Osha Attorneys
Osha Attorneys
Osha Attorneys
Osha Attorneys
Attorney search
Search by

The collective expertise of our global team distinguishes OBWB in the field of Intellectual Property Law. We align our best resources to meet each client's specific needs and we treat each matter with the highest degree of attention and care.

Emotional Perception AI v UKIPO: Is There a Change in The UK Intellectual Property Office's Assessment of AI?

日本語 简体中文 繁體中文 عربي

The UK Patents Act excludes “a program, for a computer … as such” from being eligible for patent protection. The UKIPO has used this exclusion to reject patent applications claiming inventions involving artificial neural networks (ANNs). Previous examination guidance issued by the UKIPO stated that ANNs, unless implemented in fixed hardware, were considered to be programs for a computer. This meant that ANN inventions could only be patented if the invention could demonstrate a technical effect inside or outside of the computer.

However, on November 21, 2023, the High Court of England and Wales issued a ruling in Emotional Perception AI Ltd v Comptroller General of Patents, Designs and Trademarks that trained ANNs cannot be considered computer programs "as such" for the purposes of the exclusion from patentability.


Briefly, the case concerns a patent application by Emotional Perception AI for a system that recommends data files, such as music files, to users. The ANN is trained with pairs of media files and corresponding text descriptions of each file’s content, such as “happy”, “sad”, or “relaxing”. The text descriptions are analysed using a first ANN via natural language processing software, so that a semantic embedding is found for each of the descriptions in a semantic space. The similarity or difference between the descriptions of the files is reflected by the distance between the embeddings in the semantic space. For example, embeddings for descriptions of music files that are similar will have coordinates closer together in the semantic space than descriptions that are further apart in similarity.

A second ANN is used to analyse parameters, such as tone, timbre, speed, and volume, of the same two media files and output a parameter embedding for each file in a property embedding space. The similarity or difference between the files (in term of their parameters) is reflected by the distance between the embeddings in the property space. This is the ANN which is the final trained ANN.

The second ANN is then trained, using back-propagation, to make the distances between pairs of the parameter embeddings converge or diverge in accordance with the distance between the semantic embeddings in the semantic space. The trained second ANN can then use parameters of a file without semantic labels to recommend to a user other files that are nonetheless semantically similar to the first file.

The UKIPO rejected the application in June 2022 on the basis that it constituted a “program for a computer” and thus was excluded from patentability. It was also rejected on the basis that recommending content was “of a subjective and cognitive nature” rather than a technical effect that might render the invention patentable.


On appeal, the judge considered that the process by which the artificial neural network is trained is a program for a computer as such. However, he considered that, even when implemented in software, the trained artificial neural network was not implementing code given to it by a human but was instead operating according to something that the ANN had learned. Therefore, it was not a computer program and did not fall under the exclusion.

Additionally, the judge stated that the invention provided a technical effect beyond what might be considered a computer program.  The judge said that the content being recommended “is not just any old file; it is a file identified as being semantically similar by the application of technical criteria which the system has worked out for itself”. 


Following the decision, the UKIPO announced an immediate change to the examination of patent applications involving ANNs by issuing guidance that patent examiners should no longer object to inventions involving an ANN under the “program for a computer” exclusion in the Patents Act.

This appears to be a significant shift in the UKIPO’s assessment of patentability of claims directed to AI implementations. It also indicates a divergence between the UKIPO and EPO in the extent to which ANNs are considered excluded from patentability.

However, it remains possible that inventions involving ANNs could be excluded as mathematical methods under UK patent law. This was raised as an issue in the Emotional Perception case but was not considered by the judge due to a procedural objection.

Furthermore, it has been reported that the UKIPO has been granted leave to appeal the decision - so further change may come. We will report on further developments as they occur.