IPOS has issued a public consultation on the relevance of the decision of the UK Supreme Court in Emotional Perception AI Limited v Comptroller General of Patents, Designs and Trade Marks [2026] UKSC 3 to Singapore patent practice.

The consultation is significant for AI patents in Singapore because it may affect how IPOS examines computer-implemented inventions, artificial neural networks and claims involving software, data processing and mathematical methods.

The Emotional Perception case is significant because it concerns artificial neural networks, computer-implemented inventions and the boundary between patentable technology and abstract subject matter. It is also noteworthy as the UK Supreme Court rejected the long-standing UK Aerotel approach to excluded subject matter.

UK Supreme Court decisions are however not binding on the Singapore Courts. Such judgements may be persuasive, but only to the extent that the reasoning fits Singapore law. Furthermore, Singapore no longer has a statutory list of exclusions corresponding to Section 1(2) of the UK Patents Act 1977 or Article 52(2) EPC, which were discussed at length in the case.

However, we consider that Emotional Perception raises issues that are directly relevant to how AI and computer-implemented inventions are examined in Singapore.

Key Points

IPOS is considering whether Emotional Perception is relevant to Singapore patent practice.

The UK Supreme Court rejected the Aerotel “actual contribution” approach, but the decision is not binding in Singapore.

Singapore no longer has the UK/EPC-style statutory list of excluded subject matter (FAQ: Singapore patentable subject matter and exclusions), although IPOS examination practice still uses an Aerotel-derived analysis.

Cantab IP’s position is that IPOS should review that approach for AI and computer-implemented inventions.

Current Singapore Approach to Computer-Implemented Inventions

Section 13(1) of the Patents Act 1994 states that a patentable invention must be new, involve an inventive step and be capable of industrial application.

Unlike the UK Patents Act and the EPC, however, the current Singapore Patents Act does not contain a statutory list of excluded subject matter such as mathematical methods, business methods, presentations of information or programs for computers “as such”. These were previously set out in Section 13(2) of the Singapore Patents Act, but the Act was partially repealed on 1 January 1996 to remove that provision.

In practice, however, IPOS continues to refuse patent protection for such subject matter. Chapter 8 of the Examination Guidelines for Patent Applications at IPOS explains how this is currently handled in examination practice:

  • Paragraph 8.3 says that, if there remains an issue that the claims may relate to subject matter which is not an “invention”, the Examiner should undertake a separate analysis of the claimed subject matter. In that analysis, the Examiner should look at substance rather than form, identify the “actual contribution” made by the claimed subject matter and consider the problem to be solved, how the claimed subject matter works and what its advantages are.

  • Paragraph 8.4 then identifies this “actual contribution” inquiry as the second step of the Aerotel/Macrossan test. It quotes the well-known formulation asking what the inventor has “really added to human knowledge”.

  • Paragraph 8.5 says that an objection should be raised if the actual contribution lies solely in subject matter that is not an “invention”.

For AI-related inventions, IPOS has published the Supplemental Guidance for the Examination of AI-related Patent Applications. That guidance supplements the main Examination Guidelines and gives hypothetical examples of AI-related claims. IPOS states that AI-related applications are examined case by case, and that the claims must still satisfy the other requirements for patentability.

The Supplemental Guidance recognises that AI applications are found in many fields of technology. It gives examples ranging from collision avoidance, fraud detection and traffic prediction to drug discovery. It also recognises that patent protection may be sought for many aspects of AI-related inventions, including training datasets, machine learning methods and AI-specific hardware such as processors and memory chips.

The Supplemental Guidance however also advises an “actual contribution” approach (Section B(i), paragraphs 4 to 10).

Thus, even though Singapore no longer sets out excluded subject matter by statute, unlike Section 1(2) UKPA or Article 52(2) EPC, the IPOS Examination Guidelines and Supplemental Guidance still use an Aerotel-derived contribution analysis when deciding whether a claim defines an invention. The significance of Emotional Perception is that the UK Supreme Court has now rejected that very approach in the jurisdiction from which it came.

The question is therefore whether Singapore examination practice should continue to rely on an Aerotel-derived contribution analysis, particularly for AI and computer-implemented inventions?

What was Emotional Perception About?

The patent application in Emotional Perception concerned a system using an artificial neural network, or ANN, to recommend media files. The invention sought to identify files that were similar in a way that reflected human perception. The UK Supreme Court described the invention as using an ANN to replicate similarities and differences that a person might subjectively perceive between different files, by reference to objectively measurable physical properties of those files.

In simple terms, the invention was about better recommendations. However, the case was not just about recommendation systems. It also raised wider questions about whether an ANN is a program for a computer, whether the claim was excluded from patentability under UK law and how the UK should assess computer-implemented inventions.

The UK Supreme Court held that an ANN is a program for a computer. It reasoned that, whatever the form of the machine on which the ANN is implemented, the ANN constitutes, in essence, a set of instructions to manipulate data in a particular way to produce a desired result.

The Court also clearly rejected the Aerotel approach. It held that the UK should adopt the EPO’s “any hardware” approach to Article 52 EPC. Under that approach, the presence of hardware is enough to pass the initial threshold. The analysis then moves on to identifying which features contribute to the technical character of the invention and when assessing inventive step.

The Intermediate Step

The UK Supreme Court accepted that the “any hardware” threshold is low. Where a claim involves a computer or other technical hardware, that may be enough to pass the initial threshold. But that cannot be the end of the analysis. Otherwise, an abstract idea could be made patentable merely by reciting conventional computer implementation.

The Court therefore emphasised the application of an intermediate step. Before assessing novelty and inventive step, the examiner or Court must identify which features of the claim contribute to the technical character of the invention as a whole.

This should not simply be an exercise of labelling each feature as technical or non-technical in isolation. The reason is that a feature that looks non-technical on its own may contribute to the technical character of the invention if it interacts with technical features. Conversely, a hardware feature should not count merely because it is hardware, if it does not contribute to the technical character of the invention.

We consider that such an approach is useful for Singapore practice. Thus, the question should not be whether an invention can be reduced to a computer program, a mathematical method or a recommendation scheme when described at a high level. A better question might be to ask what the invention achieves in the real world, how it achieves that result using technical means and which features contribute to the technical character of the invention as a whole.

What is IPOS Asking in its Consultation?

IPOS has launched a feedback exercise asking for views on the relevance of Emotional Perception to Singapore patent practice. The consultation asks whether the decision is relevant to the development of Singapore patent practice, whether it brings greater clarity on patentable subject matter and its relationship with inventive step and what practical issues or challenges may arise for applicants and practitioners.

IPOS appears to be asking these questions because its current AI guidance uses an Aerotel-derived “actual contribution” approach. The UK Supreme Court has now rejected Aerotel for UK law. While that does not automatically change Singapore law, it does raise a fair question: should Singapore continue to use a contribution-based approach that was developed for a statutory exclusion framework that Singapore does not have?

In our view, IPOS is right to ask the question. AI inventions often involve a mixture of hardware, software, data, mathematical processing and real-world effects. A threshold test that asks too early what the inventor has “contributed” may strip away the technical implementation and treat the claim as nothing more than an abstract idea, mathematical method or computer program. That creates a risk of rejecting claims that may otherwise protect genuine technical innovation.

Cantab IP’s Response to the Consultation

Cantab IP has sent in its response to the consultation.

Our position is that Emotional Perception is relevant to Singapore, but only as persuasive authority. As the UK Supreme Court has no direct jurisdiction in Singapore, IPOS should therefore not adopt the decision mechanically. The reasoning must be considered in light of Singapore’s own Patents Act and Rules.

We have also emphasised that Singapore does not have the UK/EPC statutory list of excluded subject matter. Much of the UK judgment concerns the interpretation of Article 52 EPC and the UK statutory exclusions. As Singapore does not have the same statutory framework, Emotional Perception should not be treated as if it simply imports UK or EPO law into Singapore.

That said, we consider that the decision exposes a genuine problem with the Aerotel-derived “actual contribution” inquiry. Asking what the inventor has contributed before deciding whether the claim defines an invention risks mixing separate requirements of patentability. It may in an extreme case import considerations of novelty and inventive step into the threshold inquiry.

We have suggested that IPOS should review the current Supplemental Guidance for the Examination of AI-related Patent Applications. In particular, IPOS should reconsider whether the “actual contribution” step remains appropriate for Singapore practice.

A Better Singapore Framework

Our proposed approach is straightforward.

First, construe the claim.

Second, ask whether the claim is directed to a technical invention, having regard to the claim as a whole and the presence of technical means.

Third, identify the technical character of the invention as a whole.

Fourth, determine which claim features contribute to that technical character, including through interaction with other features.

Fifth, assess novelty and inventive step by reference to the features that make such a contribution.

This keeps the requirements of patentability separate. It also avoids importing a UK/EPC statutory exclusion framework into Singapore law. At the same time, it preserves a proper filter against claims where the hardware is incidental and the actual subject matter is non-technical.

Artificial Neural Networks

Our response also addresses the treatment of artificial neural networks.

The UK Supreme Court held that an ANN is a program for a computer. However, we do not think IPOS should adopt that reasoning uncritically for Singapore practice.

A conventional computer program is ordinarily understood as a sequence of executable instructions. A trained ANN, however, is technically different. Its behaviour may depend on network architecture, weights, biases, learned parameters, component values or physical configuration. Those features may determine how the system behaves, but they are not necessarily executable instructions in the ordinary sense of the term.

This issue is most clearly seen in analogue implementations. An analogue ANN or analogue computing circuit may have no CPU executing code. Its behaviour may arise from circuit topology, component values and physical signal flow. In that context, asking where the “program” resides is not simply a semantic question. It exposes a genuine technical problem with treating all information-processing systems as if they were conventional computers executing conventional programs.

Our view is that treating circuit configuration, component values, weights or biases as a “program” risks replacing technical analysis with an abstract legal classification. We consider that the better approach for Singapore practice is to examine the technical means used and the technical effect achieved, rather than to decide patentability by applying a broad legal label such as “computer program”.

The same caution applies to mathematical methods. ANNs plainly use mathematics, particularly during training and in the adjustment of weights and biases. But applied mathematics should not be treated as non-patentable merely because mathematical operations are involved. We consider that the proper inquiry is whether the mathematics is claimed in the abstract, or whether it forms part of a technical system which achieves a technical effect.

Practical Points for Applicants

AI patent applications should not rely on broad statements that the invention uses AI or improves recommendations. The specification should explain the technical problem, the technical means used to address that problem, the relevant feature interactions and the real-world technical effect achieved.

For ANN inventions, applicants should carefully explain the technical architecture, the role of training where relevant, the role of weights, biases, parameters or embeddings, the data being processed, the mechanism by which the output is produced and how those features interact to produce the claimed technical effect. In any event, it is good practice to draft patent specifications with sufficient technical detail and fallback positions.

That will be important whatever approach IPOS ultimately adopts. The stronger the technical explanation in the specification, the easier it will be to show that the claim is directed to a technical invention and that the relevant features contribute to the technical character of the invention.

Conclusion

While Emotional Perception does not change Singapore law, it is still a hugely significant case.

In particular, it gives IPOS a useful opportunity to reconsider whether the current Aerotel-derived “actual contribution” framework remains appropriate for Singapore, particularly for AI and computer-implemented inventions. In our view, Singapore should move toward a framework that asks whether the claim, considered as a whole, is directed to a technical invention, then identifies which features contribute to the technical character of that invention before novelty and inventive step are assessed.

That approach would better reflect Singapore’s statutory framework. It would also reduce the risk of artificial distinctions between software, digital hardware and analogue implementations, while preserving proper scrutiny of AI inventions under novelty and inventive step.

Further Discussion

This consultation is likely to shape how AI and computer-implemented inventions are examined in Singapore. The issues are not limited to artificial neural networks. They may also affect inventions involving machine learning, data processing, simulation, optimisation, recommendation systems, digital signal processing and software-implemented control systems.

Please feel free to get in touch if you would like to discuss any of these issues in more detail, or if you would like to review how the current IPOS approach may affect a particular AI or computer-implemented invention.

Contact Cantab IP

← Back to Blog