Copyright Registration for AI-assisted Creations: How Much AI Input is Too Much?

It’s finally April. That means that summer can’t be that far behind, can it? Dreams of sitting down at the dock with a cold one in hand, watching the sunset while the loons trill in the twilight. That’s what the idyllic image above conveys. I’ve even written a poem—well doggerel verse actually—to accompany the image. Except that I didn’t write it, or even compose the title of the work; ChatGPT did. Nor did I create the image. That was done courtesy of DALL-E2, both by now well-known AI programs. I did have a finger in all this creativity. I wrote the prompt which went like this, for Chat GPT;

“Write me a poem in iambic pentameter involving watching the sun set over a lake in northern Ontario while I sit in Muskoka chair at my dock sipping on a cool drink, listening to the haunting cry of the loon”.

Not very creative, but whatever….Each time I entered the prompt I got a different bit of verse. Here’s one that is not completely terrible, labelled (by ChatGPT) “Sunset Serenity”. About the only skill and judgement (the criteria for originality with respect to copyright registration in Canada) I exercised was in picking this poem over another one that was even worse.

As the sun descends to meet the lake’s edge,

A golden glow spreads across the water’s ledge.

I sit in my Muskoka chair, calm and still,

Sipping a cool drink as the world grows chill.

The haunting cry of the loon pierces the air,

Echoing through the forest without a care.

The colors of the sky shift and change,

As the sun’s last rays slowly fade away.

In the distance, the forest creatures stir,

The rustling of leaves and the whir of a bird.

But here on my dock, I am content to be,

Watching the sunset and feeling free.

The lake becomes still as the sun disappears,

And darkness envelops the world without any fears.

But memories of this moment will forever stay,

Etched in my heart as I drift away.

I could probably improve on this poem but I won’t because I want to be sure that it is 100% generated by AI, with virtually no input from me, just like the image which was generated from a similar prompt. Why do I want to keep it “AI pure” and publish it “as is”? Because I am going to try to register this as a copyrighted work with the Canadian Intellectual Property Office (CIPO). And while a work does not have to be published to be registered, having it published gives it more credence as the date of publication becomes part of the registration process. So, if you are reading this, you have just helped me. There, it’s been published.

Why am I going to all this trouble? To prove a point? To test the system? For the hell of it? All of the above. I won’t actually be the first to register a work with CIPO that has been partially created with AI. In December 2021, IP lawyer Ankit Sahni managed to register the artistic work Suryast, listing the AI-powered painting app RAGHAV as co-author. The issue of whether an AI generated work can be considered an original work, subject to copyright protection, is a very live topic. The US Copyright Office has just waded into the fray as a result of a couple of authors who want to push the system, Stephen Thaler and Kris Kashtanova.

I wrote about Thaler last year (“The Humanity of Copyright”). Thaler has gone to inordinate lengths to try to get the US Copyright Office (USCO) to issue a copyright registration for works he has created with his so-called “Creativity Machine”. To date, he has been unsuccessful. Another run at this was taken by an author/artist Kris Kashtanova. Kashtanova submitted and received a copyright registration from the USCO for a graphic comic-book novel (“Zayra of the Dawn”) where the artwork had been produced exclusively by Midjourney, an AI application that produces images based on synthesizing millions of images scraped from the internet. In filing their application Kashtanova neglected to mention this salient fact, simply noting that their work had been “AI assisted”. When the copyright registration was issued, Kashtanova proclaimed that they (their pronoun of choice) had achieved a breakthrough, the first copyrighted work produced entirely by AI. The USCO was not amused. It initially threatened to rescind the copyright but later amended the registration to include only the elements that could be attributed to Kashtanova, such as writing the text and arranging the images. You can read the USCO letter here.

As a result of the brouhaha the Copyright Office’s stance caused, USCO has launched a “New Artificial Intelligence Initiative”. As explained by the USCO’s announcement,

“This initiative is in direct response to the recent striking advances in generative AI technologies and their rapidly growing use by individuals and businesses. The Copyright Office has received requests from Congress and members of the public, including creators and AI users, to examine the issues raised for copyright, and it is already receiving applications for registration of works including AI-generated content.”

As part of this process the Office issued new guidelines requiring that applicants disclose the extent to which AI has been used in generating a work. In setting out these guidelines the USCO tries to establish the line between an AI-assisted work (which is eligible for copyright registration) and an AI-created work, which is not, while recognizing the difficulty in drawing this line. The Office notes that it all depends on the circumstances, particularly how the AI tool operates and how it is used to create the final work. In the final analysis, whether the work is the result of human hand will have to be decided on a case-by-case basis. The line is far from bright; indeed it is exceedingly murky and grey. To assist the Office in making this determination, applicants are instructed how to complete the form. For example, applicants should take care to claim only the portions of the work that are human-authored, and “AI-generated content that is more than ‘de minimis’ should be explicitly excluded from the application”.

In addition to this guidance, the Office will conduct a number of “listening sessions” with artists, creative industries, AI developers and researchers, and lawyers with a view to gaining public input prior to issuing a formal “notice of inquiry” that will look into a range of copyright issues arising from the use of AI, including the use of unlicensed copyrighted materials in AI training.

That is what is happening in the US. The Copyright Office is clearly trying to respond to this exploding and explosive issue, show some leadership and develop a position that will enable it to come to grips with difficult questions raised by new applications of AI technologies. Now, what is happening in Canada? The same issues are at play, but the responses are different.

For its part, the Government of Canada, through the Industry and Economic Development Department (ISED) tried to get ahead of the issue by releasing a discussion paper on AI, Text and Data Mining and the Internet of Things back in July of 2021. The paper was thorough, posing all the relevant copyright-related questions arising from AI generated content. For example, even if an AI created work could be linked back to a human author, who in the chain of creation should be able to claim authorship? Or, how do we determine liability and infringement where the level of human involvement in creation of AI-assisted content is minimal?  ISED received a number of thoughtful briefs in response. However, since September of 2021 when comments closed, nothing appears to have happened on this file, although a proposal for new legislation to regulate harms generated by the application of AI, the Artificial Intelligence and Data Act (AIDA—yes, just like the opera!) has recently been introduced. The AIDA consultation paper focusses on a range of possible harmful outputs from AI for which businesses could be held accountable, but skips over the issue of inputs. The word “copyright” is not mentioned in this document.

The Industry and Economic Development Department (ISED), where CIPO is housed, has bigger fish to fry, like electric vehicles, infrastructure development and so on. To its credit, it is also trying to tackle the thorny issue of AI with respect to its potential impact on society, ranging from inherent bias affecting human rights to criminal behaviour affecting consumers. But when it comes to copyright–even though ISED is the department within the Government of Canada with statutory responsibility for this essential service—it gets treated like an orphan. The Department of Canadian Heritage, the other government stakeholder in copyright issues, is up to its eyeballs dealing with other priorities; a complex bill to regulate online streaming (C-11) as well as groundbreaking and controversial legislation (C-18) to require large digital intermediaries like Google and Facebook to reach deals with news media providers to compensate them when news content is made available on the platforms. Copyright, it seems, has been lost in the shuffle and nothing is being done to address the rapidly evolving situation of AI-generated content infringing on creator’s rights.

As for the copyright registration process and the question of whether AI-generated works qualify, CIPO has not issued any guidance. Nor, frankly, is it capable of much examination because no copy of the work to be registered is submitted to, or retained by, the Office. Unlike in the US, where the Copyright Office keeps a copy of a published, registered work, CIPO has no physical record of what is registered. It will review the details submitted in the application to ensure that a work qualifies for registration according to the information provided, but registration does not guarantee that a work does not infringe another. One wonders, frankly, whether registration is really worth the time and trouble. At least it is cheap.

Now that AI-generated Sunset Serenity has been published, I am going to pony up my $50, submit a registration application to CIPO, and see what happens. The application process has a box where the work is described. My application is going to describe the work Sunset Serenity (image and poem) as being created entirely by Artificial Intelligence (DALL-E2 for the image and ChatGPT for the poem and title) on the basis of minimal prompts (citing the prompt in the first paragraph of this blog post) provided by the applicant, (me) the owner of the copyright, in an action that was largely devoid of any skill or judgement. I will not claim to be the author. Then I will sit back and see what happens. Stay tuned. I will be sure to let you know how CIPO handles the registration process and if my AI friends and I get our copyright certificate.

© Hugh Stephens, 2023. All Rights Reserved.

This blog post has been updated to clarify that consultations on AI and other copyright related issues are carried out by ISED, not CIPO.