In January of this year, Dan Mangan addressed a question about songwriting on his Substack. A reader wanted to know his thoughts on developing an initial song idea into a "rounded-out, singable lyric."
Mangan confessed he doesn't really know how songwriting works. Part of it is like figuring out a puzzle, and part of it is more accidental, a relationship with an ephemeral, hard-to-explain muse. Quoting an old CBC interview with Leonard Cohen, he noted songwriters somehow make use of whatever they have at hand. But lest the reader think Mangan has a romantic '60s-era notion of creativity, he also compared the process to what new AI tools like ChatGPT do. When given a prompt, AI tools use whatever data they've been trained on to offer a response that looks like previous patterns.
"It's exactly like what we do," Mangan tells Exclaim! now.
These days, AI tools themselves are something that songwriters have at hand. Mangan often uses ChatGPT to search the internet instead of Google. He didn't use AI for any lyrics on his new album out May 16, Natural Light, but the "all in the room together" recording of the album resulted in problematic bleed on certain songs between the vocal and instrument tracks. So, he used Moises AI Audio Separation to remove some bleed from the mix. "I'm not a purist," Mangan says.
Full disclosure: I'm not a purist, either. In writing this article, I used an AI tool called Otter to transcribe my interviews. I also used Otter's generative AI chatbot to help me sift through the transcripts. And I used an AI tool called Perplexity to help with my research. I've been interested in computer-assisted creativity ever since I wrote a master's thesis 15 years ago about how writers can improve by studying data sets akin to the ones that current AI tools use.
Mangan acknowledges that AI tools have ethical problems. One is that they may take our jobs. He's noticed a lot of fear in his industry, which was already struggling before people started worrying about AI. While Mangan doesn't think an AI can replace the live experience a human musician provides directly to an audience, he does see AI tools replacing many jobs in disciplines like graphic design. But one possibility is that productivity gains will mean everyone will have to work less. Mangan hopes that we can regulate AI enough, or implement a universal basic income, so that the benefits of productivity gains don't go solely to the companies who own AI.
Another ethical problem is the energy consumption necessary to power the data centres AI tools run on. The environmental impact is huge. Again, we need regulation. As can be seen with things like the continued commercial omnipresence of paper coffee cups, this will likely take a while. In the meantime, people will continue to use AI tools like ChatGPT.
"The curse of convenience is that it works really, really well," Mangan says.
But regulators are tackling a third major ethical problem with AI tools, one that's directly related to songwriters' livelihoods: around the world, governments are looking at the unpaid use of songwriters' copywritten material as training data to help AI learn human patterns.
Last year, the EU passed an AI Act. It requires developers to comply with existing copyright law, which would mean no scraping songwriters' songs without permission. The US has no such legislation yet, and developers to date have claimed their training data falls under the US's "fair use" doctrine. But a court ruling against a developer earlier this year suggests that US law may be moving in a similar direction to the EU. There are many such law cases underway south of the border, including one by all major record labels against AI music services Suno and Udio.
Here in Canada, we have a similar policy to the US's fair use, which we call "fair dealing." Though we don't have an AI Act yet, the potentially good news is that fair dealing is already a narrower set of exceptions than fair use.
SOCAN chief legal officer Andrea Kokonis is optimistic for a new framework in Canada that pays songwriters when developers use their songs. "Many of us are trying to compel these AI tech companies to recognize the creators," Kokonis says. "You do need to compensate and make sure that creators are part of your payment chain."
A related issue governments are working on is whether or not new songs and art generated by AI can be copyrighted. An intellectual property lawyer in India, Ankit Sahni, has been testing jurisdictions around the world by submitting an AI-generated image to various copyright offices. In Canada, the image was initially accepted for copyright, but this is only because our copyright registration is somewhat automatic. The Canadian Internet Policy and Public Interest Clinic at the University of Ottawa currently has a law case to remove Sahni's image from the copyright registry — a case that Kokonis thinks will be successful.
"Our Copyright Act clearly ties authorship to humans, which I think is right," Kokonis says. "A machine can't be an author."
Yet, as the Canadian Radio-television and Telecommunications Commission (CRTC) has noted, "Recent developments in AI technologies have led to the creation of tools that are being integrated into the Canadian audio and music industries at all levels." The CRTC is holding consultations on sustainable broadcasting, with public hearings starting June 18. They've asked industry organizations like SOCAN under what circumstances AI-generated music could be considered Canadian content.
SOCAN has recommended that AI-generated content should never be considered Canadian, on the basis that no AI-generated content is protected by copyright in Canada. But AI-assisted content — like Caribou's 2024 album Honey, which features vocals altered using AI — is another matter.
"Creators and other creative industries have used AI as tools along the way, and that's not something that in any way anyone wants to stop." Kokonis says. "Just because they use an AI as a tool, would that disqualify them from copyright protection? No, I don't think so."
As for Mangan, he thinks the difference between AI-generated music and human-generated music — even when a human is assisted by AI tools — isn't just about copyright law. It's about how, in our yearning to connect with each other, we stumble upon truths we can all feel, even if we don't quite understand why. Somehow, we break the old patterns with something fresh and new. Machines don't have feelings, or desires, or consciousness. They can only represent back to us the patterns they see in the human expressions of these things from the past.
"I don't believe AI can accidentally articulate the truths of existence," Mangan says. "And that's what we do."