New UK Supreme Court decision paves the way for more patents relating to AI
For many years, the UK has been a somewhat unfriendly jurisdiction for innovators seeking patent protection for software-related inventions, particularly compared to countries like the USA (which, quite literally, starts from the premise that ‘everything under the sun’ should be patentable).
The UK Intellectual Property Office (the UK’s patent-granting body), on the other hand, has been hamstrung for the last two decades by a somewhat reductive judge-made legal test, which essentially ruled out all patents for any innovation which hinged on software, on the basis that this could not be a ‘technical contribution’, the reason being that ‘programs for computers’ were specifically ruled out by legislation (the UK Patents Act 1977).
The UK’s Supreme Court has now discarded this test, in a major and highly prescient decision, which should open the door to a more patent friendly regime in the UK for inventions concerned with software and AI in particular.
The patent application
The patent application in question was made by Emotional Perception AI Limited. The UK Intellectual Property Office’s patent examiner had rejected Emotional Perception’s patent application, effectively on the basis that it was concerned with a computer program. On appeal, the High Court had disagreed. The Court of Appeal then reinstated the original rejection.
All three decisions had referred to the long‑standing Aerotel test, set out by the UK Court of Appeal in Aerotel Ltd v Telco Holdings Ltd [2006] EWCA Civ 1371; [2007] RPC 7. As alluded to above, the Aerotel decision had resulted in a broad exclusion from patentability, based on an interpretation of section 1(2)(c) of the Patents Act and the reference in it to ‘programs for computers’. The key issue before the Supreme Court was whether the test set out in Aerotel should still apply.
What was the technology?
This decision thus concerned the patentability of Emotional Perception’s neural network-based technology. An artificial neural network (ANN) is a layered network of interconnected nodes that process data by passing weighted inputs through activation functions to determine how strongly the ANN should respond to the information it receives.
During training, the ANN repeatedly adjusts its weights and biases, to minimise the error between its output and a known target output. Once training is complete, the weights are set, and the ANN can infer outputs from new data using the patterns it has learned. ANNs can be implemented in software on conventional hardware, or they can be embedded directly into specialised hardware architectures.
Emotional Perception’s ANN is trained to analyse the emotional or semantic similarity between media files (such as music files), using only objectively measurable physical properties. The system is trained on pairs of files that have both human‑derived semantic similarity scores (based on natural‑language) and physical similarity scores (based on features like tempo, tone and rhythm).
In practical terms, the system learns to map quantifiable features like tempo, rhythm, tone and volume dynamics onto the kinds of subjective labels humans naturally use to describe music, such as ‘uplifting’, ‘melancholic’ or ‘euphoric’. The outcome is an ANN that is trained to recognise when two media files feel emotionally similar, even though it examines only their measurable, musical features.
By learning how the objective musical traits of a media file relate to the way humans describe their emotional responses to that file, the system can take a new song, video or text, analyse its raw properties, and recommend other files likely to trigger a similar emotional response, without relying on tags, metadata or human descriptions. Emotional Perception stated that its invention meant that it could make better file recommendations more quickly and more accurately than anything else that was then available.
It is important to note that a finding of the Supreme Court in this case was that an ANN such as this was a ‘program for a computer’. Therefore, on the face of it, it was excluded from patentability under section 1(2)(c) of the Patents Act (itself based upon articles 52(2)(c) and (3) of the European Patent Convention (EPC)).
The Aerotel test
In Aerotel, the UK Court of Appeal had sought to clarify how the exclusion of ‘computer programs… as such’ in the legislation should be applied in such cases. In Aerotel, the invention was a telephone pre‑payment system that relied on a new piece of hardware called a ‘special exchange’. The court had decided that the patentee’s contribution was the entire telephone system, which was new, technical, and not merely a business method (which was one of the other exclusions), even though it was used for selling phone calls. The system claims were therefore patentable.
The Court of Appeal suggested the following four‑step test:
- Properly construe the claim. Determine what the invention actually is.
- Identify the actual contribution. Determine the substance, not just the form, of the invention’s contribution to the state of the art.
- Ask whether it falls solely within the excluded subject matter. Determine if the contribution is only a business method, computer program etc (as referenced in section 1(2)(c) of the Patents Act) and therefore excluded from patentability.
- Check whether the contribution is technical in nature. Determine if the invention solves a technical problem.
Subsequent UK cases have refined this approach. For example, in Symbian Ltd v Comptroller General of Patents [2008] EWCA Civ 1066, an improved method for accessing files in a dynamic link library was found to be patentable, because it made a computer run faster and more reliably, which was an accepted technical effect.
Meanwhile, in the European Patent Office…
However, in Duns Licensing Associates (Decision T 154/04) [2004] EPOR 10, the European Patent Office’s Technical Board of Appeal had rejected the Aerotel approach as being inconsistent with the EPC. Instead, the EPO advocated the application of the so-called ‘any hardware’ approach (that had previously been specifically rejected by the UK Court of Appeal in Aerotel), meaning that, if a patent claim features the use of any physical hardware, the claimed invention will not be excluded from patentability under article 52(2)(c) of the EPC, even if it is partly to a computer program.
That position was later reinforced in the EPO’s Enlarged Board decision in G1/19 [2021] EPOR 30, which expressly supported the ‘any hardware’ approach set out in Duns. The key point arising from these decisions is that a patent claim to a computer program which would otherwise be prima facie excluded from patentability may not be so excluded, if the claimed invention uses or relies upon any physical hardware.
In addition, the Court of Appeal in Aerotel had interpreted section 1(2)(c) of the Patents Act as being only concerned with whether claimed (but excluded) subject-matter is an ‘invention’. According to the Court of Appeal, if the technical contribution of the patentee was excluded by this provision in the legislation (eg because it was a computer program, or a business method), then the analysis need not go further and it was the end of the road – the patent application would be rejected.
The EPO in Duns had criticised this aspect of Aerotel, for conflating the concept of an ‘invention’ in article 52(2) of the EPC with the other patentability requirement of inventive step. By contrast, G1/19 sought to keep these issues separate. Under G1/19 and Duns, the test was: first, to determine whether the claim has ‘technical character’ (and may therefore be an invention); second, and only after step 1, to assess novelty, inventive step and industrial applicability.
The Supreme Court overrules Aerotel
The Supreme Court decided to reject the Aerotel test and endorse the EPO’s significantly less demanding approach to the meaning of the relevant exclusion, based on ‘any hardware’. Thus, although Emotional Perception’s claimed method involves an ANN, which is a program for a computer, it is not to be excluded from patentability under section 1(2)(c) of the Patents Act (cf article 52(2)(c) of the EPC). This was because the claimed ANN could only be implemented on computer hardware.
In addition, the patent claims referred to a database for storing data files, a communications network and a user device, all of which required or constituted hardware. This meant that the subject matter of the patent claims had technical character and was not confined to a computer program ‘as such’. The Supreme Court therefore held that the UKIPO was wrong to refuse the application.
Having resolved this threshold issue, the court explained that it would now fall to the UKIPO to assess the remaining aspects of patentability (novelty, inventiveness and industrial applicability). These questions were referred to the hearing officer for further consideration.
Conclusion
After years of adherence to the harsh reality of the Aerotel test, the UK Supreme Court has now significantly liberalised patenting in this important area of technology. For AI developers, this means that tools and models that rely on computer hardware may now clear the threshold requirement, and inventions in this sector can be assessed on a more equal footing with inventions in other areas of technology. This should result in a significantly more patent-friendly approach of the UK Intellectual Property Office towards AI and AI-related inventions in the future.
This article co-written by James Notton, associate in the commercial, IP and IT team.