that appears to be sampled from “Carol of the Bells,” show their roots. They also lack lyrics, which would have been added if the company didn’t run out of time.

Overall, it’s a good effort and much less creepy than 2016’s AI-generated Christmas carol in which the neural network obsesses over flowers and—lest we forget—appears, ever-so-briefly, to become slightly self-aware (“I can hear the music coming from the hall,” anyone?).

Why has AI-generated art advanced so much while AI-generated music is mired in Creepytown? That’s the question Kaleigh Rogers at Motherboard posed to Hang Chu, the University of Toronto doctoral student who was one of the people behind the 2016 AI-generated carol. “Composing good music is actually more complicated than we expected,” he says. “Music is not something where if you throw enough data at it and hope the algorithm can figure it out, it will work.”

Datawise, art is relatively easy for an AI to learn since huge swathes are similar—for instance there may be hundreds of images of the human face in all sorts of styles, but they are still recognizably a face. With music, each individual song has so much more variability when it comes to instrumentation, melody, tempo, timing and harmony. It’s difficult to find commonalities between songs, much less identify what makes music a “Christmas song” (just try to find a common thread between “Silent Night” and “Disco Christmas” to understand what the AI is dealing with here).

John Smith, a fellow at IBM’s AI research center, tells Rogers of Motherboard that AI that just does one element of music at a time—like developing a melody—finds better success than one creating music from the ground up. It seems, at least for the time being, the creative element of the human mind can’t be replaced when it comes to generating some music magic. “The computer can start to do more and more of the groundwork and prep work and even suggest different ideas,” he says. “But that leap of creative thought, that spark of imagination, still has to come from a human.”

However, we may need AI to step in eventually if we want any new Christmas classics. Chris Lockery at Prospect argues that humans are just not writing the kinds of catchy Christmas tunes that are sung through the generations anymore, despite attempts from artists across the spectrum. Songwriters for a couple generations now have been writing sadder, slower, minor-scale music, which is the opposite of the major-scale, bell-based, swinging holiday music that’s binged in December. Then again, has Lockery just not been listening to Ariana Grande?

About Jason Daley

Jason Daley is a Madison, Wisconsin-based writer specializing in natural history, science, travel, and the environment. His work has appeared in Discover, Popular Science, Outside, Men’s Journal, and other magazines.

Read more from this author |
Tags

Comment on this Story

comments powered by Disqus