Navigating Copyright Amidst the Paradox of A.I. Authorship

By Nicola Dawson

In recent years, Artificial Intelligence has become increasingly prevalent across all sectors and the popularity of generative A.I. models has skyrocketed alongside it. There are numerous generative A.I. services that produce images, which are readily available to the public. These include Midjourney, Stable Diffusion, Adobe Firefly and Generative AI by Getty, and they work by allowing users to input written prompts, then producing related images. Whilst this has provided an imaginative new medium for creators to help realise their visions, the use of generative A.I. is currently largely ungoverned and as a result there are problems which have risen to the foreground. One of the biggest issues for creators who utilise generative A.I. is the current inability to gain copyright protection for their works.

It is broadly accepted that the protection offered by copyright serves two main purposes. Firstly, it means that the artistic output of a creator belongs only to them, and they can be reassured that their work will not be copied or replicated by others. By granting artists the rights to their works and the solo ability to choose how it is used, adapted, or disseminated, copyright protection ensures that creators are able to gain monetary compensation for their works through licensing and royalties. This leads us to the second aim of copyright protection, which is to provide financial incentive to artists, motivating them to create new designs and artworks.

Under current regulations in the United States, material generated by Artificial Intelligence is not eligible for copyright protection. The first factor which contributes to this is that the law stipulates an author of a work must be human. This detail is rooted in the fact that non-human authors cannot sue for copyright infringements and therefore protection would have no purpose. To circumvent this barricade, artists in the US who have utilised generative A.I. to create artworks have insisted that since they are the ones giving written prompts to the A.I., they are in fact the authors. The United States Copyright Office has rejected claims that use this defence, likening the process of giving prompts to an A.I. to commissioning a painting from a human artist. However, there is one creator who is using a different approach to try and gain protection for an A.I. generated work.

Dr. Stephen Thaler is scientist specialising in Artificial Intelligence, and the President and CEO of Imagination Engines Inc. Thaler is a pioneer within the field of Artificial Intelligence and began experimenting with neural networks—technology that simulates the architecture of the brain—in the 1980s. In 1992, Thaler created an Artificial Intelligence model called DABUS, which he trained using existing images to create associations between the data given to it and generate new ideas and images. In 2012, DABUS produced an image which it then entitled, A Recent Entrance to Paradise.

When Stephen Thaler was a child, he underwent a near death experience and, at the climax of the event, he recalls seeing an image of his grandmother and dog, produced by his brain in response to the traumatic event. Inspired by the brain’s reaction to trauma and the images we produce mentally when we are near the end of life, Thaler experimented with simulating this reaction in DABUS. By severing a portion of the DABUS neural nodes, he found that it created a reaction comparable to that which a human mind undergoes at the end of life. Similar to the visions which might appear to us before we die, DABUS created its own end-of-life review, producing and naming A Recent Entrance to Paradise.

At the recommendation of intellectual property lawyer and professor at the University of Surrey School of Law, Ryan Abbott, Thaler applied for copyright protection for the work produced by DABUS, crediting the A.I. system as the author. His application in the United States was rejected on the ground that the author of a work must be human in order to qualify for copyright protection, and the court also refused to grant copyright protection for the work with Thaler alternatively stated as the author.

This novel scenario has raised questions surrounding the notion of creativity and authorship, whilst also bringing into question the foundations of copyright protection. In Thaler’s initial application to the court for copyright protection, he listed DABUS as the author of the piece, demonstrating a genuine faith in its capacity to create works completely without creative input from Thaler himself. Showing none of the hesitance that other computer scientists typically do, Thaler is comfortable referring to DABUS as ‘sentient,’ believing it to have equal powers of creation to a human being. Whilst the notion of sentience and whether or not Artificial Intelligence models have or ever will achieve it is complex, A.I. models ultimately would not currently benefit from the ability to sue others for copyright infringement. Whilst copyright protection is not just a way for creators to make money from their creations, it is also difficult to see who benefits from A.I. models being able to register their works. Dr. Thaler’s concern partially stems from a belief that if there is no simple way for people to monetize the creations of generative Artificial Intelligence, they will simply cease to invest in the technology and the entire field risks being left behind.

Whether or not works produced by generative A.I. are eligible for copyright protection remains hotly debated and, as of yet, there is no clear answer. But the paradox of authorship which Dr. Thaler finds himself in—unable to claim authorship himself, yet also unable to claim authorship for DABUS—indicates the direction that this debate is going in. As the lines between creative intention and true authorship become more and more blurred, and as Artificial Intelligence technology advances, the answer to the copyright quandary may become more difficult rather than less.

There is another facet to the conundrum of reconciling Artificial Intelligence with copyright laws. The use of protected data in the training and development of A.I. models can result in copyright violations itself. In order to be able to produce a wide array of images in response to different prompts, generative Artificial Intelligence models are trained using enormous amounts of data which is taken from the internet through a process known as ‘scraping.’ However, some of the material collected and used to train generative A.I. models is indeed protected by copyright, and its use can result in violations. This becomes especially apparent when generative A.I. services are utilised by people to create artworks which incorporate elements from other, protected pieces within them. In response to this, many artists have petitioned Artificial Intelligence services to remove their copyrighted works from the data they use for training, and governments have been urged to put pressure on A.I. companies to disclose details of their training data. Not only does the use of copyrighted material in the training of generative Artificial Intelligence create the potential for copyright infringements, but it also raises questions regarding the legitimacy of artworks created by A.I. and whether or not they are simply recycling and recompiling existing images into new ones, rather than creating original pieces.

Art generated by Artificial Intelligence seems to clash with copyright on all sides and there is no doubt work to be done in providing guidance to users in regard to their rights to protect their creations, as well as establishing increased standards of transparency and accountability within Artificial Intelligence developers.