Emotional Perception AI: UK Supreme Court overrules Aerotel in favour of EPO’s approach to eligibility but leaves the application of the EPO’s “intermediate [inventive] step” to the UKIPO

Many readers will recognise the headline result of this decision from recent case reports. What is less obvious – and potentially more important in practice – is how the UKIPO is now expected to apply the EPO’s “intermediate step” within the UK’s established Pozzoli framework for inventive step. This article suggests a practical way the UKIPO could integrate that filter into Pozzoli without disturbing settled UK obviousness law.

In a landmark decision on computer-implemented inventions (in this case AI), the UK Supreme Court has overruled the Court of Appeal’s Aerotel approach to excluded matter under section 1(2) of the Patents Act 1977 (implementing Article 52  EPC). Instead, it held that UK courts are bound to follow the EPO Enlarged Board (EBA)’s approach in G 1/19 (the “any hardware” approach), under which the claimed invention was not excluded because it necessarily involved technical means.

However, what adds to the interest of this decision is also what it does not decide. The Supreme Court did not decide on novelty or inventive step, nor how those questions should be assessed in light of its Article 52 decision. It concluded that the subject matter of the patent was an “invention” and remitted the case to the UKIPO,  adding an “intermediate step” to filter which claim features contribute to the technical character of the invention viewed as a whole, before judging novelty and inventive step only by reference to those features.

How that intermediate step is applied in UK practice – and how it fits with the UK’s Pozzoli[1] approach to inventive step – will be the next battleground. The Supreme Court did not rule on whether this test needed to change to accommodate its ruling, leaving that to fuller argument before the UKIPO and, if necessary, the lower courts.

What has the Supreme Court decided?

The invention related to a media recommendation system. The application used a trained artificial neural network (“ANN”) to recommend files that are “semantically” similar based on objectively measurable properties (e.g. tempo, tone, volume and rhythm).

• Aerotel The Supreme Court held that the four‑step Aerotel framework should not be followed in light of the EPO Enlarged Board’s decision in G 1/19. The Court accepted the EPO’s criticism that Aerotel risks conflating the “invention” inquiry with novelty and inventive step by focusing on “contribution” to the art. It emphasised the policy and comity reasons for UK courts to follow established EPO jurisprudence unless convinced they are wrong beyond reasonable disagreement.
• ANNs are programs for computers. While it disagreed with the Court of Appeal’s description of an ANN as itself a “computer”, it held that an ANN is an abstract model capable of implementation on different types of hardware. The Court then held that, viewed as a whole, an ANN constitutes “in essence, a set of instructions to manipulate data in a particular way so as to produce a desired result” and therefore “an ANN is a program for a computer.”
• But ANNs are programs that are not excluded “as such”. Applying the EPO “any hardware” approach, the Court held that the claims were not excluded because they necessarily involved technical means (hardware implementation) and referred to hardware such as a communications network and user device. The Article 52 hurdle was therefore cleared. The Court allowed the appeal and remitted the application to the UKIPO for consideration of novelty and inventive step, incorporating the “intermediate step” used by the EPO of filtering out non-technical features before assessing those headings.

Analysis and implications

The new centre of gravity: the intermediate step

The judgment relocates the real work of patentability for computer-implemented inventions (including AI) into an “intermediate step” in the inventive step analysis – something new to UK patent law but used by the EPO. Although the Supreme Court did not apply this itself, it made clear that the intermediate step is a necessary filter. Because “any hardware” is such a low hurdle, decision‑makers must now identify which features contribute to the technical character of the invention “viewed as a whole” and disregard those that do not when assessing novelty and inventive step.

Uncomfortable bedfellows: filtering meets Pozzoli

Under G 1/19, features are assessed for whether they contribute to a “technical solution of a technical problem”, and the intermediate step serves as a filter so that “only those distinguishing features can contribute to inventive step“. That language sits naturally within the EPO’s closest prior art → objective technical problem → obviousness structure.

By contrast, the Pozzoli test for inventive step involves: (i) identifying the skilled person and their common general knowledge; (ii) identifying the “inventive concept”; (iii) identifying the differences between the prior art and that concept; and (iv) asking whether the differences were obvious. Pozzoli is not framed around an “objective technical problem”.

By remitting the case to the IPO, the Supreme Court has left open the question: where, in a Pozzoli analysis, does the intermediate step “sit”?

• One option is to build the filter into step (ii) (inventive concept). If the inventive concept is articulated broadly (including non-technical elements), this risks allowing excluded matter to drive the analysis of obviousness; but if the concept is “filtered” too aggressively, that risks stripping out features, when considered as a whole, that make the claim coherent, and leaving an artificially impoverished remainder.
• Another option is to apply the filter at step (iv), once the prior art has been identified and the “differences” are on the table. This approach has an obvious attraction: it mirrors the EPO tenet that only distinguishing features which contribute to technical character should be allowed to carry inventive step. However, it risks allowing non‑technical framing to creep into step (iii) only to be removed after it has already shaped the identification of differences relevant to obviousness.

The practical difficulty will be getting the analysis right where the alleged inventiveness lies purely in software, for example in the way the model is trained.

A practical way UKIPO might integrate the filter into Pozzoli

The Supreme Court’s most useful guidance appears at paragraph 106. It states that the intermediate step must be carried out by reference to the technical character of the invention “viewed as a whole”, and that it is open to UKIPO and UK courts to adopt “any appropriate method” to identify that technical character – giving, as an example, identifying the alleged “inventive concept” as explained in Pozzoli.

One way to reflect that guidance in the UK framework is a three‑stage approach:

1. Identify the inventive concept in the familiar Pozzoli
2. Apply the EPO’s “intermediate step” to that inventive concept. Ask, feature by feature, which parts contribute to the invention’s technical character when viewed as a whole, and which do not; only the contributing features are carried forward. This is where the UKIPO must be disciplined: the question is not whether a feature is “technical per se”, but whether it contributes in context (technical features can be filtered out; non‑technical features can survive if they interact with technical subject matter).
3. Proceed with comparing differences/obviousness using the filtered concept compared to the state of the art + CGK. Compare the prior art against the filtered inventive concept, identify differences, and assess whether those differences (looked at collectively) would have been obvious to the skilled person armed with common general knowledge. In doing so, one would treat only differences arising from the filtered concept (i.e. differences in contributing features) as capable of supporting inventive step.

This structure has two advantages. First, it implements the Supreme Court’s instruction to anchor the filter in the invention “viewed as a whole” (the inventive concept) rather than treating the claim as a collection of disconnected features. Second, it keeps the UK’s obviousness inquiry recognisably Pozzoli, while still ensuring that inventive step cannot be propped up by features that do not contribute to technical character.

Where next for AI and other computer-implemented inventions in the UK

Because the Supreme Court has remitted much of the practical application of its decision back to the UKIPO, many questions remain to be decided. Parties are likely to test how the UKIPO identifies the invention’s “technical character … viewed as a whole” and how strictly it filters “non-technical” features out or features that are not contributing to the technical character of the invention. If the UKIPO’s approach is seen as either too permissive or too strict, it is conceivable that the methodology itself becomes the subject of further appeals – potentially returning, in time, to the Supreme Court.

For applicants, the practical message is that the UK is now less likely to refuse AI and software applications at the Article 52 eligibility gateway. However, the battleground now shifts to showing that the features relied on for inventiveness contribute to technical character and survive the filtering exercise, before inventive step is assessed. For AI‑adjacent inventions, that places increased emphasis on how the technical character is articulated “viewed as a whole”, and on explaining how features interact with technical means.

[1] Pozzoli SPA v BDMO SA [2007] EWCA Civ 588