Is patent protection for inventions created using AI an emotional topic?

Section 1(2)(c) of the Patents Act 1977 excludes a program for a computer from protection. These are only prevented from being treated as an invention for the purposes of the Act to the extent that a patent or application for a patent relates to that thing “as such”. A recent High Court case raises new questions of law as to whether the use of an aspect of Artificial Intelligence, an Artificial Neural Network engages the exclusion or not. 

The case

Emotional Perception AI Ltd v Comptroller- General of Patents Designs and Trade Marks [2023] EWHC 2948 (Ch) involves an appeal against a decision made by a Hearing Officer of the UK Intellectual Property Office (IPO) to refuse patent protection for an artificial neural network (ANN). 

The patent application was for an improved system for providing media file recommendations to an end user. Typically for use in music websites to suggest tracks the user might like based on ones they already knew or had. Existing websites could suggest music using human tagging and categorisation (rock, folk, happy, sad etc) but this “semantic categorisation” was limited to types of music. The stated advantage of the patent application was that it could suggest similar music in terms of human perception and emotion, irrespective of the genre. It does this by passing music through a trained ANN which can assimilate the semantic characterisation of a track and allocate coordinates to it. Two tracks can then be compared in “semantic space”. A separate ANN can characterise tracks by “physical” properties such as tone, timbre, speed etc and produce coordinates in “property space”. The ANN is then trained until it can take any music track, calculate its property space coordinates and find semantically similar music to make recommendations to end users by sending a file and message to them.

An ANN can either be a hardware device or emulated by software on a computer. The hardware ANN consists of layers of neurons akin to those in the brain, arranged in layers and connected to each other (or at least some others), and to layers below. Each neuron can process inputs and produce outputs according to the weights and biases of that neuron. The outputs are passed on to other neurons except the last layer which produces outputs from the system. In a software ANN a computer runs a piece of software to enable the computer to emulate the hardware version.

Even though this case involved the computer program exclusion, the questions: 

  • where is the computer? and 
  • where is the program? 

hadn’t been addressed before the Hearing Officer. The judge thought these required more focus as a self trained ANN was involved. 

Where was the computer for the purposes of the exclusion?

The hardware ANN and the software emulation were considered separately.

  • The hardware ANN processes data so (with reference to dictionary definitions of a computer), would be regarded as a computer.
  • The software emulation of the ANN has to run on a computer, so a computer “is involved” for the purposes of the exclusion.

Where was the program?

In written submissions the IPO conceded that, in the case of a hardware ANN, there’s no relevant computer program to which the s1(2)(c) exclusion applies. The judge therefore couldn’t consider the correctness of this point but commented that there would have been an interesting debate had the concession not been made, hinting that future judicial consideration of this aspect would be welcomed.

The software ANN involved two programs: one at the training stage and one in the trained ANN.  The “knottier problem” was whether the internal training and subsequent operation of the trained emulated ANN is a computer program for the purpose of the exclusion. This was the crux of the decision. 

The judge held that the emulated ANN existed at a different level to the underlying software. There was no program at that point because the ANN was not operating a set of program instructions given by a person, but rather applying its own weights, biases etc to produce relevant coordinates. The ANN had trained itself. It was emulating a piece of hardware which had physical nodes and layers, so the software ANN was no more operating a program than the hardware ANN was. Therefore the emulated ANN was not running  a program for a computer for the purposes of s1(2)(c).

Was the invention a claim to a computer program?

Although there was a computer program at the training stage of the software emulated ANN, it wasn’t correct to view the whole thing as some sort of programming activity. What is said to be special in the applied for patent is the idea of using pairs of files for training and setting the training objective and parameters accordingly. The actual program is a subsidiary part of the invention and is not what’s claimed. The claims go beyond that. So as a matter of construction the claim is not to a computer program at all, and the exclusion can’t apply. 

Technical contribution

The judge also considered the question of technical contribution. He commented that, if he was wrong about the claim not being a claim to a computer program, or if there was a computer program elsewhere, it would be necessary to consider this issue. The mere involvement of a computer program in an invention doesn’t invoke the statutory exclusion. If there’s a technical effect or contribution which lies outside the excluded subject matter, the invention may be patentable. 

The judge decided that the claimed invention made a technical contribution. The sending of an improved recommendation message in the form of a file sent to an end user was a matter external to the computer and the ANN had gone about its analysis and selection of the file in a technical way. It wasn’t just any old file; it was a file identified as being semantically similar by the application of technical criteria which the ANN has worked out for itself. The output was of a file that wouldn’t otherwise be selected. That was a technical effect outside the computer, and when coupled with the purpose and method of selection, fulfilled the requirement for technical effect. There ought to be no difference between a hardware ANN and a software emulated ANN for this purpose. 

What does this mean for future patent applications?

The immediate effect of the judgement is that the UK IPO has announced (in statutory guidance) that patent examiners shouldn’t object to inventions involving artificial neural networks under s1(2)(c) of the Patents Act 1977. So we might see an increase in patent applications for this type of AI in the coming months. Although each application will need to satisfy the usual requirements of novelty, inventive step etc.

The IPO guidance doesn’t address s1(2)(a) of the Act, so objections to patent applications for ANNs could be raised on the basis that an invention enabled by the use of an ANN (or the algorithm by which it’s trained) relates wholly to a mathematical method. Arguably an ANN is an abstract model which takes a data input, applies a series of mathematical operations (applying weights, biases etc) and outputs data at successive layers. This could be said to be a mathematical method. Arguments on this aspect of the exclusion weren’t fully considered in Emotional Perception AI for procedural reasons but could be tested in future patent applications, or objections to them. 

The UK IPO has also suspended its broader AI patentability guidelines pending consideration of the judgement. It will be interesting to see whether the guidelines are deemed permissive enough in light of this case, however, as always, each patent hinges on its individual claims. 

Before drafting any claims, parties should carefully consider what exactly is the computer and the program involved. This will be fundamental to any analysis made by the IPO or, if it comes to it, a court. So any applicant will need to carefully consider these two questions, then be prepared to justify their reasoning. 

Ultimately for inventions created by self trained AI which truly exist above a software platform, the door may now be slightly more open for patent protection.  

Our content explained

Every piece of content we create is correct on the date it’s published but please don’t rely on it as legal advice. If you’d like to speak to us about your own legal requirements, please contact one of our expert lawyers.

Posted by


Mills & Reeve Sites navigation
A tabbed collection of Mills & Reeve sites.
My Mills & Reeve navigation
Subscribe to, or manage your My Mills & Reeve account.
My M&R


Register for My M&R to stay up-to-date with legal news and events, create brochures and bookmark pages.

Existing clients

Log in to your client extranet for free matter information, know-how and documents.


Mills & Reeve system for employees.