The end of 2023 featured two significant judgments concerning AI inventions: (i) a highly awaited decision from the Supreme Court in Thaler on the ability of AI systems to be named inventors of patents; and (ii) a decision from the High Court in Emotional Perception considering the application of the computer program exclusion in the UK, leading to prompt changes in patent examination practices by the UKIPO. The Thaler decision was unsurprising and consistent with decisions in other jurisdictions. Consequently, this article focuses on the second of these judgments, especially as Emotional Perception could have ramifications for life sciences companies utilising artificial neural networks (ANN); inventions using ANNs will no longer be excluded from patentability on the basis that it engages the computer program exclusion to patentability in the UK.


ANNs consist of interconnected layers of nodes that resemble the structure of neurons in the human brain. Training data is fed at the input layer and passes through multiple layers of the network, with each node applying its associated weights and biases. The data is transformed as it travels through the network, eventually leading to a result that is reflected at the output level. An ANN can exist in both hardware and software form. In Emotional Perception, the judge referred to a hardware ANN as “a physical box with electronics in it” and software ANN as an emulated ANN “in which a conventional computer runs a piece of software which enables the computer to emulate the hardware ANN”.

The High Court case was an appeal of the UKIPO’s refusal to grant a patent for a trained ANN that could identify semantic similarity or dissimilarity between media files and recommend a specific file to end users (for example, a similar music track) that is semantically similar to the input media file. While the invention was applicable to various media files, including images, text and videos, music was put forward by the parties as the most likely usage example, and was the focus of the decision.

The process of training the ANN as described in the patent application involves comparing music files in the dataset by analysing: (1) semantic properties of the file, i.e. the listener’s perceived emotional response to the music; and (2) measurable physical properties of the file, such as rhythm, tonality, timbre and/or musical structure.

The first ANN is given instructions to produce vectors in the “semantic space” via natural language processing. Semantically similar tracks produce vectors that are closer together; conversely, tracks which are semantically different create vectors that are more distant from each other. The second ANN analyses the physical properties of the same tracks, producing vectors in a “property space”, again with the differences or similarities in the tracks corresponding to the proximity of the vectors.

The second ANN is then trained to make the distances between pairs of the property coordinates converge or diverge in alignment with the distancing between them in the semantic space. For example, if two songs are very semantically similar, the property vectors will converge so that they are also closer together. The training process that corrects such as “errors”, known as back-propagation,, is repeated until the distance between the property vectors reflect semantic similarity or dissimilarity. This trains the ANN to identify semantic similarity between a pair of music tracks based on an analysis of their physical properties. This results in the ANN’s ability to take a track provided by a user, and recommend a semantically similar music track from an overall database.

The UKIPO’s decision

Holding that the invention was not patentable, the Hearing Officer was of the view that the application was a “computer program as such”, thus falling under the exclusion to patentability under s.1(2)(c) of the Patents Act 1977. In reaching this decision, the Hearing Officer applied the four stage test in Aerotel:

  1. Properly construe the claim;
  2. Identify the actual contribution;
  3. Ask whether it falls solely within the excluded subject matter; and
  4. Check whether the actual or alleged contribution is actually “technical” in nature.

The Hearing Officer also considered the five AT&T signposts below to assess technical contribution of the invention in question before concluding that the actual contribution was “no more than a computer program” and that “the ANN-based system for providing semantically similar file recommendations is not technical in nature”:

  1. Whether the claimed technical effect has a technical effect on a process which is carried on outside the computer.
  2. Whether the claimed technical effect operates at the level of the architecture of the computer, i.e., whether the effect is produced irrespective of the data being processed or the applications being run.
  3. Whether the claimed technical effect results in the computer being made to operate in a new way.
  4. Whether the program makes the computer a better computer in the sense of running more efficiently and effectively as a computer.
  5. Whether the perceived problem is overcome by the claimed invention as opposed to merely being circumvented.

He further decided that there was no suggestion of any technical effect existing over and above the running of a program on a computer.

Emotional Perception appealed the UKIPO’s decision on the basis that:

  1. The computer program exclusion was not engaged at all;  and
  2. If there was a computer program, the exclusion did not apply because the claim revealed a technical contribution and was therefore not a “computer program as such”.

UK High Court ruling

Was the invention a computer program?

The High Court disagreed with the UKIPO, holding that the trained ANN was not a computer program. Mann J disagreed with the UKIPO’s view that the trained ANN could not be “decoupled” from the underlying software platform that facilitated it. He acknowledged that, although the initial training stage of the ANN involved a computer program, the subsequent operation of the trained emulated ANN did not. His view was that there was a distinction between the underlying software which created the emulated ANN, and the emulated ANN itself. No instructions were being provided by a human for the emulated ANN to carry out its function; it applied its own weights and biases to produce relevant vectors. Accordingly, the emulated ANN operated at a different level from the underlying software on the computer.

When it came to hardware ANN, the UKIPO conceded that it did not operate as a computer program, something not disputed by Emotional Perception. Mann J decided that he was not inclined to consider the correctness of this concession on the part of the UKIPO noting that  “presumably it is because the hardware is not implementing a series of instructions pre-ordained by a human. It is operating according to something that it has learned itself”. The Judge went on to rule that, similarly, the emulated ANN was not deemed to operate as a computer program as it operated in the same way as hardware ANN (as described by him in this judgment).

Was there a technical contribution?

After establishing that the invention did not involve a claim to a program for a computer, Mann J nonetheless went on to consider whether there was a technical contribution, in case the computer program exclusion was invoked. He concluded that the invention made a substantial technical contribution, and therefore could be patentable.

The UKIPO Hearing Officer deemed that a subjective reaction from the end user after being sent a recommended song, meant a technical effect did not exist. Mann J disagreed with that analysis; the ANN’s recommendation of a semantically similar track to the end user was based on technical criteria which the system had taught itself, and therefore the system did have technical effect which could not be disqualified because of the possible subjective effect on the mind of an end user. 


Within days of the High Court’s decision, the UKIPO published statutory guidance which immediately changed examination practice, such that inventions involving an ANN should not be rejected under the “program for a computer” exclusion.  This statutory guidance was promptly followed by a related update to the UKIPO Manual of Patent Practice.

The High Court’s approach diverges from the European Patent Office’s (EPO) current practice, which applies a more limited scope in which AI inventions can be patentable. The EPO’s view is that AI and machine learning are based on computational models and algorithms, such as neural networks, and are considered to be of an abstract mathematical nature, irrespective of whether they can be “trained” based on training data. The EPO’s stance is that claims of mathematical nature are excluded from patentability, and to fall outside of the exclusion, inventors must show that the features of the claim contribute to the technical character of the invention as a whole. Only then will the claim be considered in the assessment of inventive step.

On the face of it, this is a favourable outcome to AI developers using ANNs for any applications. In the digital health sector, for example, ANNs can assist in clinical diagnosis, by analysing medical images and reports to provide more accurate predictions. ANNs can also be useful for early stage drug discovery, from initial screening to predicted success rate of particular drug candidates. As ANNs are able to process large volumes of data quickly, they can help accelerate diagnosis and drug development. However, this judgment may be of limited commercial impact. Innovators typically follow a strategic path that results in patent protection in multiple jurisdictions, and not just in the UK.  Consequently, the fact that it may now be easier to obtain patents for inventions concerning ANNs in the UK, may have limited impact on overall R&D and patent prosecutions strategies.