Copyright and generative AI: Creative infringement or timeless innovation?

Experts unequivocally agree that ChatGPT, Bing, Bard, and Artificial Intelligence (AI) based systems like them have the potential to change not only internet searching but the whole of the internet — and the world economy.

AI is already seeing use in the workplace. For instance, some might use AI to power chatbots that help answer frequently asked questions, or to create image backgrounds for presentation at an upcoming meeting. Some organizations even have their employees use AI systems to analyze and enhance their information systems’ code. Under the umbrella of the aptly named “generative AI,” these systems can be used to create expressions of ideas and content.

The music industry is also grappling with the impacts of generative AI — fans of artists like Drake and The Weeknd are using generative AI to create original songs using their famous voices. These songs were previously uploaded to popular streaming services until the largest music label in the world, Universal Music Group, had them taken down due to intellectual property issues.

How do these systems work? As with a standard search engine, you can ask ChatGPT a question, like what the weather will be like in a city to which you will be travelling. The platform and tools like it expand on that basic premise with the ability to handle more sophisticated requests, generating answers to unique demands. Writing social media posts, creating a form for customers to use or making a presentation with sophisticated graphics are all within the ability of these new systems.

At a high level, these systems often work by feeding tons of data and information — usually in the form of art made by artists and content creators around the world — into a machine that combines it all and is able to produce something “new” in its ultimate output.

Naturally, this has led artists and organizations relying on their own intellectual property for their business model to question the legal consequences of such technology.

More systems appear to be arriving every day, with platforms like Midjourney (an image request and output system) and Cohere (an AI art and music creation service) disrupting the world of content creation and ownership. The thrust of concern from copyright owners is the desire to know if their works have been infringed by being fed into these systems. 

Is that Legal?

Interested parties are trying to determine the ramifications when copyrighted content is fed into generative AI platforms, asking questions like “is the training tracked” and “should a royalty be paid to the content creators?” So far, neither the platforms nor the law have definitively answered these questions.

AI systems are a mix of open and closed source code for the time being, so there’s a shroud of secrecy as to how each exactly works and on what data they’ve been trained. Most platforms claim their system creates works that are new and unique: in other words, works that should be considered original.

If a generative AI system uses protected works as input data in order to produce “new” content, are the rights of the original creator being infringed? Defenders would say each generative AI platform is no different than being inspired by old ideas and outputting new ideas the same way humans do — in the same manner that some creators might pay homage to different themes, characters, and plot points in their creations.

The legal argument that the creator’s rights are being infringed stems from the idea that the original works are being copied in an unauthorized way when they’re fed into the AI’s learning model — the content creators’ position is that they are the muses bringing creativity to the machines, and thus the machines’ outputs necessarily infringe the rights of others. In addition, humans don’t necessarily make a “copy” to pay homage or create works that take inspiration from other works — whereas the AI systems technically do so as they learn. 

Is it Unique Enough to be Original?

Canada’s Copyright Act protects original works made in a fixed form, irrespective of whether they are registered. The Supreme Court of Canada has held that works are original and able to be protected by copyright if there is sufficient skill and judgment used in their formation. Additionally, the Copyright Act is intended to protect only human-authored creations showcasing an exercise of this skill and judgment.

Given that AI platforms often just produce content after a simple text-based request is made, it could be said that there is no demonstration of skill and judgment. Conversely, there is the argument that human interaction with the AI’s content output can be original and worthy of protection.

The United States has grappled with the issue in more depth. Recently, the US copyright office registered the copyright for a comic book by a human author. Soon after, the copyright office noted that the images in the comic book were made by Midjourney, the image-based generative AI mentioned earlier. Upon notice, the registration was amended to only grant copyright protection for the author’s personally-made text, and the arrangement of the visuals and words. The content of the visuals themselves were not protected because they were made entirely by Midjourney; the human authorship requirement for copyright was not made out in the images.

In the United Kingdom, patent creation and ownership by AI currently is being tested. AI certainly has played a role in milestone inventions like the COVID-19 vaccine, but it has never been credited as an inventor — that is a role reserved for a human under UK patent law. In UK patent law, every invention must have a human inventor to be patentable. In the recent case of Thaler v. Comptroller-General of Patents, Designs and Trademarks, a UK researcher wanted to have his AI system identified as the author of a new invention it devised. Afterwards, the researcher wanted to transfer ownership of the invention’s patent to himself. Though already denied by the second-highest court in the land, the case has been appealed and will soon be decided by the UK Supreme Court.

This mirrors the researcher’s attempts in the United States, in Germany, and with the European Patent Office (among other jurisdictions). The researcher is largely unable to get past the initial stages of registration, because the relevant legislation in the above jurisdictions requires a human inventor. The United States Court of Appeal for the Federal Circuit confirmed as much in Thaler v. Vidal, rejecting the researcher’s application.

In Canada, the researcher’s patent application is still under review. The Canadian Intellectual Property Office responded to the patent application by saying that an application must be filed by the inventor or by the inventor’s legal representative. They also noted that listing a machine as inventor meant it was not compliant with the Canadian Patent Act, emphasizing that it is not possible for a machine to have rights under the relevant Canadian laws. Nevertheless, the Canadian Intellectual Property Office allowed the researcher to submit a statement on behalf of the machine as its legal representative.

The Supreme Court of Canada’s prior rulings have established that inventors are natural persons, and machines do not qualify. The Patent Act seems to require natural persons as well. However, this novel situation is still being considered by the Canadian Intellectual Property Office in depth.

Many are watching to see which courts or IP offices will ascribe the necessary personhood to an AI for inventorship, or provide a reasonable way to get around the requirement like allowing a legal representative to be considered the “inventor.” That could have the knock-on effect of supporting or impairing AI’s link to authorship within every area of intellectual property.

Is AI Really Fair?

Generative AI proponents and legal experts have tried to apply the lens of fair use in the US, and fair dealing in Canada. Both of these create exemptions for the use of copyrighted material. The infringing use is analyzed by the courts in an all-inclusive way, and is examined in the particular context of the usage. Though there is no settled law on fair dealing or fair use in the context of AI platforms, there are some cases making their way through the system that could have major implications.

Intellectual property experts are anticipating the results of the United States Supreme Court decision in Andy Warhol Foundation for the Visual Arts Inc v. Goldsmith. In this case, the court will address what is considered transformative inspiration (and thus not-infringing) versus what is derivative and infringing. Considering what we know about how generative AI works, the findings will be instructive for the United States, and could provide some persuasive positions as to how the rest of the world should view the use of generative AI.

In different cases, a class-action lawsuit has been started against Microsoft and OpenAI (the owners of ChatGPT) based on the alleged illegal use of copyrighted materials to train their machine learning system named Copilot (an AI-based software writing system). As well, Getty Images, one of the world’s largest visual media companies and stock image providers, is suing Stable Diffusion due to Stable Diffusion’s AI system allegedly training on millions of images protected by Getty’s copyright. Ultimately, these cases all attempt to determine whether these systems are infringing when training on these systems, what the consequences are, and whether they can be stopped from doing so.

Aye, Aye, Robot?

Though the legal case for or against generative AI’s copyright usage has not yet been tested in Canada, courts around the world are grappling with an issue that’s sure to become more common.

In the United States, artists have formed movements against the use of their copyrighted creations to train generative AI. Some are developing technology that prevents it from being scraped by machines; others like retailers and user-generated content sites are preventing the submission or sale of content produced by generative AI.

Conversely, other parties are embracing it more, seeing it simply as a way to produce significantly more output with less work and using it to create accompanying images for children’s books, and even endless (albeit rudimentary) internet streams that simulate popular TV shows from the past.

It is unlikely for the toothpaste to be put back in the tube, but there are certain to be surprises in new freedoms and restrictions that develop around this technology. Until legislators and the law catch up, the legal questions surrounding this technology will need to be evaluated on a case-by-case basis while establishing clear signposts for where the law should go next.

Fillmore Riley’s Intellectual Property and Technology Practice

If you plan to use AI for your business’s intellectual property or content, consider talking to an IP professional in order to ensure you’re up to date on the law. Not doing so may mean you do not have protection when you need it most.

The lawyers in the Intellectual Property and Technology practices at Fillmore Riley LLP can help you get the most out of your intellectual property. Contact one of our lawyers to learn more.