Generative AI and Copyright – Part I

You would likely agree if we told you the featured image was a hyper-realistic painting of a boy on a beach and be interested in knowing the talented artist who pulled off this impressive masterpiece.

What if we told you the artist here was not a natural person but a generative AI tool known as DALL-E, and all it took to ‘paint’ this piece, was to type the words, ‘an African boy sitting on the beach staring at the ocean with the evening sun over him’. After typing in this phrase, it took DALL-E less than 10 seconds to develop four unique images, the above image being one of them, each conveying the message in the text prompt.

With tools like DALL-E, everyone can play at being an artist. But would that be enough to make us all artists, compared to those who use cameras, paint, and manipulate digital tools like CorelDraw to create original images? If I use ChatGPT to write a poem or an essay, can I rightly call that essay my original work?

That is the central question this article seeks clarity on, considering the copyright laws of a few countries. It is important to note that DALL-E belongs to a family of AI tools, broadly described as generative AI, alongside the popular ChatGPT and other similar devices.

To do this, we consider what generative AI is, how it works and the relationship between copyright, generative AI, and the human effort assumption that usually underlies copyright eligibility. We also consider how some copyright laws treat AI-generated works, as well as posit on treatment under Nigerian law.

Generative AI: An Overview
Generative AI is a class of AI tools capable of generating original text, images, audio, and audiovisual content based on user text-based prompts. The most famous example today is ChatGPT, developed by OpenAI. Examples of different generative AI tools in various content categories are text, e.g. ChatGPT and Bard; images, e.g. DALL-E and Midjourney; audio, e.g. Amper Music and Aiva; and audiovisual, e.g. Descript and Runway.

Generative AI tools can create original content in whatever category they belong, mainly because they are trained on large datasets (sometimes running into billions) of pre-existing content using unsupervised machine learning models. Unsupervised machine learning models are models designed to discover patterns in datasets without human intervention. For instance, unsupervised learning is fundamental to AI applications used in medical imaging for patient diagnosis and recommendation engines like YouTube and Netflix.

After learning to detect patterns in the training data, generative AI models can generate original content similar to the training data. There are two broad categories of generative AI tools -Generative Adversarial Networks, or GANS; and Transformer-based models.

GANs create images, audio, and other multimedia. They are trained using multimedia data. DALL-E is an example of a GAN. On the other hand, transformer-based models are trained on the information scraped from the internet to create textual output. ChatGPT is a good example of this.
With this background, this article’s next heading will consider copyright in relation to generative AI.

Copyright and Generative AI
Copyright covers a body of rights that creatives are vested with automatically upon creating an original literary work. Copyright generally allows the creator to control the reproduction of the original work, its adaptation, distribution, performance, display, and derive royalties from the commercial proceeds of the work—copyright grants exclusive rights over their creation, usually for 70 years plus the creator’s lifetime.

Copyright laws in most parts of the world presume creativity and human effort to assign rights or authorship over a work. Generative AI tools, however, challenge this presumption as they separate creativity or effort from authorship. In creating the image at the beginning of this article, there was little creativity or human effort; as such, it might not qualify for copyright protection.

With tools like DALL-E, everyone can play at being an artist. But would that be enough to make us all artists, compared to those who use cameras, paint, and manipulate digital tools like CorelDraw to create original images?

However, some climes create a different category of copyright to recognise works created with minimal human intervention, granting them protection, albeit for a shorter duration. These climes are thus more friendly for protecting generative AI works. The USA and UK are examples at both ends of the copyright spectrum.

Treatment of Generative AI in the USA and UK

A. USA
The copyright regime in the USA broadly holds the position that there must be human effort or creativity for authorship to exist. This position was affirmed in early 2023 when the US Copyright Office (“the Office”) revoked the copyright granted to the comic book author because the images in her book were generated using Midjourney. The Office believed that the book’s text was still covered by copyright as they were the author’s original words. The images, on the other hand, were not her work and were not eligible for copyright.

The Office subsequently released copyright guidance to clarify its position on generative AI. It believes that entering a prompt in a generative AI tool cannot be considered authorship, likening it to issuing instructions to a commissioned artist. For a work from a generative AI tool to be eligible for copyright, a human must have selected to arrange the AI-generated content sufficiently creatively, such that the resulting work becomes an original work of authorship, meeting the standards of copyright protection. Where the human modifies or edits the AI-generated content to a certain degree, it may also meet the criteria for copyright protection.

The Office further stated that the term ‘author’ under the US Constitution and Copyright Act excludes non-humans and generative AI tools by extension. The Office now requires applicants to disclose the use of AI-generated content in work submitted, explaining the human author’s contribution to the job. AI content that exceeds the threshold should not be included in the application. Where the Office learns that a work has significant AI content and the copyright owner did not disclose this upon application, the Office will revoke registration.

B. UK
The UK Copyright, Designs, and Patents Act 1988 (“the Act”) takes a dual approach to copyright protection. On the one hand, it aligns with the US position, deeming creativity as key to authorship. Works that involve significant human effort are granted copyright protection for 70 years plus the author’s lifetime.

On the other hand, the Act further recognises works with limited human involvement, separating creativity from authorship while granting them protection, albeit for a shorter duration. The Act achieves this by adopting a futuristic definition of computer-generated works. It defines them as works generated by computers in circumstances where no human author exists. This definition is broad enough to capture modern generative AI tools. When the Act was passed, the rationale behind defining computer-generated works this way was to deal with the advent of early AI systems, such as expert systems, which could generate original content in response to queries.

The Act also broadly defines who an author is under copyright. An author of a computer-generated work is the person ‘by whom the arrangements necessary for the creation of the work are undertaken’. Providing text prompts to a generative AI tool may be considered ‘arrangements necessary’ because the AI tool would not create any content without it. There is no requirement to arrange further or edit the work for it to be eligible. Thus, the images in the comic book would be copyright-protected on their own and as part of the book.

Considering the reduced human involvement, the Act defines the length of protection to be 50 years from when the author generated the work. As such, the copyright over AI-generated works may expire during the author’s lifetime and enter the public domain.

We have now engaged two spectrums of how copyright law treats generative AI through the USA and UK. In the final part of this article, we will consider how Nigerian copyright law treats, or will treat AI-generated works.