Copyright or Copywrong?

As AI systems increasingly generate creative outputs, rely on intricate algorithms and process vast datasets, questions around ownership, licensing and protection emerge. These issues underscore the need for innovators and legal professionals to adapt traditional IP principles to the novel challenges posed by AI technologies.

Training for success

Training data lies at the heart of AI development, and its use highlights some of the pressing IP issues and challenges. Successfully navigating this space requires an understanding of a complex web of copyright, database rights and licensing obligations.

Under the Copyright, Designs and Patents Act 1988 (CDPA), for example, materials that are used to train an AI system that individually qualify as literary, artistic, musical or dramatic works can be protected. However, using such content without proper authorisation may result in copyright infringement, exposing developers to legal liability where they scrape or potentially learn from third-party content.

Once assembled, the dataset used to train an AI system may also qualify for protection in its compiled form. Copyright can protect databases as literary works if the selection or arrangement of content is sufficiently creative, while sui generis rights can protect databases where substantial investment has been made in obtaining, verifying, or presenting the content.

However, databases that are generated by computers or machines may not qualify for these protections. This distinction can be important when it comes to considering how best to protect the inputs in an AI system, particularly when it comes to systems that autonomously create or compile their own training data, as can be common where adaptive learning, self-supervised learning or reinforcement learning techniques are employed.

Output or outlaw?

The creative outputs of AI systems, such as images, music, or text, can also challenge the traditional concept of authorship and copyright, although in the UK, the CPDA may afford protection for AI outputs and creative works.

Copyright law is generally based around the principle that only content created by human beings can be protected. However, in the UK, under section 9(3) CDPA, copyright in a literary, dramatic, musical or artistic work which is computer-generated may subsist in the person “by whom the arrangements necessary for the creation of the work are undertaken”. This provision could theoretically extend copyright protection to AI-generated content.

However, the absence of definitive case law interpreting this section of the Act leaves some uncertainty, particularly as to who the person undertaking “the arrangements” might be. Is it the entity that developed the system in the first place? Or might it be the person who provided the instructions to the system—such as when an end-user asks a system to generate an image of a dog? What if it were the system that undertook the task of creating the content in the first place, such as might occur in a fully autonomous or self-learning system?

All this uncertainty makes one thing clear at least: robust contract terms are essential for system providers. However, the temptation of contractually exerting copyright must be weighed against the risks of the system outputting materials which infringe the rights of third parties—perhaps coincidentally, or because the end-user supplied copyright materials as part of their prompt when using the system.

While copyright issues abound, the courts are quite clear when it comes to patents. In the landmark case of Thaler (Appellant) v Comptroller-General of Patents, Designs and Trade Marks (Respondent) [2023] UKSC 49, the UK Supreme Court confirmed that only human inventors could be named on a patent.

The ruling excludes AI systems from being recognised as inventors, which could create a potential barrier when it comes to protecting certain AI innovations in critical fields, such as pharmaceuticals or advanced engineering. However, as the Court comprehensively explained in Thaler, under the Patents Act 1977, “any person” may make an application for a patent, leaving the door open to other parties making the application for the patent in place of the system.

Algorithms in action

Algorithms form the core of AI systems, but their protection under IP laws remains a contentious issue in the UK as well as broader European jurisdictions. Understanding the nuances is critical for innovators seeking to safeguard their technological advancements.

Algorithms are generally excluded from patent protection because they are considered abstract mathematical methods. This exclusion stems from the principle that mathematical methods lack the technical application required to qualify as patentable inventions.

However, this exclusion is not absolute. Courts and patent offices can distinguish between purely abstract algorithms and those applied in specific, technical contexts. For example, if an algorithm solves a specific technical problem or interacts with hardware to improve its functionality, it may be patentable.

Where a patent proves problematic, trade secrets offer an alternative form of protection. By keeping the algorithm confidential, developers can safeguard their competitive advantage. However, trade secrets have their limitations. They cannot be licensed or monetised as easily as patents, and they are vulnerable to reverse engineering or independent discovery. Unlike patents, when the trade secret genie is out of the bottle, the protection is essentially lost forever.

The bigger picture

IP considerations and laws extend far beyond these areas, of course. Trademarks, for example, also play a crucial role in protecting the branding of AI technologies, including product names and logos that distinguish solutions. Similarly, design rights can safeguard the aesthetic aspects of AI-related products, such as the hardware designs for things like robotic systems and autonomous vehicles, or user interfaces for software. Contractual agreements also often play a pivotal role in defining ownership and usage rights when multiple parties collaborate on AI development.

These additional IP domains and tools highlight the multifaceted nature of AI innovations, demonstrating the importance of a comprehensive intellectual property strategy that encompasses more than just the technical aspects.

Conclusion

The legal landscape for AI and intellectual property reflects the tension between traditional IP principles and the disruptive nature of AI technologies.

While in the UK, laws like the CDPA provide some unique provisions for computer-generated works, gaps remain, particularly when it comes to algorithm patentability and AI-generated outputs. Ongoing debates surrounding the scraping of public internet content when training an AI system can also leave developers at risk of building solutions that infringe on other parties’ intellectual property rights, or of having their own IP rights infringed.

As AI continues to evolve, so too must the legal frameworks that govern it.

While the debates continue and we await further legislative change though, adopting proactive strategies and staying abreast of legal developments remains critical for businesses and innovators, if they are to effectively navigate these challenges and ensure their AI-driven innovations are both legally compliant and commercially viable.

---

In this fifth video in my new 6-part series, “Artificial Intelligence: Navigating the Legal Frontier”, I delve into the complex intersection of AI and intellectual property, exploring how evolving technologies challenge traditional laws on patents, copyrights and more.