< Back to all publications

Protecting AI inventions in Singapore

The rapid advancement of artificial intelligence is profoundly impacting business and society, with AI now possessing the ability to write code, design drugs, develop products, and overhaul processes, among other innovations. This emergence of AI has sparked inquiries into the role of patent laws in protecting AI inventions.

Patent protection can significantly enhance the commercial opportunities for AI inventions. Conversely, without patent protection, there is a risk that these inventions may be kept as trade secrets, which is a less secure option and can affect transparency requirements in fields like life sciences. Such a scenario would weaken the core principle of providing protection in exchange for disclosing inventions to the public, which is a vital element in stimulating further innovation.

Countries worldwide have faced several key issues on AI inventions, such as AI inventorship for patents, the patentability of AI inventions, and the ownership of AI inventions. This article specifically addresses these legal issues at the intersection of AI and the patent system in a Singaporean context.

Can an AI invention fulfil the patentability requirements?

Patent protection grants the inventor an exclusive, time-limited right to prevent others from using the invention, in exchange for disclosing the details of the invention to the public, enabling others to innovate based on this knowledge. This reciprocal arrangement creates a beneficial cycle that drives societal progress.

Patentability criteria:

  1. Patentable subject matter: To be patentable, the subject matter must not fall into the prohibited categories under the Singapore Patents Act, including discoveries, scientific theories, mathematical methods, aesthetic creations, schemes, rules, methods for mental acts, gaming, business and information presentation.
  2. Novelty: The invention must possess new features not found in existing knowledge within its technical domain, known as the prior art.
  3. Non-obviousness/inventive step: A patentable invention must demonstrate an “inventive step” which is not obvious to a person skilled in the art, considering the state of the art (Section 15 of the act).
  4. Industrial application: The invention should be capable of use for commercial or industrial purposes.
  5. Sufficiency of disclosure: The patent application must also completely and clearly describe the invention, enabling replication by someone with ordinary skills in the relevant technical field, for it to be patentable.

AI-related or AI-generated inventions fall into the non-patentable mathematical method category if described solely through mathematical processes. Merely stating that an AI method runs on standard computer hardware is usually insufficient for patent protection.

An AI method is considered patentable if implemented on a computer to solve a specific problem such as machine learning for speech or image recognition and must be functionally limited or causally linked to solving that problem.

However, there has been substantial debate on whether and how AI-related and AI-generated inventions can satisfy the requirements of non-obviousness and sufficient disclosure.

Non-obviousness: The requirement of non-obviousness was initially conceived with human inventors in mind. What may be perceived as non-obvious by a human might appear insignificant to AI considering AI’s capabilities. Legal standards might require adjustment to accommodate these technological advancements, particularly as inventors increasingly employ AI to innovate.

Sufficiency of disclosure: In Singapore, Sections 25(4) and 25(5)(c) of the act require clear and complete disclosure, enabling replication by a person skilled in the art.

However, determining sufficient disclosure for AI-related or AI-generated inventions may be difficult due to the rapidly evolving AI field. There are suggestions that AI-related patent applications should include disclosing the correlation between input and output data, explaining the AI decision-making process, predicting AI decisions and outputs, and disclosing the data sets used for AI training to meet the sufficiency of disclosure requirement.

Although the Act mandates that a patentable invention should meet the criteria of non-obviousness and sufficient disclosure, it is yet unclear how these criteria will be fulfilled concerning AI-related and AI-generated inventions.

Can an AI system be identified as an inventor?

An AI system cannot be designated as an inventor of a patent in Singapore.

Section 24(2) of the act necessitates the identification of the inventor or justification of patent rights acquisition within a patent application; failure to adhere thereto results in abandonment of the patent application. This section further requires the inventor to be a natural person.

Section 2(1) of the act defines the inventor as “the natural person who came up with the inventive concept,” as interpreted by Singapore courts.

Section 24(1) of the act grants the inventor a moral right, which is personal and non-transferable, exclusively applicable to natural persons, suggesting corporations cannot be designated as inventors.

Who is the owner of an invention?

AI lacks legal personality, capacity, or rights, thus cannot own patents under the current Singapore patent laws. Possible proprietors of patents related to AI or generated by AI can be the AI’s owner, user or developer.

Section 19 of the act allows patent ownership by inventors, those entitled by law, foreign laws or agreements, and successors in title. Unlike inventors, who are natural persons, owners can be legal entities like corporations, with transferable ownership for flexibility in rights assignment.

Conclusion

Singapore is well-positioned to embrace the era of artificial intelligence, acknowledging its crucial role in driving the digital economy. Addressing challenges in protecting AI inventions is essential, given the rapid evolution and widespread integration of AI technologies in various sectors of the society.

This article was first published in the IP Analysts column of Asia IP.