Create Your Album Now
EXPLORATORY RESEARCH | This project is an exploration of the capabilities and limits of AI technology as it pertains to the creation of music and associated arts, rooted in the history of New York City. The project was featured in an article about design and AI in Parsons School of Design’s re:D magazine.
The Goal
To explore music generated by AI platforms and contrast it with music by human artists, and to investigate this contrast with the residents of New York City.
The Timeline
February-May 2023.
The Process
Ideation, music generation, art & lyric generation, comparative analysis, graphic design, testing, marketing, statistical analysis.
This project was conducted in early 2023. The results of this study are rooted in that context. With the speed at which publicly available ML technology has been advancing, we would find drastically different results if we were to conduct this project today.
Stage 1 | AI Music Generation
To begin exploring the process of ML music generation, we devised a set of prompts designed to emulate music from different time periods and different artists localised in various regions of New York City. The prompts followed the format:
“write a song about location in New York City in the genre genre set in the time in the style of artist”
We filled in this prompt with a list of music that evolved in different areas of New York:
We then used Riffusion AI to create music based on these prompts, and Midjourney for album art:
Riffusion seemed to follow a similar pattern of music - beats and counts - with every iteration. The tunes observed were definitely based on Western music. It read genres like jazz pretty well for a computerised system.
ChatGPT came up with lyrics for the songs we generated:
Verse: Harlem, Harlem, oh what a sight / In the roaring twenties, it's a city of light / From the Cotton Club to the Savoy Ballroom / The sounds of jazz make the night bloom
Chorus: In Harlem, where the music flows / From the trumpets to the saxophones / The rhythm of the city never slows / In Harlem, where the jazz is home
ChatGPT was able to identify key touchstones of jazz as it developed in Harlem, and used those in the lyrics of the song. Each iteration of the prompt produced a more refined output. However, it did seem to get stuck in a systematic songwriting structure, and could not capture the fluidity of much of Duke Ellington’s music:
Harlem has those southern skies / They're in my baby's smile / I idolize my baby's eyes and / Classy up-town style
Stage 2 | Comparision
To see how distinctive and recognisable songs generated by AI were, we compiled them into a playlist, mixed with songs by the actual artist, and asked people around the city to vote on which songs were created by AI and which were by the artist. We put up posters that would lead people to the form, so they could listen and vote, in and around the areas relevant to the music.
Most people correctly identified who created the music. Most people also demonstrated interest in listening to more AI music.
Stage 3 | Listen to the Playlist!
We compiled a playlist of songs by AI on Figma, so everyone can listen to them!
* best viewed on a desktop or laptop computer.MORE PROJECTS
