War Against The Machines: How AI Is Changing The Way We Make Music
Friday, January 12 marked the date of Flow Machines’ first musical release. Spearheaded by the highly venerated French composer Benoit Carré, the 15-track pop album entitled Hello World—a nod to the text traditionally used to test the functionality of various computer programs—is the collaborative fruit of many artists’ labor, including the Canadian folk artist Kyrie Kristmanson, the Belgian production team The Bionix and the Mercury Prize-nominated artist C. Duncan—as well as the Artificial Intelligence algorithm that ultimately crafted all of the album’s songs.
AI has increasingly become associated with modernity and the age of convenience. While sophisticated Artificial Intelligence has been, until recent years, only a speculative feature of science fiction, it now drives our cars, provides us with medical diagnoses and plays—and conquers—the world’s greatest chess masters. Now, it’s even bleeding over into our creative industries. The research project Flow Machines, an outgrowth of the Sony Computer Science Laboratories in Paris, is one of a handful of emerging enterprises exploring the possibility of using algorithms to create music.
Such machine learning-powered software is unique in its ability to produce complex outputs that can serve as decisions, predictions and recommendations that are based on patterns extracted from large data sets. In the case of Flow Machines, a project first conceptualized in 2012 and led by AI researcher François Pachet, a catalog of 13,000 songs was used to train statistical models that represent information about how atomic musical events, like notes and chords, follow in succession across different musical styles. These statistical models are then used to generate new melodic and harmonic sequences in a chosen style, serving as suggestions to musicians using the software during composition.
“The idea is that when an artist uses the system, the first thing he has to do is to decide which songs he wants the machine to be inspired by, whether that’s in the form of scores, lead sheets, or audio stems” Pachet said in an interview at the Flow Machines public launch in Paris. “The machine then analyzes all of these inputs, and the user asks, ‘Please generate a score based on whatever I gave you.’” The ostensible aim of the software, however, is not to replace musicians; it’s to help them generate new and unique ideas by giving them access to harmonies and melodic structures otherwise outside of their usual purview, not unlike a modern reimagining of Brian Eno’s Oblique Strategies.
Carré and Pachet both maintain that Flow Machines is the logical next step in the history of music production. Visitors to the album’s launch party in Paris were greeted by posters depicting milestones in the history of music technology, such as the advent of Pythagorean tuning or the invention of Pro Tools. The effect was clearly intended to underscore the inevitability of AI-produced music. The installation also forced visitors to challenge their traditional perceptions of music-making. “What is a musician?” One sign on the wall asked. “We enter a new realm of music technology, producing music that couldn’t possibly be done before created by people who might not ordinarily think of themselves as musicians,” it answered.
By loading the video, you agree to YouTube's privacy policy.
Learn more
This emphasis on technology’s role in democratizing and advancing supposedly antiquated forms of music production was echoed throughout the rest of the evening. The exhibition provided a primer for a presentation on the development and application of the Flow Machines algorithm as well as a preview of one of the album’s songs. As the event’s video explained, each of Hello World’s tracks was created by fusing every artist’s individual input with the algorithm’s own variations on lyrical sequences and rhythms.
The result, which was played on surround-sound speakers, was a summery, undeniably pastiche take on classic radio ballads, featuring harmonized chords and melodic progressions not dissimilar from its non-AI-produced cousins. Although it’s clear that Carré had agency over each song’s basic structure, the algorithmically-indebted stylistic flourishes were uncomfortably artificial. Tracks like “One Note Samba,” “Magic Man” and “Mafia Love – 16 Bits” were sung over by seemingly normal pop vocals that had been inordinately skewed, chopped and transposed by the algorithm. Their synthetic feel was off-putting and clumsy rather than novel and innovative.
If the Flow Machines algorithm is designed to act as a creative tool for composers and musicians, then, it’s only in its first steps. Hello World is unusual in its algorithmic approach to composition, but the music itself is only a kitschy and conventional take on modern pop. And while this vein of AI-produced music doesn’t currently appear to be a boon or a bane to musicians, it does raise significant questions about the implications of using algorithms to generate music and populate our listening outlets. How are listeners supposed to feel about AI-generated music? At what point will these algorithms be able to make “hits”? In a feature in The Guardian, music industry consultant Mark Mulligan suggested that AI music is not inherently about the quality of the music that it creates. “As long as the piece has got the right sort of balance of desired instrumentation, has enough pleasing chord progressions and has an appropriate quantity of builds and breaks then it’s good enough,” he said.
An approach to music production that sidesteps the creator points to the potential financial benefits that streaming services can receive by funding AI. Given that Flow Machines is publicly affiliated with the streaming service Spotify, it’s possible to assume that by padding playlists with music made by algorithms—and not by people—the company can avoid paying royalty fees to copyright holders after the music’s publication. When asked about the legal copyright procedures associated with Hello World’s production, Pachet merely answered, “The machine is never credited—it isn’t technically possible to credit a machine. And the music that we’ve put into the machine we’ve already received the right to use.”
It’s not difficult to overlook the possibly adverse applications of artificial intelligence in music. But it also seems that this conceivable reality won’t be realized for quite a few years, and that even if it is, algorithms like those pioneered by Flow Machines will likely be yoked with the more formulaic compositional styles of pop, EDM and modern folk. For now, AI music is a novelty at minimum and a creative tool at most, and Pachet and Carré are dedicated to exploring and expanding the contours of the Flow Machines tool. They’re even planning a second album to be released on their nascent record label, Flow Records. “The next album won’t be the same story,” Pachet said. “It will have different musicians and a different style. More focus on lead sheets and composition and less on orchestration. We’re trying to start songwriting in a more classical way. Maybe the third one will even be about rap. We’ll see.”
Hello World will be released on vinyl this March. Listen to the single “Magic Man” below and stream the entire album here.
By loading the video, you agree to YouTube's privacy policy.
Learn more
Read more: How Artificial Intelligence will change music forever
Published February 05, 2018.