On 21st and 22nd July 2025 the UK Supreme Court heard the much-anticipated case Emotional Perception AI Limited (Appellant) v Comptroller General of Patents (Respondent), which is the first time the Supreme Court has considered the patentability of inventions relating to AI.
Now the dust has settled, we discuss the arguments put forward by the parties in the Supreme Court.
Background
Emotional Perception (EPAI)’s patent application was directed towards an artificial neural network (ANN) for providing improved file recommendations (e.g. songs) which are semantically similar to an input file based on the file's physical characteristics.
The UKIPO initially rejected the patent application on the basis that the ANN falls within the “program for a computer” exclusion of section 1(2)(c) of the Patents Act.
EPAI appealed this decision to the High Court, where it was overturned (see more here). The UKIPO then appealed the High Court judgment to the Court of Appeal, where it was, once again, overturned, restoring the UKIPO’s original findings (again, you can read more here).[1]
EPAI was then granted leave to appeal the Court of Appeal’s judgment to the UK Supreme Court.
The questions central to the appeal were:
- Does the statutory restriction on patenting a “program for a computer…as such” apply to artificial neural networks (“ANNs”)?
- If so, does it prevent EPAI’s application from being patented?
The submissions by both the UKIPO and EPAI focus on definitions of a computer and a computer program, and how these apply to ANNs, and the appropriate approach to determining whether a claim is excluded under section 1(2) of the Patents Act.
UK vs EPO Approach to assessing exclusions for computer programs
The four-step Aerotel test is currently used to assess the exclusions set out in section 1(2).
This test is performed before any assessment of novelty or inventive step, and involves determining whether the actual contribution disclosed in the claim falls “solely within excluded subject-matter”.
In contrast, the established EPO approach to assessing excluded subject-matter under Art. 52 EPC is the “two-hurdle” approach consisting of 3 steps, as set out in G1/19, also referred to as the “any hardware + Comvik” approach[2]:
- Does the invention include at least one non-excluded feature?
- (Intermediary step) If yes, which features contribute to the technical character of the invention?
- Inventive step assessment using the Comvik test (problem-solution approach) of those features contributing to the technical character of the invention
Before the Supreme Court, EPAI advanced that the UK approach should align with the EPO approach when construing a claim in order to assess it for inventive step by adopting an “any hardware + Comvik or Pozzoli” approach, emphasising the essential ‘intermediary step’ 2 of the two-hurdle approach.
But what might an “any hardware + Pozzoli” approach look like?
The Pozzoli test is currently the preferred test for assessing inventive step in the UK, and involves identifying the inventive concept of the claim; identifying what differences exist between the inventive concept and the knowledge of the skilled person and the prior art; and then lastly determining whether the inventive concept would be obvious to the skilled person.
In contrast, the Comvik approach involves determining a technical effect that solves an objective technical problem (step 3 above), so under the EPO approach, the technical character of the invention is considered once again in the assessment of inventive step, not just in the initial any hardware stage (a fairly low bar).
Lord Stephens pointed out that an “any hardware + Pozzoli” approach might lead to a single assessment of excluded subject-matter occurring only at the ‘any hardware’ stage (step 1 above).
The UKIPO argued that the proposed new approach could lead to situations in which a narrower construction of the claims is adopted during prosecution (by limiting the construction of the claims in the ‘intermediary’ step 2 of the two-hurdle approach) while leaving open the possibility wider construction for post-grant proceedings, such as infringement, which they argued would disrupt the principle of using the same construction of the claims throughout a patent’s lifecycle for all legal proceedings.
They went on to argue that removing the well-known and understood Aerotel test could cause difficulties for the UKIPO, and that the test being put forward by EPAI was not sufficiently clear. The UKIPO defended the Aerotel test and stated that the ‘hybrid’ approach of combining any hardware with Pozzoli would constitute a “Frankenstein test”.
The UKIPO also noted that the Aerotel test, provides an opportunity for examiners to determine whether an application is allowable under section 1(2) before search, keeping costs at the UKIPO comparatively low (although often much to the frustration of patent practitioners and applicants alike!).
Definitions of “computer” and “program for a computer” applied to an ANN
In their judgement, the Court of Appeal provided broad definitions for a computer: “a machine which processes information”; and a computer program: “a set of instructions for a computer to do something”.
The breadth of these definitions has raised concerns that they might cause the UK’s approach to ANNs and computer-implemented inventions to diverge from EPO outcomes.
During the Supreme Court hearing, EPAI advanced that the law must be read using a conventional understanding of the language, in the absence of any policy reason or clear indication that something else is intended.
They put forward that a “computer” in the Act must be understood as a ‘conventional computer’, which contains a central processing unit (CPU), and that a ‘conventional program’ is a series of logical commands that instruct the CPU to execute a specific function. EPAI advocated that an ANN does not include either of these features as they are conventionally understood.
The UKIPO modified the Court of Appeal’s proposed definition specifying that a computer is a programmable machine which processes information, arguing that this specification resolves any potential perverse results that may lead from the CoA’s broader definition (think electromechanical jukeboxes and toasters).
The UKIPO argued that the definition of computer cannot simply be narrowed to a machine containing a CPU by citing analogue and quantum computers as machines that are commonly understood to be computers but do not contain a classical CPU. The UKIPO submitted that the structure or architecture of an ANN (links and nodes) is a computer and that the weights and biases are the computer program that allows the computer (ANN architecture) to perform its function, in line with the reasoning provided by the Court of Appeal.
Application to the EPAI claim
EPAI asserted that the interactions between the ANN and the hardware features, the database, the user device to which the recommendations are sent, etc., are essential and must be considered as contributing to the technical contribution of the invention.
They argued against “salami slicing” features that contribute to the technical character of the invention for the assessment of excluded subject-matter. EPAI also stated that even if the invention were considered to be a program for a computer it would be allowable as a program that makes the computer a “better computer” as defined in the AT&T signposts.
The UKIPO presented an assessment on excluded subject-matter performed by the UKIPO using Aerotel and a second assessment performed by the EPO in the first examination report of the co-pending EPO application, which includes an excluded subject-matter objection using the any hardware + Comvik approach. They also rely on Yahoo! (T 306/10) to assert that a file recommendation is not a technical contribution.
Conclusion
The Supreme Court itself seemed somewhat divided on the issues during the hearing, with counter opinions on ‘subjective’ technical effects making themselves known from the questions asked.
Certainly, nothing was given away by the judges, who gave both EPAI and the UKIPO quite a grilling. Given Lord Kitchin returned from the supplementary panel in order to sit in, we predict a relatively short period between the hearing last week and the handing down of the judgement.
Plainly, the outcome of this case has the potential to shift the direction of AI and other computer-implemented invention patentability in the UK. The Supreme Court has reserved judgement, and if pressed, we would suggest that greater alignment with the EPO will be the end-result of any decision (whether that be direct adoption of the COMVIK approach, or a modification to the existing UK practice).
We will watch this space and will provide further updates on this case.
References
- https://supremecourt.uk/cases/uksc-2024-0131
- https://www.epo.org/en/legal/case-law/2022/clr_i_a_4.html
