The intersection of AI and IP law: Insights from IPOS’ Landscape Report

How is Singapore law likely to treat issues involving IP and AI?

Numbering almost 100 pages, the Landscape Report is a lengthy read. For readers with an interest in Singapore IP law, the most noteworthy portions of the Landscape Report are likely to be those which touch on how AI-related issues might be addressed under Singapore law. We summarise the key issues of the Landscape Report with the possible position under Singapore IP law below. 

Patents

Can an AI system be named as the inventor of a patent?

It is unlikely that an AI system can be designated as an inventor of a patent in Singapore because: 

• Section 24(2) of the Patents Act 1994 (Patents Act) requires a patent applicant to file a statement identifying ‘a person’ who is the inventor. It appears that this refers to a natural person.
• Section 2(1) of the Patents Act defines “inventor” as “the actual deviser of the invention”, which has been interpreted by the Singapore courts to mean the natural person who came up with the inventive concept. 

Can an AI-generated invention fulfil the patentability requirements?

In order for a patent to be granted over an invention, the invention must fulfil certain requirements. Two of these requirements are:

• The invention must be non-obvious, i.e. it would not have been obviously deduced by a person with ordinary skill in the relevant technical field.
• The invention must be sufficiently disclosed, i.e. described clearly and completely in the patent application, so that a person with ordinary skill in the relevant technical field can replicate the invention.

There has been significant debate on whether and how AI-related and AI-generated inventions could fulfil these two requirements:

• Non-obviousness: the requirement of non-obviousness was conceived in relation to human inventors. What might be considered non-obvious to the human mind could appear trivial to AI. The Landscape Report suggests that with the increasing use of AI by inventors to create new inventions, legal standards may need to be adjusted to accommodate these developments.
• Sufficient disclosure: there are suggestions that AI-related patent applications (i.e. inventions that form AI technologies, and involve the use of AI) should disclose information such as the correlation between the input and output data, explain the AI-decision making process and predict the AI’s decisions and output, and disclose the data sets used to train the AI.

While the Patents Act requires a patentable invention to satisfy the requirements of non-obviousness and sufficient disclosure, it remains to be seen how these requirements would be satisfied in relation to AI-related and AI-generated inventions.

Who is the owner of an AI-generated invention?

AI systems are machines without legal personality or capacity, and therefore cannot be owners of patents. Potential candidates for ownership of an AI-generated patent include the AI’s owner, user or developer. The question of who will be the default patent owner should be determined by policy review.

In Singapore, the Patents Act allows for a patent owner to be a natural person or a non-natural person with legal personality, such as a corporation. It is therefore permissible for ownership of patents to be assigned to a company, e.g. in the employment context, where inventors assign their invention to their employers that are companies. 

Copyright Law

Can AI be named as an author of a copyright work?

Singapore copyright law distinguishes between authorial works, which must have an author in order to enjoy copyright protection, and non-authorial works, which (as the name suggests) need not have an author. 

In relation to authorial works, under Singapore law, an AI system cannot be named as an author because: 

• Certain provisions of the Copyright Act 2021 (Copyright Act) imply that the author must be a natural person. For example, the duration of copyright in an authorial work is linked to the death of the author.
• The Singapore courts have previously emphasised that without identifying a human author from whom a work originates, there is no "original work" eligible for copyright protection.

As for non-authorial works, there is no requirement to name an author. Copyright may vest in an AI-generated non-authorial work if a maker can be identified: see below.

Are AI-generated works protected by copyright?

(a) For authorial works

If AI systems cannot be designated as authors, then the question arises: can copyright exist in the output that they generate?

The Landscape Report does not express a view on the Singapore law position. However, it mentions that: 

• If a human author can be assigned to an AI-generated work, this may fulfil the baseline requirement of originality for a copyright work. Depending on the circumstances, the human author could be the AI developer, programmer, machine operator or the user who fixes the work in its final form.
• It is generally easier to argue that copyright exists in a work if it is AI-assisted (rather than completely AI-generated).
• Under Singapore law , the standard for originality is the "creativity" approach. The Court of Appeal has described an "intellectual creation" as requiring the application of intellectual effort, creativity, or the exercise of mental labour, skill, or judgment towards the authorial creation. An AI-assisted or AI-generated work would have to cross this threshold in order to be protected by copyright; it remains uncertain whether de minimis human contribution to the generation of AI output amounts to the exercise of a sufficient creative choice.

(b) For non-authorial works

Non-authorial works that may be AI-generated include sound recordings and films. The Copyright Act provisions which deal with sound recordings and films refer to a “maker”, which includes any company, association, or body of persons. Copyright may exist in a sound recording or film generated by AI if a maker can be identified e.g. the company that owns the master sound recordings or made the arrangements for the film. 

However, the underlying components of the sound recording or film, such as the music, lyrics, and script, are authorial works. They would remain unprotected if no human author had been involved in their creation.

Who is the owner of an AI-generated work?

To the extent that AI-generated works are eligible for copyright protection, the first owner of the works may potentially be the owner of the AI system, the developer who wrote the algorithm, the trainer of the AI, the user who contributed to the final output – or no one. There are diverse views on this and arguments have been made in favour of each of these permutations.

Note that the owner of a copyright work may be a non-natural person with a legal personality, as opposed to an author who must be a natural person.

Trade Marks

Will the use of AI affect traditional trade mark law concepts?

Unlike patent and copyright law,  trade mark law does not require a trade mark to have the equivalent of an inventor or author. As long as the trade mark is duly registered, the method of its creation, even if facilitated by AI, is irrelevant.

Trade mark law relies on the concept of the “average consumer” when determining the likelihood of confusion between two conflicting trade marks. The growing use of AI for consumer recommendations may someday influence how courts define the "average consumer". However, we will have to wait for case law to develop on this issue. 

Passing Off

How does AI affect the law of passing off?

AI systems are able to generate material that convincingly replicates the voice and appearance of real persons. Some of this material has generated a lot of controversy, including deepfake content and AI-generated fake celebrity endorsements. 

From an IP law perspective, the tort of passing off may provide a legal remedy for celebrities whose likenesses have been replicated by AI without their consent. The Landscape Report suggests that in these circumstances, the success of a claim in passing off is likely to turn on whether the celebrity can demonstrate to the courts that there is a misrepresentation by the trader that has used the AI-generated likeness. The misrepresentation must mislead the public into thinking that the trader’s goods or services are in some way associated with the celebrity.

Impact of AI on Infringement

Does the use of copyright works for machine learning infringe copyright?

The Copyright Act introduced an exception allowing the use of copyright works for computational data analysis, so long as certain conditions are complied with (including lawful access to the works being copied). The exception does not distinguish between computational data analysis for commercial and non-commercial purposes. 

Additionally, the use of copyright work for machine learning may potentially be protected under the general fair use exception in the Copyright Act. 

Could AI models and tools infringe IP?

IP infringement may occur when the output of an AI model is substantially similar to an existing work or product, or when an AI model potentially infringes on a patent during the invention development process. There have been cases in other countries where copyright owners have sued on the basis that generative AI models infringed their copyright by (i) copying their works for training, (ii) producing AI-generated output that is substantially similar to their works, (iii) the model itself being an infringing derivative work, and (iv) providing the means for users to create infringing works. 

It remains to be seen whether current laws concerning IP infringement are adequate to deal with AI-generated output, or if reform is required.

Who would be liable for IP infringement?

The Landscape Report suggests that the issue of liability will likely be a fact-specific determination. The AI system itself cannot be held liable, but the user, programmer, developer, or owner of the AI system may be liable depending on the facts. It remains to be seen whether this issue can be resolved by the courts, or whether legislative intervention is required.

Key Takeaways

The Landscape Report provides a comprehensive summary of the complexities and legal issues surrounding AI innovation and IP rights, and a useful comparison of the views from commentators, courts and policymakers across multiple jurisdictions. As the AI landscape continues to evolve rapidly, particularly with the emergence of generative AI, this report will be a crucial resource for all stakeholders in the AI ecosystem, including owners, developers, and deployers of AI systems.

While the Landscape Report is not legally binding and is meant to collate information on key issues (rather than provide definitive answers), the views in the Landscape Report will be persuasive in any potential Singapore dispute involving AI. The Singapore courts often refer to commentary from academics and legal professionals when deciding on novel issues, and a report prepared by the local IP regulator will carry significant weight. 

Last but not least, it is important to remember that IP issues are but one aspect of the discussion and debate surrounding AI systems. There are numerous other issues which stakeholders should be aware of – our team’s commentaries on these issues can be accessed in the links below.

• AI governance: Singapore proposes Governance Framework for Generative AI | Inside FinTech
• Data protection: Singapore releases New Guidelines on the Use of Personal Data in AI Systems | Data Protection Report
• Employment: Employment hot topics: AI discrimination and bias

We would like to thank our trainee Judeeta Sibs, practice trainee at Ascendant Legal LLC, for her contribution to this post.