AI and copyright: the Stable Diffusion case

AI and copyright: the Stable Diffusion case

Artificial intelligence

The concept of “artificial intelligence” (AI) was coined in 1956 at the Darthmouth conference to refer to “the science and engineering of making intelligent machines, particularly intelligent computer programs”, using the definition proposed by John McCarthy, one of its founders, in 2007. Along the same lines, the Merriam Webster Dictionary defines AI as (1) “a branch of computer science dealing with the simulation of intelligent behavior in computers”, (2) “the capability of a machine to imitate intelligent human behavior”.

Although in its beginnings the use of AI was limited to applications in very specific technical fields, such as voice recognition or data analysis, in recent years we have seen a huge growth of AI applied to products intended for the public, such as, for example, autonomous vehicles by Tesla® or personal assistants such as Sirius® by Apple®, or Alexa® by Amazon®. Very recently, the use of AI has become massively popular with the development of tools such as the already well-known ChatGPT, an AI system capable of interacting with a user based on questions and instructions formulated in an “almost human” way, and capable of performing, leaving aside whether better or worse, tasks ranging from gathering information to “creating” poetry, programming, and practically anything it is asked for. In the artistic field, tools such as DALL-E or Stable Diffusion stand out, and are capable of generating from instructions, images with a multitude of styles, details, and even worthy of a museum.

In this context, in addition to the debate as to whether an AI can “invent” and whether it should be listed as an inventor in a patent (see the DABUS case), the ability to “create” that current AIs already have has triggered the debate on the extent to which it is possible to attribute to them authorship of a piece of work. Added to this is the fact that AI models, to be functional, need to be “trained” with pre-existing content and that this content, whether text or images, may be subject to copyright, and therefore, its use is not permitted without prior authorization from the original author.

Copyright

Copyright encompasses the rights that creators have over their literary and artistic works. This includes books, music, artistic works such as paintings, photographs, and other audiovisual media, sculptures, and films, as well as software, advertisements, and architectural works, among others.

Despite differences according to each national law, two types of copyright can be distinguished: economic rights, i.e. the right of the copyright holder to receive financial compensation for the use of his/her works by third parties, and moral rights, which include the right to claim authorship of the work and to oppose any modification of the work that may be detrimental to the author’s reputation.

At the international level, copyright is regulated by the Berne Convention, to which 181 states are currently part of, and is obtained automatically without the need for registration or other formalities. Although there is a high level of harmonization between member states, each legislation has its own particularities and copyright is a territorial right, i.e. it is applied independently in each country, and that’s why national legislation must be taken into account in each case. As for most intellectual property rights, the owner of a copyright may grant “licenses” for his/her works. These licenses are authorizations to third parties to use or exploit the work and may or may not involve remuneration for the rights holder.

The Stable Diffusion case: when creating involves the use of copyrighted images

Copyright and AI have recently clashed in what’s known as the Stable Diffusion case (or cases we should say).

As explained by lawyer Matthew Butterick in his article, a group of artists have filed a class action lawsuit in the United States of America, against three companies for the use of the AI tool Stable Diffusion. This tool, which belongs to the category of generative AI systems, allows to generate images based on a set of training images including millions of images obtained, according to the plaintiffs, without any authorization by the original creators despite it being copyrighted content.

Besides the technical details (very well explained in the previously mentioned original article, see link in the footnotes), the AI of Stable Diffusion uses a technique called “diffusion” to generate new images. This technique consists in storing compressed copies of the training images for the model, and use these to recombine them to generate derivative images. The plaintiffs, that define the AI as a XXI century “collage” tool, claim that beyond the similarities or differences of the new images with regards to the original images, the images generated by the software are derived in any case of use of the original included in the training set. Therefore, Stable Diffusion could generate an unlimited number of infringing images that would compete in the markets with the originals, causing permanent damages to artists and to art in general. 

Days after the news about this case in the US, Getty Images, a well-known online image repository, published a press statement informing that they have commenced legal proceeding in the UK against one of the companies mentioned above. They claim that the company has infringed intellectual property rights, particularly “by copying and processing of images protected by copyright and the associated metadata owned or represented by Getty Images absent a license to benefit companies’ commercial interests and to the detriment of the content creators”.

While we wait for the developments in these cases, and to see how the use of copyrighted content to train is legally interpreted in the different jurisdictions, it is important to bear in mind that intellectual property does not protect ideas, but only their expression. Therefore, it makes sense to wonder to what extent the inspiration that an artist gets from existing works, and the influence that these have on his/her creations, is different from the training process of an AI, and if it is possible to consider that the AI gets “inspired” by the existing content, instead of using it without permission. We will remain vigilant!

Author: Ricard Valdés-Bango Curell, Ph.D. in Biotechnology

Footnotes:

Photo by Possessed Photography on Unsplash.com