Ahead of the next chapter in the ongoing UK proceedings regarding the ”Emotional Perception” patent application (application number GB1904713.3), in this article we comprehensively re-cap the story so far and provide our thoughts on what we hope might happen next.
The case is being heard by the UK Supreme Court on 21 – 22 July 2025, though we do not expect their final written decision until the Autumn.
This article, and the Supreme Court’s decision, will be of interest to anyone wondering whether their or their clients’ AI innovations can be protected via the patent system in the UK and if so, under what conditions.
Introduction
In the UK (as in many jurisdictions), the Patents Act 1977 explicitly states that programs for computers are excluded from patentability as such. Of course, the Patents Act itself does not define “programs for computers”, and instead, patent attorneys rely on how this term has been interpreted by the UK courts over time (more on this below).
Originally filed in April 2019, this Emotional Perception patent application has called the UK’s current interpretation of “programs for computers” into question, which was established in 2009. In particular, the case is concerned with whether inventions relating to artificial neural networks (ANNs) are excluded from patentability as being “programs for computers”.
The UK’s current position (explained in more detail below) essentially states that only a relatively narrow set of software and AI inventions fall outside the exclusion. To fall outside of the exclusion, they must make a “technical contribution”.
Whilst most people understand ANNs to be computer-implemented inventions, one of the key questions that arose during the case is whether an ANN itself constitutes a “program for a computer”.
The case therefore asks two key questions:
- Is the ANN described by the Emotional Perception application considered a computer program?; and
- If the ANN is considered a computer program, does it fall solely within the exclusion of being a program for a computer, thereby excluding it from patentability?
Subject matter of the patent application
The Emotional Perception patent application relates to systems and methods for providing media file recommendations for a user. Specifically, the application relates to an arrangement which recommends media files to a user based on a similarity to media files that they have previously interacted with, where the similarity is measured both in terms of measurable quantities of the media files and a semantic (typically human-assigned) description of the media files.
The claims of the application are not limited specifically to music content however, it is against this backdrop that the matter has primarily been considered since this provides a readily understandable example use case. Considering the music content example, the claimed subject matter is able to recommend new music content based on a user’s previous listening history. This is enabled by the use of trained ANNs (where the claims include both the use of the trained ANNs and the training of the ANNs).
How are ANNs able to identify similar music files?
Broadly speaking, when performing analysis of a particular file (be it an audio file, image, video etc.), ANNs “see” the file as a vector of numbers, where each number effectively relates to one feature of the many features of the file. In order to determine similarity of two files, an ANN will typically measure a distance between the vectors of the two files in a multi-dimensional space known as embedding space, where a shorter distance means the files are more similar.
In the Emotional Perception application, two independent embedding spaces are considered: one representing the semantic descriptions of the music (e.g. “happy,” “soothing,” “exciting” etc), and one representing measurable properties of the music (e.g. speed, loudness, timbre etc). The separation distances of files in both spaces are provided to the ANN.
In the Emotional Perception application, the ANN is trained to adjust the separation distance between two files in the property embedding space to reflect the separation distance in the semantic embedding space. That is, if the two files are closer together in semantic space than in property space, the property separation space is reduced.
This results in a trained ANN that is able to account for both semantic and measurable property similarity when determining whether two music files are similar, which is thereby able to make music file recommendations to a user based on this similarity between music files they have already listened to, and a database of other available music files.
What is a computer program “as such”?
The legal test to determine whether a computer-implemented invention falls solely within excluded subject matter was established over 15 years ago in Aerotel Ltd v Telco Holdings Ltd & Ors Rev 1 [2007] RPC 7.
This test comprises four steps:
- Properly construe the claim.
- Identify the actual contribution (although at the application stage this might have to be the alleged contribution).
- Ask whether it falls solely within the excluded matter.
- If the third step has not covered it, check whether the actual or alleged contribution is actually technical.
Thus, the assessment of excluded subject matter hinges on whether or not the claimed invention makes a technical contribution. Subsequent cases have helped flesh-out what is considered to be “technical” in the “program for computers” space.
These are now summarised in the AT&T signposts (AT&T Knowledge Ventures/Cvon Innovations v Comptroller General of Patents [2009] EWHC 343 (Pat)). The signposts essentially say that a program for a computer is technical if it has a technical effect on a process outside the computer on which the program is run, or has a technical effect of improving the functioning of the computer itself when run on that computer.
The Aerotel test and the AT&T signposts only need to be applied if the claimed invention is considered to be a computer program falling within the exclusion. As noted above, the Emotional Perception case asks whether an ANN is a computer program itself.
Outcome during patent prosecution before the UK IPO
The Emotional Perception application progressed through several rounds of examination before ultimately arriving at a hearing between the Applicant and the UK IPO in June 2022. The hearing addressed whether the claimed invention fell within the “program for a computer” exclusion, and if so, whether it makes a technical contribution.
The Applicant presented arguments highlighting that the claimed ANN could be implemented through hardware (i.e. a physical box with electronics in it) and therefore should not be construed to only be a program for a computer. However, this was found unpersuasive in light of the lack of detail provided in the application as filed regarding such a hardware implementation.
It was also argued that the provision of the file itself was technical in nature. However, this was again found to be unpersuasive since the file recommendation had no technical effect on a process outside of the computer (as per the AT&T signposts). The recommendation is simply to provide files that are semantically relevant for the enjoyment of a user.
High Court Ruling
The Applicant decided to appeal the negative decision of the UK IPO at the hearing. This meant the case was reviewed by the UK High Court (Emotional Perception AI Ltd v Comptroller-General of Patents, Designs and Trade Marks [2023] EWHC 2948 (Ch) (21 November 2023)). The Applicant firstly submitted that the computer program exclusion should not apply at all since they did not consider their ANN to be a program for a computer. The second submission was that even if their ANN was considered a computer program, it makes a technical contribution.
In relation to the first of these points, the High Court judge considered the inherent differences between the use of an ANN which is implemented purely in hardware, and an ANN that is emulated in software. Due to the submissions presented to the High Court, the judge reasoned that in the case of a hardware ANN, there is no program to which the exclusion applies, whereas in the case of a software emulated ANN, there must be. The judge concluded that since the claimed ANN could be hardware-based, the exclusion could not apply.
Furthermore, the judge concluded that, since the software emulated ANN is simply operating according to something it has learned itself (i.e. the weights of the ANN), the software implementation is not implementing code provided by a human programmer. As such, the software emulated ANN is not running a “program” per se and as a result, the exclusion equally does not apply to this implementation.
In essence, in November 2023 the High Court effectively ruled that a claim to the use of ANNs (either software or hardware based) should not be considered to fall within the programs for computers exclusion. This was a sharp departure from the long-established approach to computer-implemented inventions in the UK.
The immediate result of this decision was, unsurprisingly, a flurry of AI and machine learning applications being filed and prosecuted quickly (and granted) at the UK IPO in order to take advantage of this newly relaxed stance.
Court of Appeal Ruling
Unsurprisingly, the UK IPO appealed the decision given how much it departed from established law, and the case was heard by the Court of Appeal ((Comptroller General of Patents, Designs and Trade Marks v Emotional Perception AI Ltd [2024] EWCA Civ 825 (19 July 2024)).
The judges in the Court of Appeal firstly considered what is meant by a “computer”, and then by a “program for a computer”. They concluded that a computer is a machine which processes information, and a program for a computer is a set of instructions for a computer to do something.
Importantly, the judges further remarked that a program for a computer does not necessarily have to be written by a human programmer. Therefore, the weights of an ANN could, according to the judges, be held to be a computer program, even though they are not determined by a human programmer, but learned during training of the ANN.
While there was a distinction made between hardware and software ANNs in the High Court judgement (i.e. ANNs that are implemented using hardware components, and those that are implemented using a general purpose computer), the Court of Appeal held a different view. In particular, the judges stated that whether the ANN is implemented in software or hardware makes no difference in whether it is a computer program. That is, both the hardware configuration, and the software used to implement ANNs, were held to be computer programs.
Given all these definitions, the judges found the actual contribution of the invention (as set out in step (2) of the Aerotel test above) to essentially relate to providing improved file recommendations using semantic characteristics of the files, which is not itself a technical contribution under the guidelines of the AT&T signposts. This is the same contribution as that initially identified by the Hearing Officer. As mentioned above, using semantic properties, i.e. properties related to language, or subjective qualities of language, is usually not considered technical. The invention in this case uses natural language descriptions of sound files to find sound files with other semantically similar descriptions.
This was not considered technical by the Court of Appeal judges. This is because both using semantic properties, and making recommendations based on subjective user preferences and taste are not considered to be technical purposes, and therefore are not associated with a technical effect.
While the claimed invention includes an additional step of transmitting the file, it was also held that this step could not contribute to a technical effect because it is entirely conventional – the unconventional aspect of the invention is the way in which a file is recommended, and because this is not considered technical, neither is the entire claim.
Thus, the Court of Appeal disagreed with the findings of the High Court, causing a quick end to the relaxed stance that had existed for about eight months.
The judgment also emphasised that this puts ANN inventions, and artificial intelligence inventions in general, on equal footing with other computer implemented inventions. That is, the same criteria applies to ANNs as it does for “traditional” software-based inventions.
Supreme Court Ruling
Unsurprisingly, the Applicant was not happy with the outcome of the Court of Appeal, and they have therefore appealed the decision to the highest court in the UK, the Supreme Court. The five Supreme Court judges are due to consider the case on 21 and 22 July. They will be considering whether the UK Patents Act exclusion for a “program for a computer” applies to ANNs and if so, whether it prevents the Emotional Perception patent application from being granted as a patent. We will report the decision of the Supreme Court when it has been handed down.
Importance of this case
The outcome of this case is important to deeptech innovators and businesses in the UK, as well as to non-UK businesses seeking protection for their AI inventions in the UK market, and so we will be waiting for the judgment with bated breath.
Unless the Supreme Court decides to radically depart from established UK law and from keeping the UK approach in line with the European Patent Office, we think it is likely that the Emotional Perception application will be considered to fall within the exclusion. Whatever they decide, we would like clarification and certainty on whether AI innovations, including those that use neural networks and those that use other types of model, should be assessed in the same way as non-AI-based software. Although the UK has been a somewhat difficult jurisdiction in which to patent computer-implemented inventions of any type, an outcome that ensures consistency between all types of computer-implemented invention is much more desirable than one that positively favours ANNs only.
In our minds, there should be no distinction between the type of computer-implemented invention when it is assessed by the UK IPO. The question should be about whether the computer-implemented invention or program for a computer makes a technical contribution to the state of the art. This would also bring the UK IPO more in-line with the European Patent Office. If the Supreme Court decides that ANNs are not programs for computers, that would leave doubt about whether all types of machine learning fall outside of the programs for computers exclusion. For instance, it would be undesirable for ANNs to be patentable, but for Bayesian networks and diffusion models to not be patentable.
It would also be undesirable, and short-sighted, for the Supreme Court to try to define a computer or program for a computer in order to avoid such questions being raised again. It is almost inevitable that definitions will be stretched, questioned and possibly become redundant in a field that is moving so fast.
Ideally, the Supreme Court will use this opportunity to expand out what “technical contribution” means in the computer-implemented invention space. As noted above, the AT&T signposts suggest limited ways one can demonstrate that a computer-implemented invention is technical. However, the AT&T signposts were collated in 2009, in a world before AI-based image and text generation, AI-driven telecoms networks, AI-powered drug discovery, and before talented researchers and scientists worked out how ordinary people could use AI on their own smartphones and laptops. The UK IPO currently dismisses extremely clever AI-based inventions if they do not fall within the narrow AT&T signposts. It would be better for innovators and UK-based deeptech businesses for there to be a more modern approach to deciding what is technical that keeps pace with innovation.
