- A musician made an entire album using only AI-generated samples.
- "patten" said AI can widen access to music and allow more people to express themselves creatively.
When Damien Roach first discovered the AI-music tool Riffusion in December, he had no intention of releasing an album.
But four months later, Roach, who performs as "patten," released "Mirage FM," the first album made entirely from text-to-audio AI samples.
"I set about exploring and spent 48 hours intensively working with this tool and recording lots of material," he told Insider.
London-based Roach built up a catalog of AI-generated music samples before eventually releasing the 21-track album through his creative agency and label 555-5555 in April.
He said every sound on the album was generated with AI tools from Riffusion, and he spent weeks searching for interesting tiny fragments, as well as cutting and editing samples.
The album can be streamed for free on a number of platforms, including SoundCloud and YouTube, and covers genres including house, garage, pop, and grime.
The album is available for purchase on Bandcamp for £7 ($8.64) and Roach said he could see a future for the commercialization of AI-assisted music-production methods — and he's encouraging others to embrace the new technology rather than oppose it.
AI music is divisive
AI-generated music is booming — but many people aren't happy about it. The music seems to be getting more sophisticated and the possibilities are unnerving some creators who also fear the financial impact.
A song generated in the style of Drake and The Weeknd went viral in April, sparking a fierce debate and a wave of imitations that picked up millions of views on TikTok. Some Frank Ocean fans were also recently duped into buying "leaked tracks" that turned out to be AI-generated.
Major record labels and streaming platforms have indicated that they will work to stamp out AI-generated content or protect artists from unwanted imitation.
Spotify, for example, has removed thousands of songs generated with the AI-startup Boomy, which allows users to generate music and then earn royalties from streams.
Roach, however, doesn't think the use of AI in music should be a "black-and-white issue."
He said that although it's important to address concerns about deep fake songs made in the likeness of other artists, AI tools also offer the potential for increased accessibility.
"The idea that more people are able to express themselves with the assistance of tools like this I think is really important," he said. "We should embrace these new tools for expression and think about how widening the means of expression for as many people as possible can benefit our society as a whole."
Commercializing AI-generated content is complicated
There's been mounting anxiety over the legality of AI-generated content across creative industries.
Mark Weston, a partner at the law firm Hill Dickinson LLP specializing in intellectual property law, told Insider: "We're seeing concern in music and in text. We're seeing it in anything which has traditionally been the preserve of creative geniuses in the industry."
"It's causing concern because a lot of the issues around ownership and the ability to make money from creative content are gray areas," Weston added.
Although the US copyright office has said that content generated entirely by AI lacks the human authorship necessary to pursue a claim, some work has still slipped through the cracks.
In February, the copyright office said it was wrong to grant copyright protection to certain images in a comic book, as the images had been made with the generative AI art tool, Midjourney, for example.
But "Mirage FM" was not trying to imitate anyone.
Riffusion lets users input a textual description of a genre or type of sound and then it spits out sound bites based on the prompts. It's trained on a large number of pieces of recorded music that have been labeled with certain descriptors like "R&B" and "rap."
"It analyzes characteristics that those pieces of music have from the text description that you put in and produces new music based on that data set," Roach explained.
Weston said that for AI-generated music to be protected, a creator must do something significant with the output.
"If you've got music and lyrics and you put them together, the combination may be copyrightable. If you edit the music substantially, then the output might be too," he said, noting issues were currently being considered on a case-by-case basis.
Nick Eziefula, a music and intellectual property lawyer at law firm Simkins, said that although the technology driving these concerns was new, the legal questions were not entirely.
"We've always had in the creative sectors where sometimes a new piece of work is actually based on someone else's," he said, noting that the new issues were around AI-powered imitation.
If it is directly copied from the work of another artist without permission, then it would likely be an "infringement," he said, adding that "copyright protects the expression of an idea, rather than the idea itself."
But Eziefula, who also makes music, was optimistic: "I think we'll have AI-powered styles and approaches to music which could be great to enjoy as long as it's done in a fair way, in a lawful way, and in an ethical way."