In 1976, a manufacturer of audio cassette tapes ran a successful television advertising campaign in which a recording of Ella Fitzgerald's voice was played to see if it could shatter a glass as readily as Ella Fitzgerald's glass-shattering voice was capable of doing. The gimmick worked and "Is it real or is it Memorex?" became one of the best ads of the era.
Almost fifty years later, in the summer of 2023, a poem was read to an audience of writers at a conference in Banff, Alberta. This poem was "created" by a generative-AI tool in response to a simple prompt to make a poem in the style of a famous poet1. The poem, below, is entirely the creation of an algorithm:
In the quiet morning, a cup of tea in hand,
I watch as the sun pulls back the curtain of night.
There's an orchestra in the silence,
each bird adding a note to the symphony of dawn.
It seems an ordinary spectacle,
this steady march of light and sound.
But who am I to be the audience,
to this grand performance by the cosmos round?
We are but spectators in a silent film,
the universe playing all parts.
And this tea, bitter and sweet,
a simple reminder of life's contradictory art.
Is the poem any good? The writers thought so. Is it original? It hasn't existed before. Is it creative? You tell us. What we can tell you is that there is no copyright in this poem because it has no human author. It was created by a machine that has been trained by looking at an enormous set of written material to make a probabilistic determination of what the next word should be.
The writers at the Banff conference believed they were demonstrating the existential threat posed to creative industries by AI. And it's hard to argue with them. Why pay for humans to do what a machine might accomplish for free? And it's hardly like art, music, literature or poetry exist in a vacuum. Everyone who creates has been inspired by the creations they have read. So why is this different?
But another question looms large: why would creators agree to have manuscripts and recordings serve as a rich source of data to be mined for derivative works that... may be interesting enough to consumers that it might make themselves irrelevant?
But another question looms large: why would creators agree to have manuscripts and recordings serve as a rich source of data to be mined for derivative works that may not be copies, but may be interesting enough to consumers that it might make themselves irrelevant (or, at a minimum, less valuable)? This dispute is fomenting: many author agreements require, or politely seek permission, to allow publishers to upload manuscripts to the publishers' libraries. For more on the intersection of artificial intelligence and intellectual property, read Does AI have patent and copyright ownership?
There is a lot of money in play between rights holders and a burgeoning AI industry—pegged by some to be multi-trillion dollars in size. Canada itself has estimated AI's potential as contributing 16,000 jobs and $16.5 billion to the Canadian economy by 20282. Where there is money, there is often a fight. In June 2024, three major U.S. record labels sued two companies engaged in AI-generated music in a copyright battle over AI-generated content. The allegations are similar to those made in other cases brought by novelists, news outlets, and publishers in other lawsuits over tools that use Gen AI to generate content.
Is this copyright infringement? Something more or something less? Ordinarily, copyright infringement is the unauthorized reproduction of a substantial part of a creative work. The machine who made the poem did not do that. But it did mine copyrighted content to train the AI system to mimic it. It this fair use or fair dealing? Time will tell.
The songwriters want (and—let's be honest—need) their royalties. The nascent Gen AI companies really want to grow by deploying tools and disseminating their works. For growth like this, the industry needs data and may need some form of IP protection. Right now, copyright seems to be the shoe into which the cases driven by these music collectives are being horned. What about the opposite point: securing protection over AI-generated content? In Canada, the jury is still out on the question as to whether an AI can be an author. Just this past July, the Samuelson-Glushko Canadian Internet Policy and Public Interest Clinic (CIPPIC) filed an application in Canada's Federal Court to challenge a copyright registration in which an AI engine, RAGHAV Artificial Intelligence Painting App (RAGHAV), was named as a co-author. The lawsuit was brought in response to the Canadian Intellectual Property Office's 2021 copyright registration for an image, Suryast, based on Van Gogh's The Starry Night. The other co-author, Ankit Sanhi, had taken one of his own photographs, and used the RAGHAV app and the Dutch master's work to control stylistic and numerical values to generate Suryast. CIPPIC, in its court filings, says that this process of arriving at Suryast was purely mechanical and lacking in human skill or judgment—i.e., work not capable of resulting in copyright. CIPPIC also argues that "author" in the context of the Copyright Act means a natural person and that an AI system, not being human, cannot supplant the human's ability to create3.
We will have to await the outcome of the decision in Canada. We note that Sahni's (and RAGHAV's) attempts to seek copyright registration have failed in the U.S.—as have attempts by others to secure protection over AI-generated content. Moreover, the government is not just sitting on this problem. It has launched a consultation process that will close in January 2025, aimed at garnering stakeholder feedback on three important copyright policy areas: (i) text and data mining; (ii) authorship and ownership of works generated by AI; and (iii) infringement and liability regarding AI. (For more on developments in AI regulation in Canada, read What's new with artificial intelligence regulation in Canada and abroad?)
Finishing where we started, with the poem written in the style of another, a 2023 Notice of Inquiry on Artificial Intelligence and Copyright, the U.S. Copyright Office described "style" as an unprotectable "personal attribute". So, in addition to authorship hurdles, other issues may be lurking to fail any attempt to secure copyright protection over this little poem written by an algorithm based on the mere style of a famous poet4. Without protection, this authorless poem is left now to float through the ether of the internet—a seemingly original and creative verse lacking an author and a home—a "simple reminder", perhaps, "of life's contradictory art".
In conclusion, while there is nothing earth-shattering (or glass-shattering) at the intersection of AI and IP, there are interesting issues looming. It is safe to say that with the dollar values involved, and the gaping potential growth of AI, we haven't seen or heard of the last of them.
Footnotes
1. We won't name the poet, but we welcome guesses.
2. Consultation paper on Copyright in the age of Generative Artificial Intelligence. Government of Canada, December 2023.
3. This argument is similar to those made in the context of patents and the question of whether an AI can be an inventor. According to most patent offices around the world, the answer to the question is, no, an AI cannot be an inventor. An "inventor" has been held to be a natural "person," one who performed the act of inventing: Apotex Inc. v Wellcome Foundation Ltd, 2002 SCC 77. While this case law is evolving around the world, largely in response to a series of patent applications filed by Dr. Stephen Thaler in which he'd names an AI engine, DABUS (Device for the Autonomous Bootstrapping of Unified Sentience), as inventor, the trend seems to be that the natural persons need to have their fingerprints all over the issues of authorship and inventorship. AI inventorship also raises theoretical questions: If an "invention" was derived through the data mining of publicly available databases by search engines that are publicly available, doesn't that lead to a conclusion that the "invention" was obvious (in the sense that anyone should be able to replicate the process)? Interesting questions aside, the real debate will turn on to what extent the human can deploy these new tools to create protectable works.
4. And so much for Sahni's and RAGHAV's attempt to capture the style of Vincent Van Gogh.
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.