This article continues our series on Emotional Perception AI v Comptroller-General, which concerns the patentability of an artificial neural network (ANN) for recommending media files based on semantic similarity. The case returned to the spotlight with a two-day hearing before the UK Supreme Court on 21st and 22nd July 2025. The appeal follows the Court of Appeal’s decision, which defined “computer” and “program for a computer” under Section 1(2) of the Patents Act 1977, ultimately finding the invention excluded from patentability.
In this article, we summarise the main points of discussion from the hearing. For further background information, see our previous articles on the Court of Appeal decision here and here, and the Court of Appeal hearing here and here.
CIPA and IP Federation intervention
Prior to the Supreme Court hearing, the Chartered Institute of Patent Attorneys (CIPA) and the IP Federation filed a joint intervention addressing the definitions of “computer” and “program for a computer”, and the approach taken by UK courts when assessing the patentability of ANNs and other computer-implemented inventions. They stressed the need for harmonisation between the UK IPO and the EPO, expressing concern that the Court of Appeal’s approach, particularly the adoption of broad definitions, could lead to increasing divergence with EPO outcomes.
Correct approach for implementing the statutory exclusions to patentability
The statutory exclusions to patentability in Section 1(2) of the Patents Act 1977, including to programs for computers, are currently assessed in the UK using the four-step Aerotel test. During the hearing, Emotional Perception advocated that the UK courts adopt an approach based on the EPO’s two-hurdle approach confirmed by the EPO Enlarged Board of Appeal in G 1/19. The first hurdle of this approach involves assessing whether the claim includes at least one non-excluded (“technical”) feature, and is referred to as the “any hardware” approach because it is overcome by claiming any hardware. The second hurdle is the assessment of inventive step using a modified version of the EPO’s problem-solution approach, as set out in COMVIK (T 641/00), in which only features contributing to the technical character of the invention are considered for the purpose of inventive step. This latter step would be a particular point of contention for the Comptroller-General, especially as the Windsurfing/Pozzoli test for inventive step is favoured in the UK over the EPO’s problem-solution approach, as confirmed in Actavis v Novartis, and changing this would have major ramifications. Whilst Emotional Perception contended that using the problem-solution approach for assessing inventive step would not be necessary, and that an intermediate step of determining technical contribution could be introduced before assessing inventive step using Windsurfing/Pozzoli, the Comptroller-General referred to the theoretical combination of “any hardware” with Windsurfing/Pozzoli as a “Frankenstein test”. The parties did not agree whether the EPO approach theoretically and actually leads to different outcomes than the Aerotel test, but it was noted that this is the practical experience of the interveners. Reference was made to previous instances of the UK courts and EPO considering, and historically criticising, the other’s assessment of the statutory exclusions.
Revisiting the definitions of “computer” and “program for a computer”
In the decision on appeal, the Court of Appeal broadly defined a computer as “a machine that processes information” and a computer program as “a set of instructions for a computer to perform tasks”. Emotional Perception argued that these definitions go beyond normal language and the conventional understanding of these terms without a policy reason for doing so, and that the definition breadth would lead to perverse results, including the identification of ANNs as computers. The latter was also highlighted in the intervention. Emotional Perception also contended that the EPO does not use such broad definitions and cited Mitsubishi (T 0702/20), which the lower court had used to say that an ANN is a computer, in support. The Comptroller-General sought to reason that, taking into consideration the lower court’s decision as a whole, a computer must be understood to be not just a machine that processes information, but a programmable machine that processes information. This would mean an ANN is a computer, and the perverse results Emotional Perception and the interveners speak of fall away. It also remarked that a conventional understanding of “computer” and “program for a computer” is not sufficient because it fails to capture earlier technologies such as analogue computers and emerging paradigms such as quantum computers.
Beyond the program for computer exclusion
There was some debate about whether consideration of the mathematical method exclusion in Section 1(2) of the Patents Act 1977 was actually before the Supreme Court, since the Court of Appeal had not decided on this point. Nevertheless, the issue was discussed. The Comptroller-General said that the mathematical method exclusion generally applies to ANNs and would ultimately lead to unpatentability via the application of the Aerotel test. Consideration of ANNs as a program for a computer was signalled as being the UK IPO’s preference, whilst the EPO prefers treating ANNs as a mathematical method. Emotional Perception disagreed with the Comptroller-General’s conclusion, stating that the mathematical method exclusion only applies to abstract numbers, not where there is a practical application of the mathematical method, such as application to a digital media file, as shown in VICOM (T 0208/84). It also remarked that just because ANNs can be described as a mathematical function does not make them abstract mathematical methods, indeed most of science and engineering can be described in an abstract way using mathematics but that does not make such technology unpatentable. Exclusions to aesthetic creation were also discussed briefly but their relevance swiftly rejected.
Application of statutory exclusions to patentability to Emotional Perception’s ANN
The parties seemed to be aligned that the contribution of the claimed invention relates generally to the provision of file recommendations that are better in terms of their semantic similarity. This is said to be primarily provided by the special training performed by the ANN. However, the parties did not agree whether the contribution is a system that enables these improved file recommendations or just their enablement per se. The Comptroller-General maintained that the contribution provided by the claimed invention was wholly excluded, reasoning that the claimed hardware was not part of the contribution because it was old and irrelevant from the perspective of the ANN. The Comptroller-General also discounted better semantic similarity in the recommended files as providing a technical contribution, reasoning that semantic similarity is not technical, but subjective or cognitive. Reference was also made to on-going prosecution at the EPO, where the file recommendation is said not to qualify as a technical purpose in view of YAHOO! (T 306/10). Emotional Perception, on the other hand, argued that the claim includes many hardware features, such as the database, system and digital media file, which are technical, and because these interact with the ANN to provide the file recommendation, “salami slicing” them from the contribution is inappropriate. Emotional Perception also suggested it was an error of law to reject the application on the basis that the benefits of the invention are subjective or cognitive in nature. It was further argued that, even if the lower court’s definition of “computer” stands, such that ANNs are deemed to be computers, the claim results in a better computer, in line with one of the AT&T signposts. The AT&T signposts are used by the UK IPO and UK courts in conjunction with the Aerotel test to indicate when a computer program makes a technical contribution.
Conclusion
The Supreme Court has reserved judgment, and a decision is expected in due course. The hearing raises significant questions for how the statutory exclusions to patentability are assessed in the UK, and the outcome has the propensity to impact not only inventions involving ANNs and programs for computers, but also the broader list of statutory exclusions as well. We will continue to follow the case closely and provide further analysis once the judgment is handed down.