The UK Supreme Court’s decision in Emotional Perception AI Ltd v Comptroller-General of Patents, Designs and Trade Marks was widely regarded as a watershed moment for the patentability of artificial intelligence inventions in the UK. Yet, less than a year later, the UKIPO has refused the application on remittal, concluding that the claimed artificial neural network (ANN) recommendation system lacks an inventive technical contribution.
Although the deadline for appeal has not yet passed, an appeal does not presently appear likely. Attention can therefore now turn to the broader questions: what does the UKIPO decision tell us about technical character, inventive step and the patentability of AI inventions in the UK?
From Exclusion to Inventive Step
As reported previously when ruling on the Emotional Perception appeal, the Supreme Court fundamentally reshaped the legal framework for computer-implemented inventions.
Whilst confirming that artificial neural networks can fall within the concept of a computer program, the Court moved away from the traditional Aerotel analysis and embraced principles drawn from G 1/19. Most significantly, the Court introduced an intermediate step requiring identification of those claim features that contribute to the technical character of the invention before inventive step is assessed.
The Court deliberately avoided deciding whether the claimed recommendation system possessed technical character. Instead, that issue was remitted to the UKIPO.
The UKIPO’s Response
On remittal, the UKIPO accepted that the application was not excluded from patentability merely because it involved an ANN.
The decisive issue became whether the claimed ANN architecture, training objectives and techniques for closing the “semantic gap” contributed to the technical character of the invention.
The examiner and subsequently the Hearing Officer concluded that they did not.
In broad terms, the invention was characterised as seeking to improve the semantic or emotional relevance of recommendations provided to users. Once that objective was regarded as non-technical, the specific ANN learning objectives and convergence mechanisms were treated as differences that could not support inventive step.
With those features discounted, only conventional computer hardware and implementation remained. The result was a finding of obviousness.
The Debate Over G 1/19
The applicant relied heavily on passages in G1/19 recognising that:
- technical effects may arise within a computer system;
- measurements may constitute technical activities;
- technical contributions do not necessarily require a direct interaction with physical reality; and
- sophisticated information-processing techniques can, in appropriate circumstances, contribute to technical character.
The Hearing Officer accepted that key principles from G 1/19 had been endorsed by the Supreme Court and that the Enlarged Board of Appeal’s opinions represented settled law which the Supreme Court had indicated the UKIPO should follow. However, he rejected the proposition that the entirety of the Enlarged Board’s reasoning had become binding UK law.
As a result, much of the wider reasoning of G 1/19 was treated as highly persuasive rather than mandatory.
The Real Battleground: Technical Effect
The Applicant’s case focused heavily on the technical implementation of the ANN, including its learning objectives, vector-space representations and semantic-gap-closing mechanisms. Whilst those features were said to contribute to technical character, the case ultimately centred on whether they could be linked to a clearly identifiable technical problem and technical effect.
The Hearing Officer ultimately characterised the remaining technical contribution at a high level as “Computer hardware running a program (the ANN) that selects and outputs a file”.
The Applicant argued that the processing undertaken by the ANN was such that the ANN outputted a “semantically better file”. However, the Hearing Officer concluded that how a file was chosen was a non-technical difference and therefore did not contribute to the technical character of the invention.
The Hearing Officer’s decision illustrates that the debate over technical character is, in many cases, really a debate about technical effect. Once the Hearing Officer concluded that the claimed ANN produced only a “semantically better” recommendation and that this result related to human perception, which is inherently subjective, rather than technology, the remainder of the inventive-step analysis largely followed. This reflects long-established EPO practice under which features that do not contribute to a technical effect cannot support inventive step.
Such an approach is entirely consistent with EPO practice.
In G 1/19, the EPO’s Enlarged Board of Appeal stated that “patent protection is reserved for creations in a technical field” and that claimed subject matter must have a “technical character” or, more precisely, involve a “technical teaching“, namely an instruction addressed to a skilled person as to how to solve a particular technical problem using particular technical means. The Enlarged Board further noted that the Basic Proposal for the Revision of the EPC expressly stated that these considerations regarding technical character apply equally to the assessment of computer programs.
Under established EPO jurisprudence, features that do not contribute to the technical character of an invention cannot support inventive step. Similarly, EPO case law makes clear that where no relevant technical effect can be identified, inventive step will generally fail regardless of the sophistication of the underlying implementation.
The decision also resonates with established principles from English patent law.
In Actavis UK Ltd v Novartis AG, Jacob LJ observed that arbitrary claim limitations which possess no technical significance cannot support inventiveness. Similarly, in Sandvik Intellectual Property AB v Kennametal UK Ltd, Arnold J emphasised that a selected feature is not inventive merely because it differs from the prior art. Its alleged technical significance must be established.
The real disagreement in the case was therefore not whether non-technical features should be disregarded. Both UK and EPO practice broadly accept that proposition. Rather, the key question was whether the ANN learning objectives, semantic-gap-closing mechanisms and resulting file selection in the Emotional Perception System should themselves have been regarded as producing a technical effect.
Is This the End of the Road?
Comments published following the decision suggest that some supporters of the application believe there are serious grounds for appeal. Criticism has focused on the operation of the new “intermediate step” and allegations that the UKIPO has, in substance, drifted back towards a contribution-based analysis reminiscent of Aerotel.
Yet there are also practical realities.
The dispute has already been litigated before the UKIPO, the High Court, the Court of Appeal and the Supreme Court before returning to the UKIPO on remittal. Any further appeal would involve substantial cost, delay and uncertainty.
At present, there appears to be no indication that a further appeal will be pursued. If that remains the case, the October 2026 decision may stand as the first significant UKIPO attempt to apply the Supreme Court’s new framework in practice.
What Next for AI Patenting in the UK?
If the Emotional Perception saga is indeed approaching its conclusion, its legacy may be less about the patentability of ANNs and more about the future meaning of technical character.
The Supreme Court removed many of the barriers associated with excluded subject matter. The UKIPO’s remittal decision demonstrates that the focus has now shifted elsewhere.
For applicants seeking protection for AI inventions, the key question is unlikely to be whether machine learning techniques are sophisticated or novel. Instead, the crucial question will be whether those techniques can be linked to a clearly identifiable technical problem and a technical effect arising from their implementation.
In that sense, Emotional Perception may mark not the beginning of a more permissive era for AI patents, but rather the beginning of a new and more nuanced debate about the relationship between technical character and inventive step.
If no appeal is filed, the Emotional Perception litigation will have achieved the unusual feat of reshaping UK law whilst leaving many of the most important practical questions unanswered. The Supreme Court settled the law on patent eligibility. The UKIPO’s remittal decision shows how the debate has now shifted to technical character and inventive step. Whether future applicants can successfully demonstrate technical effects in AI systems will be key.
