Striking a new chord: Will the UK harmonise with the EPO on excluded matter?

Following a two-day hearing last month, the UK Supreme Court is deliberating on a significant ruling concerning the patentability of artificial intelligence. The case, Emotional Perception AI Ltd v Comptroller General of Patents, tackles how the statutory exclusion for "a program for a computer ... as such" applies to Artificial Neural Networks (ANNs). The judgment is poised to clarify the UK's stance on AI inventions and address whether it should align with the European Patent Office's (EPO) approach on excluded subject matter.

Recommending Media by 'Emotional Similarity'

EPAI’s patent application (GB1904713.3) describes a system for providing “improved” media file recommendations, such as for music, based on their perceived emotional character. The system uses a trained ANN to analyse the physical properties of a new music file (e.g., tone, rhythm) to find other files likely to evoke a similar emotional response.

The core of the invention lies in its training process. One ANN processes natural language descriptions of music (like "happy" or "sad") to create a "semantic vector" representing human perceptions of the music. A second ANN is then trained to generate the same kind of vector, this time by analysing the music's physical properties. As Lord Briggs noted during the hearing, this ANN has then

"...learned through the training process what no human knows, namely, precisely what physical differences between different tracks produce the semantic difference on the vector".

Lord Briggs also reflected wryly on the implications of this approach: "It's distressingly utilitarian. I mean, I like to think that I have some wonderfully elevated reasons for thinking that there are certain similarities between Handel and Mozart, but actually it's all to do with some terribly undiscoverable physical characteristics”.

A Machine or a Program for a Computer?

A central theme of the hearing was whether ANNs engage the UK’s computer program exclusion.

EPAI’s "Machine" Argument: EPAI's counsel portrayed a trained ANN as a "structural machine". They argued that whether implemented in hardware as a physical circuit of "resistors and wires, amplifiers" or emulated in software, the ANN is a fixed functional structure. Once trained and "frozen", it no longer takes instructions but simply "works in accordance with its characteristics", much like any electronic circuit. This led to questions from the justices about how a hardware ANN could be trained, with counsel explaining that its components would be adjustable during training but fixed thereafter. From this perspective, the weights and biases are inherent properties of the machine's components, akin to "the resistance of the resistor or the gain of an amplifier".

The Comptroller's "Program" Argument: The Comptroller’s counsel presented a contrasting view, describing an ANN as having a "programmable architecture". They argued that the generic, untrained ANN is the hardware, while the set of weights and biases determined during training is the "program". These weights and biases are the "set of instructions which cause the machine to process the information in that particular way". This view was met with some scepticism, with Lord Leggatt questioning whether mere numbers could truly be considered a set of "instructions" or "commands" when they are not written as such at any level. Counsel countered by drawing an analogy to self-driving cars, where a file of weights and biases is loaded into the car's hardware to tell it how to drive, and can be updated just like any other computer software.

To counter EPAI’s narrower definition of a computer, counsel for the Comptroller introduced the examples of analogue and quantum computers, neither of which relies on a conventional CPU architecture. Counsel explained that analogue computers, which pre-date the 1977 Patents Act, were programmed using physical patch panels to wire up components to solve specific mathematical problems. They argued this was directly analogous to configuring an ANN, stating the analogy is "very, very close" and that an ANN is "much closer to an analogue computer than what we are familiar with as a digital computer". They then pointed to modern quantum computers, which also lack a conventional architecture but are supported by dedicated programming languages, reinforcing the idea that a "program for a computer" should not be limited to one type of hardware.

EPAI’s counsel challenged the broad definition of a computer adopted by the Court of Appeal as “a machine which processes information”, arguing it would lead to "perverse results" by capturing technologies that could not reasonably be considered computer programs, such as a simple electronic filter circuit or an electromechanical toaster.

In response, the Comptroller’s counsel drew a crucial distinction: the toaster's knob is merely a "user input" for a dedicated mechanical timer that "is always a mechanical timer". It lacks a "programmable architecture" that could be repurposed to perform a different task, unlike a modern dishwasher's microprocessor which could be reprogrammed to control an oven. They argued that the exclusion applies to devices with this underlying programmable nature, even if the program is permanently fixed for the end-user.

A Technical Output or a Subjective Benefit?

Beyond the debate over the ANN itself, the hearing also scrutinised the nature of the invention’s output. The UKIPO had dismissed the invention partly because its benefit—a better music recommendation—was deemed "subjective and cognitive" in nature.

Counsel for the Comptroller attempted to anchor this objection in the statutory exclusion for "aesthetic creations". However, EPAI’s counsel argued this was an error of law, as the statute does not exclude inventions based on their benefits, and the system itself does not create anything aesthetic; it only analyses physical properties and outputs a data file.

A key point of contention was the Court of Appeal’s reliance on the EPO's Yahoo case (T 0306/10). EPAI's counsel argued this case was misunderstood and irrelevant. They explained that Yahoo was a "broken chain" case, where a technical effect is dismissed because it is dependent on the unpredictable actions of a human user (e.g., a user choosing to click less frequently). In contrast, EPAI argued its invention produces a direct technical output—a new and improved recommendation file sent over a network—that is not dependent on any subsequent human behaviour to exist, meaning the chain of causation remains unbroken.

The Legal Frameworks: A Tale of Two Tests

The appeal's core legal dispute revolves around how to assess patent eligibility for computer-implemented inventions.

The UK Status Quo: The Aerotel Test: In the UK, courts and the Intellectual Property Office (IPO) use a four-step framework known as the Aerotel test to assess whether a claim is excluded from patentability:

Properly construe the claim.
Identify the actual (or alleged) contribution.
Ask whether the contribution falls solely within excluded matter.
Check whether the contribution is actually technical in nature.

A key aspect is the second step, which seeks to identify what the inventor has genuinely "added to human knowledge". In practice, this often means that known, conventional features (like a database or network) are not considered part of the contribution unless they interact with other features in a novel way. The test also incorporates a "rider" which clarifies that a contribution that is novel and inventive but consists purely of excluded matter (for instance, an aesthetic creation or a business method) does not count as a “technical contribution” for the purposes of the test.

The European Approach: Any-Hardware-Plus-COMVIK: The EPO uses a different methodology, often called the "two-hurdle" or "any-hardware-plus-COMVIK" approach. The first hurdle is a low bar: if the claim includes any physical hardware (the "any hardware" part), it is considered an "invention" and passes the initial eligibility stage. The substantive analysis happens at the second hurdle when assessing inventive step. Here, the COMVIK approach is used, in which all features that contribute to the invention's technical character are considered. Crucially, non-technical features (those that would be excluded if claimed alone) can still be part of the assessment if they interact with technical features to solve a technical problem.

The Opposing Views

EPAI: A Call for Harmonisation: Counsel for EPAI argued that the UK should align with the EPO's approach to avoid inconsistent outcomes when assessing excluded subject matter, by properly assessing an invention as a whole. Their primary criticism was that Step 2 of Aerotel wrongly "strips out" features before considering their interaction, which could lead to a different conclusion from the EPO's more holistic analysis. For example, they argued, under Aerotel, a conventional computer and data file might be stripped out, leaving only an unpatentable program, whereas the EPO's Vicom decision (T 0208/84) would see the interaction as a patentable, improved method of file analysis. This characterisation was met with some scepticism from the bench, with Lord Kitchin noting that he would “… need some persuading that it was not a part of conventional English law to take into account interactions in deciding whether or not there is an overall contribution or technical advance which justifies a patent”.

The justices showed a keen interest in the status of EPAI's parallel application before the EPO, though EPAI’s counsel appeared reluctant to be drawn on the details. After pressing the point, the court learned that the European application was on hold pending the UK decision and that an EPO examiner had raised an objection based on the "mathematical method" exclusion, but not the computer program exclusion. Counsel for both sides explained this was consistent with the EPO’s usual practice for ANNs. The "mathematical method" objection represents a significant alternative hurdle for EPAI, with the UK Court of Appeal noting that even if an ANN was not a computer program, an objection on this other ground "might well have had traction". The potential for divergence prompted concern from Lord Stephens, who described the EPO's contrary provisional view as a "fly in the ointment" and noted that,

“It would be somewhat inconsistent if we arrived at one decision and the EPO arrived at a different decision”.

The Comptroller: Defending a Distinct UK System: The Comptroller's counsel defended Aerotel, arguing it is substantively equivalent to the EPO's test, merely arriving at the same result through a different procedure. They characterized EPAI’s suggestion of combining the EPO's "any hardware" starting point with the UK's Pozzoli test for inventive step as a "Frankenstein test" that would be wrong in principle. They warned that simply adopting the EPO's approach would cause systemic disruption, particularly as the COMVIK approach is tied to the problem-solution method for assessing inventive step, which UK courts have deliberately chosen not to mandate. This would also fracture the UK's unitary approach to claim construction, which is used for both validity and infringement.

Several of the justices sharply questioned the Comptroller’s position. Lord Kitchin asked that if the outcomes were the same, “… would there not be an advantage in having the approaches brought into alignment?” Lord Leggatt similarly remarked that if there was no substantive difference, why was the Comptroller arguing that "the roof will fall in" if the UK were to adopt the EPO's method? Lord Stevens highlighted the central contradiction in both sides’ argument:

"The irony in this case is that the appellant [EPAI] says it would make a difference if we move to the any-hardware-plus-COMVIK, and you [the Comptroller] say it doesn't. And on the appellant's case, you've got problems about prospective/retrospective, and on your case, you don't".

Lord Leggatt also noted Lord Justice Lewison's previous "scathing" comments on the Aerotel test, suggesting it was not universally lauded in the UK, and going on to remark that,

“It doesn't seem as though everybody here thinks it's wonderful and that they do funny things on the continent which we should not follow”.

The justices also repeatedly raised the principle from cases like Symbian, which suggests UK courts should follow a settled approach of the EPO unless it is clearly wrong. With the EPO's approach now settled by the Enlarged Board of Appeal in G 1/19, the court is perhaps under pressure to justify any continued divergence.

Summing up

The Supreme Court is left to weigh two competing pressures.

On one side is the compelling principle, raised repeatedly during the hearing, that UK courts should follow a settled approach of the European Patent Office unless it is "clearly wrong". With the EPO's methodology for computer-implemented inventions now clarified by its Enlarged Board of Appeal in G 1/19, the pressure for harmony is significant.

On the other side are the Comptroller’s stark warnings about the practical consequences of such a change. Counsel argued that abandoning the established Aerotel test would require a major overhaul of the UK’s approach to assessing inventive step and would fracture the unified way patent claims are construed for both validity and infringement—a cornerstone of UK patent law. They cautioned against creating a "Frankenstein test" that could throw decades of domestic jurisprudence into "disarray".

Submissions made by CIPA and the IP Federation, as interveners in the case, also underscored the real-world stakes, arguing that the divergence between the UK and EPO already leads to different outcomes for inventors.

Ultimately, the court's decision, therefore, is not merely about whether a clever music recommendation system is a "machine" or a "program". The outcome will set an important precedent for how artificial intelligence and other software-based technologies are treated by UK patent law.