A big question we have asked ourselves in this IP class is “Who owns the copyright in an AI work?” It is a question that not only we—but courts and governments across the world have been struggling to answer.
For my project I decided to write a song about copyright, and specifically, the different ways that using AI in the making of a song could lead to tensions with various stakeholders. I had initially planned to create a music video based on the song I had written, but decided to make a storyboard you can use to follow along as you listen instead. If you are interested in seeing a blooper from my attempt to make a music video just scroll down to the bottom of this post!
Disclaimer: The images in the storyboard were each generated using ChatGPT. The song was created with the assistance of Suno AI, although I wrote the lyrics for it. I created the storyboard manually using Canva. No AI was used to assist with the writing of this post.
LEGAL ISSUES OF THE COMPLAINANTS IN THE SONG
DJ Romeo
The first complainant, DJ Romeo, is a producer who makes beats and recognizes that one of his beats made it into the AI-generated song. This points to a key focus of many artists in the AI debate: what happens when AI scrapes or ingests their copyrighted works? And especially when it subsequently includes such works in its outputs. We saw various examples of this problem pop up throughout the term. A notable example is the lawsuit that Warner Bros launched against Midjourney because its AI was allowing users to generate images of copyrighted characters like Superman and Scooby-Doo in its outputs. The typical defence that has been relied on for such scenarios is that it falls under fair dealing (or fair use in the US)—both for the ingestion and the output.
Section 29 of the Copyright Act in Canada states, “Fair dealing for the purpose of research, private study, education, parody or satire does not infringe copyright.” Criticism, review and news reporting are also exceptions under this section. CCH Canadian Ltd. v. Law Society of Upper Canada, sets out the test for fair dealing, the dealing must fall under 1) one of the allowable purposes and 2) it must be fair. At para 51 the Supreme Court of Canada also said, “‘Research’ must be given a large and liberal interpretation in order to ensure that users’ rights are not unduly constrained…[it] is not limited to non-commercial or private contexts.”
In the US, the Trump Administration seems strongly in favour of a fair use exception for AI companies scraping copyrighted materials, and US lower courts have ruled in favour of fair use for AI companies, although so far it seems to be in very fact-specific contexts (read article here). On the other hand, Australia has taken the opposite approach with its Attorney General declaring that a text and data mining exception will not be made available to AI companies who use copyrighted materials to train their large language models or LLMs (read article here).
Given that Australia is the only nation so far to have made this move, and that Canada has typically given “research” a very liberal interpretation, it is more likely that Canadian courts will find that fair dealing applies to the ingestion of copyrighted works. As for AI including copyrighted materials in its outputs, that may end up being dealt with on a case-by-case basis depending on how each user prompted the AI to reproduce copyrighted materials, and whether that use meets the fair dealing test. It seems unlikely to me that Canadian courts would institute a blanket prohibition on copyrighted materials in AI outputs; however, it is also hard to predict anything with certainty.
Fine Artist Lily
The second complainant, Lily, has a predicament that is similar to the one shared by Studio Ghibli artist, Hayao Miyazaki, and perhaps Vincent Van Gogh in the Suryast copyright case. What recourse does an artist have when their artistic work is not copied exactly, but their art style clearly is? The defendant AI Music Label in my song got AI to generate an album cover for it; however, the art was generated in an art style that is clearly Lily’s. Unfortunately for Lily, the Agent for the AI Music Label was correct in retorting that Lily does not have a strong case for a remedy since styles and genres are not copyrightable under Canadian IP law.
Unnamed Vocal Artist
The final complainant, an Unnamed Vocal Artist, unfortunately has her voice copied and used to perform the AI song. The plight of AI voice cloning, and the theft of artist voices by AI is a topic I tackled more in-depth in my previous post titled, “AI Voice Cloning & Artist Remedies”. The Unnamed Vocal Artist whose voice was stolen by AI, recognizes that she would not have a claim in copyright under Canadian IP law. However, she also knows that she may have several potential claims in tort, namely: violation of privacy, the tort of false light and the tort of appropriation of personality. And she definitely plans to sue!
IS THE COPYRIGHT SONG ITSELF COPYRIGHTABLE?
Now that we have thought about the various complainants’ issues raised within the song’s lyrics, let’s think about the song itself. Would this song I made be copyrightable in Canada?
Let’s break it down step by step.
The Process
- I wrote the lyrics to the song.
- I recorded myself singing the song, with the melody I had in mind.
- I fed the recording and the lyrics into Suno AI with the following prompt:
“This is a song for a scene in a musical. There are 4 singers in this song. One male singer and three different female singers. The chorus is sung jointly by all 4 singers at the same time. Each verse is sung by an individual singer. The song should have a hip-hop/r&b feel.”
Also note that, when I fed the lyrics into Suno AI, I added descriptions for what kind of voice I wanted for each verse. For example: “[Verse 1 – Black male singer]”, “[Verse 2 – White female singer]” and “[Chorus – All 4 singers]”.
The Analysis
The Law
For any work to be copyrightable, it requires: 1) originality 2) expression 3) fixation. The song has been fixated in a sound recording, and is clearly a form of expression (not merely an idea), so the key issue we will be looking at here will be originality. Songs usually have separate copyrights for various aspects of them; however, I will focus my analysis on the lyrics and the melody. The applicable case law and legislation is as follows:
In section 5 of the Copyright Act, copyright is said to exist “in every original literary, dramatic, musical and artistic work”.
In section 2 of the Copyright Act, “every original literary, dramatic, musical and artistic work includes every original production in the literary, scientific or artistic domain, whatever may be the mode or form of its expression…”
Finally, in the case CCH Canadian Ltd v Law Society of Upper Canada, 2004 SCC 13, the Supreme Court of Canada stated the standard for originality: “…an original work must be the product of an author’s exercise of skill and judgment.”
Lyrics
I wrote the lyrics in their entirety, so, as an original literary work I should have no issues with claiming my copyright over those. There was no AI-assistance, I am a human that is ordinarily resident in a Berne Treaty Convention country, and I exercised the requisite skill and judgement to trigger copyright protection.
Melody
What about the melody? In my original recording of the song, I used a melody that I came up with in my head. I didn’t write it down or create sheet music, however, the melody was fixated in the original demo recording I made using my voice, and would likely also be copyrightable in Canada.
That said, no matter how many times I fed the demo recording into Suno, the AI did not replicate the melody exactly like I had it in the recording. The AI took quite a bit of creative licence in its output. If you listen to both recordings—the demo and AI versions—you can tell they are related, but the melody has definitely changed in the AI version. Would this still count as a work copyrightable by me? Especially since the output is not exactly what I intended/came up with?
I suspect my best argument is that since I had copyright over the original melody (the one in the demo recording), therefore the version created with AI is a derivative of that original work which I am allowed to make under s. 3 of the Copyright Act. If that is the case, both works would belong to me. However, whether this derivative work would count as being copyrightable in its own right is another question.
Since AI is not currently allowed authorship in Canada, it is unlikely I will be able to establish separate copyright in the AI version of the song. This is because all the “new” elements added to the song in which the new copyright might subsist—namely, the voices, adjusted lyrics/adlibs and changed melody—were all made by Suno AI. I would have to fight to establish that there was sufficient human authorship and “skill and judgement” exercised in creating the AI version of the song. Would my prompt be sufficiently complex to override these hurdles? I do not think so.
Given AI’s propensity to not follow our prompts, and instead produce unexpected outcomes, how might courts be able to answer the question of when a prompt is sufficient to count has human authorship? Good food for thought on this question can be found in the article, Thirty-Six Views on Copyright Authorship by Jackson Pollock. In it, Pollock argues that the addition of an intent/causation/volition element in the copyright test could help clarify copyright law and make it better suited to handle questions raised by AI. I encourage you all to check it out!
A NOTE ON MUSICAL COMPILATIONS
Finally, I wanted to share a thought on musical compilations and the potential loophole they may present to get around the AI/human authorship issue in copyright—at least to some extent.
According to s. 2 of the Copyright Act, “compilation” means, (a) a work resulting from the selection or arrangement of literary, dramatic, musical or artistic works or of parts thereof, or (b) a work resulting from the selection or arrangement of data.
In Robertson v Thomson Corp, 2006 SCC 43, the SCC stated that the “Copyright Act establishes a regime of layered rights” at para 6. This means that freelance authors who submit their works to newspapers maintain copyright over their individual literary works, and newspaper publishers have copyright over the collective work of their newspapers. It stands to reason then that the same thing could be said about fine artists maintaining copyright over their individual pieces of art, and museum galleries perhaps having copyright over the curation of the collective artistic gallery/exhibition. Similar scenarios could apply to musical and dramatic works as well.
Therefore, I think that if someone made a series of AI songs almost mechanically and with little to no skill or judgment, but then put parts of those songs together to make a musical compilation or remix, it would be enough to get a compilation copyright. The initial AI songs that were generated may not qualify for copyright due to insufficient human authorship or skill and judgment. However, the very act of putting the different AI song elements together to make a new piece should be enough skill and judgement for a copyright in the compilation.
I would be interested to know what your thoughts on this (or anything else in this post) are in the comments!
Cheers,
Elsie
BLOOPERS!
Here is a short clip of my attempt to make an animated music video using Runway AI. It didn’t quite go as I planned, but you get the idea haha!
Copyright & Social Media
Communications Law