Support Centre

You have out of 5 free articles left for the month

Signup for a trial to access unlimited content.

Start Trial

Continue reading on DataGuidance with:

Free Member

Limited Articles

Create an account to continue accessing select articles, resources, and guidance notes.

Free Trial

Unlimited Access

Start your free trial to access unlimited articles, resources, guidance notes, and workspaces.

South Africa: AI's unavoidable impact on intellectual property - litigation risks

The past month has witnessed a surge in the number of allegations regarding the infringement of intellectual property (IP) rights by artificial intelligence (AI) models. In this Insight article, Tasmiya Patel, Davin Olën, and Amaarah Kapdi, from Dentons, unpack the broad international legal framework that is applicable in such cases, the potential defenses available, and discuss the remedies accessible to parties claiming the infringement of their IP rights.

To navigate this landscape, the article first articulates the methods used by AI to develop neural networks. It then proceeds to address the applicable international IP rights regime, which is subsequently developed, and concludes with an examination of the likely relief that a court may grant in such cases.

GeorgiaCourt / Signature collection / istockphoto.com

Data and AI

From the use of AI in diagnostic imagery in healthcare applications to travel itineraries, numerous sectors have been disrupted by recent AI platforms. While AI lacks legal personality, it can assess and analyze hundreds of existing inventions and artistic works created by humans to generate new inventions or artistic works. One example among many is the Next Rembrandt, a 3D-printed artwork that was created by analyzing hundreds of Rembrandt's paintings and mimicking the Dutch artist's work.

To illustrate the use of data by AI, this article will use Large Language Models (LLMs) as an example as most recent litigation centers around LLMs. Broadly, LLMs operate by predicting the likelihood of each word in a sentence that the LLM generates. LLMs draw from existing patterns identified between words in the language that they are trained on. Utilizing the available data, LLMs form connections between different words, creating a neural network. This neural network discerns patterns in word and phrase relationships, determining how words link together while using probability to predict the most likely follow-up word. Since an LLM's neural network relies on the data that it is provided with, the volume and quality of the data provided significantly influence an LLM's ability to establish connections.

Given the extensive datasets required, many LLMs source data from databases, websites, social media, user inputs, and other publicly available textual content. However, not all information on the internet is free to use, and available information may be subject to copyrights or usage exclusions. Consequently, it is crucial that all data provided to LLMs undergo pre-processing to ensure the protection of IP rights.

IP rights

As a first step, it is relevant to distinguish IP rights from human rights. Human rights are inherent, universal, and cannot be forfeited or restricted. In contrast, IP rights are granted by a state and come with a finite duration. IP may be defined as the creations of the human mind or intellect, including, but not limited to, designs, literary and artistic works, names, symbols, and images. IP is an asset which the owner earns a financial benefit from and/or recognition and is therefore protected by law through, amongst others, copyrights, patents, and trademarks.

IP rights are also transferable and are accessible not only to individuals but also to legal entities. The rationale behind the protection of IP rights can be framed as granting exclusive rights over a creation, effectively barring the public from infringing upon these rights. It is argued that the exclusion serves the greater public interest by fostering innovation, encouraging creativity, and motivating individuals and organizations to develop new ideas.

The legitimacy of IP rights finds support in various national laws and multilateral agreements. Chief among these are Article 27(2) of the Universal Declaration of Human Rights (UDHR) and Article 15(1) of the International Covenant on Economic, Cultural, and Social Rights (ICESCR). These international legal instruments acknowledge the significance of IP rights and strive to strike a balance between safeguarding these rights and promoting the broader public interest.

One legal right stemming from the ownership of IP is copyright. The foundation for international copyright protection was introduced by the Berne Convention, which built upon fundamental principles, three of which are relevant to this article. Firstly, the national treatment principle provides that countries must extend the same level of protection to foreign authors as they do to their own domestic authors. Secondly, the principle of automatic protection ensures that copyright protection is granted without the need for certain conditions to be met, such as registration. Finally, the principle of independence of protection stipulates that the enjoyment and exercise of copyright are not contingent upon the existence of protection for the work in the country of its origin. Collectively, these three principles provide a framework for international copyright, further articulated in the Trade-Related Aspects of Intellectual Property Rights Agreement (TRIPs Agreement).

The TRIPs Agreement was initially conceived to establish a minimum global standard for IP. The founding countries proposed that aspects of IP rights required fundamental protection. These standards were originally intended to be based on international norms prevailing at the time, and the TRIPs Agreement aimed to align and provide uniform protection to the IP rights of countries.

Copyright protection

Article 9 of the TRIPs Agreement and Article 2 of the World Intellectual Property Organization (WIPO) Copyright Treaty (WCT) establish that copyright protection extends to expressions and not ideas, procedures, methods of operation, or mathematical concepts. However, unlike many other forms of IP, copyright does not mandate registration for its existence.

The following principles are relevant for the purpose of this article for an IP holder to enjoy copyright protection:

Creativity and originality

While a work must be original, the principle does not require complete novelty. It simply requires that the work must be the creator's own creation and not a direct copy of another work. So long as this requirement is met, a work is considered original.

Use of existing subject matter

As an extension of the creativity and originality requirement, using existing material does not demand entirely new thinking. An author will only need to prove that their skill and effort have been invested in creating the work for it to be considered original.

Originality of an infringing work

When a person infringes copyright in creating a work, that work will not be afforded copyright protection.

Material form

The only requirement is that the work must exist in some tangible form. A work (except for broadcasts or program-carrying signals) is not eligible for copyright unless it is in written form, recorded, represented in digital data or signals, or otherwise reduced to tangible form. Traditional works can fulfill the tangible form requirement in works that are substantially preserved in the collective memory of the indigenous community. A digital literary work meets the requirement as it is encapsulated in digital data. Broadcasting is susceptible to protection once it has been transmitted and a program-carrying signal once it has been transmitted. Speech made without being written down or a composer playing music by ear without notation will not enjoy copyright protection.

 

Fair use

Recent cases dealing with AI have raised fair use as a defense for copyright infringements, as directly mentioned in the Getty Images v Stability AI case and the Andy Warhol Foundation for the visual Arts Inc. V Goldsmith case. In Getty Images, Getty accused Stability AI of developing an AI system using 12 million photos that were sourced from Getty Images, and the court has not yet reached a verdict on this case. Conversely, in the Goldsmith case, the U.S. Supreme Court had to hold that a depiction had been adequately altered to adhere to the originality requirement and, therefore, constituted fair use.

Fair use serves specific purposes, including research, private study, personal or private use, and criticism, review, or reporting of current events. In essence, fair use provisions aim to strike a balance between the rights of copyright holders and the broader public interest, allowing limited use of copyrighted material for specific purposes without infringing on the rights of the copyright owner. The determination of fairness relies on various contextual factors to ensure a just and equitable approach to copyright protection.

Including copyrighted material as part of the neural network of an AI system creates the possibility of copyright infringement. On the other hand, the fair use defense implies the alteration of copyrighted material. When using this defense, the defending party will need to indicate how the AI system output has met the threshold of originality and creativity. For this reason, it is crucial that AI systems incorporate adequate safeguards to prevent potential copyright infringement, or alternatively, that the data utilized in creating AI neural networks undergo pre-processing to remove any content subject to copyright.

From the perspective of an organization integrating AI as part of its data processing activities, the organization is likely to be considered both a controller and processor. However, these roles can be separated depending on the parties and processing activities involved. Regardless of the function, a controller will require the informed consent of the data subject to undertake processing unless a separate ground for lawful processing exists.

Relief

In the context of copyright law, compensation is addressed through compensation as a valid recourse when a copyright holder's rights are deemed to have been infringed by a judicial body. This may take three main forms. A significant component of copyright disputes involves determining hypothetical royalties, particularly in cases where no license agreement was in place. Reasonable royalties represent the amount that would have been paid had a licensing agreement been established under prevailing circumstances. Additionally, the notion of additional compensation comes into play, potentially allowing for further remedies in cases of copyright infringement. Lastly, the legal recourse known as 'delivery up' serves as a mechanism to enforce injunctions against copyright infringement. While courts may issue injunctions, there remains a risk that the defendant might disregard them. In such instances, the court may compel the surrender of the infringing 'goods' to the plaintiff, thus ensuring the efficacy of the injunction and safeguarding copyright rights.

Ownership of works created by AI

The question of ownership and authorship of works created by AI is nuanced. In the US, according to the rules of the U.S. Copyright Office, works produced by a machine or through a mechanical process that operates randomly or automatically, without creative input or invention from a human author, will not be registered. Similarly, the Federal Court of Australia held that inventions created by a computer cannot receive IP protection as they lack human origin. In the UK, courts determined that the inventor of a creation is the 'natural person' who conceived the inventive concept. In South Africa, a similar view is taken, offering little to no IP protection to creations produced by AI.

Although AI poses challenges to IP and IP law, it can also aid the protection and implementation of IP law. AI algorithms may be used to detect, identify, and monitor IP infringement across various online platforms by recognizing trademark misuse, copyright violations, and patent infringements. However, IP law and policy have not kept pace with the rapidly changing technology landscape globally. The dilemma surrounding inventorship, authorship, ownership, and infringement is a legal gap with no definitive answer.

Conclusion

Given the broad scope of protection afforded to copyright holders, it is likely that IP-related litigation will continue to mature in the foreseeable future. Courts globally will need to navigate delicately the extent to which AI may infringe on existing rights, informing their interpretation of fair use. Similarly, AI neural networks may be safeguarded from copyright infringement through data pre-processing or the development of adequate preventative systems.

Tasmiya Patel Partner
[email protected]
Davin Olën Associate
[email protected]
Amaarah Kapdi Candidate Attorney
[email protected]
Dentons, South Africa