Mishcon de Reya page structure
Site header
Main menu
Main content section
Abstract white architecture

Patentability of Artificial Neural Networks – revolution or evolution?

Posted on 24 January 2024

In Emotional Perception v Comptroller-General, the High Court was asked to consider, on appeal from the UK Intellectual Property Office, a patent application for an invention involving a form of artificial intelligence known as an Artificial Neural Network (ANN). The eligibility of ANNs for patent protection had not been considered by the UK courts before this decision.

The High Court found that the trained ANN claimed in the subject application was not excluded from patent protection under s1(2)(c) Patents Act 1977 either because no relevant computer program was claimed or because, if it was, then the technical effect claimed meant it was not excluded as a computer program "as such" under the Act.

The decision opens the way for the patenting of trained ANNs in the United Kingdom, but also serves as a reminder of the care that must be taken when drafting claims involving computers and computer programs. The UKIPO has acted quickly to update its guidance in relation to ANNs, but has also sought permission to appeal the decision to the Court of Appeal.

The ANN and the Application

An ANN uses neurons connected in layers to process inputs and produce outputs. The ANN in this application concerned a system of making media recommendations. It did so based on comparisons of the physical properties of pairs of training data files (in the case of a music file this might be its tone or tempo etc.) with their human-assigned semantic properties (which might be descriptions such as "happy", "sad" or "relaxing").  

The ANN in the invention plots the notional co-ordinates of the physical properties of pairs of files in one "property space" and does likewise, in a separate "semantic space", for the files' semantic properties. It then compares the distances between the files' co-ordinates in the two spaces and, using an algorithm provided by a human, adapts the co-ordinates in the second space as required to make the distance between the files in that space align with the distance between them in the semantic space. In doing so, the ANN trains itself iteratively, by repetitive correction and "without being told how to do it by a human" to adjust the weights and biases in its nodes. It continues this training process until it is able to sufficiently accurately match the physical properties of input files to its semantic descriptions. Once it has been suitably trained, the ANN's structure is "frozen" and is capable of being used to recommend new files based on their similarity to known files.  

It appears to have been common ground that, once the training of the ANN was complete, "no activity which might be called programming activity takes place in relation to the ANN or the data".

The UK IPO decision

The UK Intellectual Property Office refused to grant the patent for the reason that what was claimed was a computer program as such, which is excluded from patent protection under s1(2)(c) Patents Act 1977, and made no technical contribution to save it from exclusion.

That decision was challenged on multiple bases, including that there was no computer program engaged in the ANN or, if there was, then the claims claimed a "technical contribution" and were not claims for a computer program as such.

The legal test for assessing whether a claimed computer program invention makes a contribution which is not excluded from patent protection entails the following four step test from the decision in Aerotel v Telco: (1) construe the claim; (2) identify the contribution made by the invention; (3) consider whether the invention falls within excluded subject matter (i.e. it is no more than a computer program); and (4) check whether the contribution is actually technical.

The High Court decision

Whether a computer program was claimed

Applying the relevant legal principles to the ANN in this case, the Judge concluded that the ANN could be considered a computer (whether it was implemented in hardware or in a software emulation), and that the training of the ANN involved a computer program. However, the application did not claim that program. Furthermore, once the training was complete and the ANN was operating to provide its recommendations to users, there was no program running on that computer because that operation did not depend on a set of program instructions given by a human, but rather on how the ANN had trained itself. 

That conclusion meant the "computer program as such" exclusion was not engaged at all.

Whether any computer program had a relevant "technical effect"

The Judge then went on to consider whether the application was excluded if he was wrong about whether or not a computer program had been claimed. On the facts of this case, he considered that the effect of the ANN, namely a recommendation to a user and the movement outside of the computer system of the recommended file, was a technical effect with a "real world" output.  Accordingly, the 'computer program as such' exclusion did not apply.

Similarly, the Judge held that a trained hardware (and emulated) ANN could by itself be an external technical effect and be unaffected by the exclusion.

Mathematical methods

The Judge also briefly considered the "mathematical methods" exclusion from patent protection under s1(2)(a) of the Act.  The point was not "in play" before the High Court for procedural reasons, but the Judge appeared to agree with the IPO's finding that the ANN was not so excluded as a mathematical method because of its specific application in recommending media files.

Consequences for the application and IPO practice

The application will be remitted back to the UK IPO for further consideration on substantive eligibility grounds now that it has been determined that the claimed invention is not excluded (as a mathematical method, as such or a computer program, as such).

In the meantime, however, the IPO has both suspended its existing guidance on the examining of patent applications relating to Artificial Intelligence inventions and has issued a practice note instructing examiners not to object to inventions involving an ANN under the 'computer program' exclusion under section 1(2)(c) of the Patents Act 1977. 

Whilst this would seem to suggest the UK IPO's acceptance of the decision, in parallel, it has sought and been granted permission to appeal the decision. So, there may be more chapters to this story yet.


As always, caution must be exercised when applying ground-breaking decisions like this to new technologies (particularly as the decision may not yet prove final), and the High Court's judgment makes clear that the decision in Emotional Perception turns in large part on the underlying 'technical effect' provided by the invention and equally on the drafting of the claims. With that in mind, the exclusion for "computer programs as such" remains and must continue to be considered carefully before patent applications for 'AI' inventions are filed. It is equally important to ensure skilled and proper drafting of such applications is carried out, by engaging with patent attorneys proficient in this technology area.

Nonetheless, the Emotional Perception decision (to the extent it remains final) provides an important and warm-welcomed shift in approach in the UK for patent applicants seeking protection in this technology space. As a review of the prior decisions of the UK IPO prior to Emotional Perception will demonstrate, it had been generally very difficult to convince the UK IPO to grant patents for inventions in this technology space. It is hoped therefore that the decision will result in the UK IPO being more willing to grant patents for ANN inventions (and AI inventions more generally) in the near future.

How can we help you?

How can we help you?

Subscribe: I'd like to keep in touch

If your enquiry is urgent please call +44 20 3321 7000

Crisis Hotline

I'm a client

I'm looking for advice

Something else