The Emotional Perception saga has come full circle. On 1 October 2026 the UKIPO refused the remitted application again, this time for lack of inventive step (BL O/0894/26). It's one of the first decisions to apply the approach set by the Supreme Court in Emotional Perception AI v Comptroller [2026] UKSC 3, and in particular its new "intermediate step", discussed in our previous article.
As a quick recap, the Supreme Court did away with the funk of twenty years of Aerotel and brought the UK into line with the EPO's approach in G 1/19. It adopted Duns principles A to F. In short:
- there are four requirements for patentability: an invention, novelty, inventive step and industrial application
- an invention must have technical character
- eligibility, novelty, inventive step and industrial application are separate requirements
- anything with technical character isn't excluded, even if it relates to an excluded category such as a computer program
- the claim must be construed to identify its technical features
- a claim can mix technical and non-technical features, but only features that contribute to technical character count towards novelty and inventive step
That last principle is where the intermediate step comes in. It sits between the "any hardware" test, which almost any computer implemented claim now passes, and the assessment of inventive step. It acts as a filter. Features that don't contribute to the technical character of the invention, viewed as a whole, are set aside, and only the remaining features are compared with the prior art. A non-technical feature can still count if it interacts with the technical subject matter to help solve a technical problem. The Supreme Court left it to the UKIPO to decide how the step should be carried out.
How the decision was reached
The claim concerns an ANN (artificial neural network) trained to recommend files, such as music, that a user will find emotionally suitable. Emotional Perception's representative declined to define the technical character of the invention, and so the Hearing Officer therefore defined it himself: computer hardware running a program that selects and outputs a "semantically better" file.
For the Hearing Officer, that reduced the analysis to one question: is a semantically better file a technical result? If it were, he was confident of finding an inventive step. He held that it isn't. The file is better only in the subjective sense that a user would find it emotionally closer to what they wanted. It isn't found more quickly or with less computing power. The training details fell away as ineligible differences, and what remained was held to be obvious.
The process described and the route taken
This is where the decision gets interesting. The Hearing Officer described a structured process: identify the technical character, then take each feature of the claim in turn and place it in one of four categories, technical or non-technical and contributing or not. Only contributing features then count as differences under Pozzoli.
The starting point comes from the Supreme Court decision. It describes identifying the technical character of the invention, viewed as a whole, as the necessary first part of the intermediate step. The Hearing Officer built that into his approach: identify the technical character of the claim as a whole and then, "while holding this character in mind", go through each feature in turn.
The first part, identifying the technical character, is carried out by reference to what the invention achieves in the real world. The second part, however, assessing each feature in turn, doesn't appear in the reasoning, beyond a single conclusion covering all the features at once.
The assessed claim sets out several distinct features, from training the ANN on pairs of files through to storing reference files, comparing vectors and sending files over a network to a user device. In the Hearing Officer's decision, none is discussed individually against the four categories, and the feature by feature stage the decision describes isn't visibly carried out.
The feature by feature stage exists because an individual feature, or its interaction with others, can contribute to technical character even where the overall output isn't technical. Some of these features may well have merited that closer look. The training step works from measurable properties extracted from audio, video or image files. The applicant had pointed him to G 1/19's statements that measurements have technical character, and he said he would bear that guidance in mind, but it would have been helpful to see how the UKIPO applied that guidance to the training features specifically.
Parallels with Aerotel
This is where we see some parallels with Aerotel. Letting a single answer about the output settle the status of every feature, without the individual assessment, looks a lot like Aerotel's search for the contribution. The Hearing Officer identified what the invention achieves, asked whether that single result was technical, and let the answer decide the fate of every training feature at once. In a thrilling and seasonal twist, technical character took the sound before the training features could make it.
The Supreme Court said the UKIPO may adopt "any appropriate method" of identifying technical character, as the first part of the intermediate step. That flexibility relates to identifying technical character. The Hearing Officer's own test treats deciding which features contribute as a separate stage, with attention to interactions between features. Practitioners will want to know which reading the UKIPO intends examiners to follow.
The Emotional Perception refusal can be compared with the decision of 29 July 2026 in FLIR Belgium, which we assess here. In the FLIR Belgium decision, the Hearing Officer noted that the intermediate step "requires a feature-by-feature analysis". She broke the claim into labelled features and filtered out three. Several features the examiner had struck out were found to contribute, so the application went back for a full Pozzoli analysis.
The two Hearing Officers therefore set out the new framework in near identical terms in near identical terms, but applied it in very different ways.
What this means for applications now
Technical character should be set out up front, so that the examiner isn't left to frame it. The Hearing Office is of the view that Emotional Perception's representative declined to say what the claim's technical character was, so the Hearing Officer defined it in one sentence, and that sentence decided the case.
Applicants should to draw the examiner's attention to a feature-by-feature analysis. It's worth explaining why each training or implementation feature contributes in its own right, as well as why the overall output is better. The four category test set out in this decision gives a useful structure for doing that. The FLIR Belgium decision shows that this can bring features back into the inventive step analysis.
The description should be drafted with care. The Hearing Officer relied on the specification's language about "human subjective responses" and "emotive perception" to read "semantic" as subjective. How the description explains the benefit of the invention matters, both at drafting stage and when choosing what to rely on in prosecution.
