The Court of Appeal's Ruling on Emotional Perception’s ANN Application

Emotional Perception AI Ltd’s (EPL’s) artificial neural networks (ANNs) patent application has sparked considerable debate regarding patentability of computer programs in the UK, asking the question of whether a patent application for an ANN-based music recommendation system is excluded as a computer program ‘as such’. The invention in this case did was deemed not to meet the criteria for a technical contribution necessary for patentability.

For patent practitioners, this case maintains the importance of demonstrating how an innovation makes a technical contribution beyond the mere use of computer programs. We’ll be eagerly waiting to see if EPL appeal the decision to the Supreme Court.

The Dispute

EPL developed a novel system leveraging ANNs to recommend media files based on human emotional perception. Specifically, the system utilises a trained ANN to analyse and recommend music based on both semantic (e.g. emotional tone) and physical properties (e.g. tone, timbre). Initially, the Comptroller of Patents rejected EPL's application, arguing that it constituted a computer program "as such" and was therefore unpatentable. However, this decision was overturned in November 2023 by the High Court, which ruled that the patent did not fall under the exclusion. This resulted in the Comptroller of Patents appealing the High Court’s decision to the Court of Appeal, which issued its judgment on 19 July 2024.

Understanding ANNs and the Patent Application

ANNs, inspired by biological neural networks, are sophisticated systems composed of interconnected artificial neurons that process data through various layers. These networks can be implemented in software or hardware, with their performance improved through a training process that adjusts their parameters to optimise results.

EPL’s patent application describes a method that involves two main analytical steps:

1. Semantic Analysis: Music files are processed using natural language processing (NLP) to generate semantic vectors that describe their emotional content. These vectors are placed in a semantic space where their distances reflect the emotional similarity between tracks.
2. Physical Property Analysis: Simultaneously, the ANN evaluates the physical properties of music files, creating vectors in a property space. Initially, these vectors do not align with the semantic vectors.

 The key innovation is training the ANN so that the distances in the property space align with those in the semantic space. This alignment is achieved by adjusting the ANN's parameters through backpropagation to minimise discrepancies between semantic and physical distances, enabling more accurate music recommendations.

Legal Framework and Proceedings

Section 1(2) of the Patents Act 1977, which aligns with Article 52 of the European Patent Convention (EPC), excludes patents for discoveries, scientific theories, mathematical methods, literary or artistic works, and computer programs "as such." The relevant cases for interpreting these exclusions include:

•  Aerotel Ltd v Telco Holdings Ltd [2007] RPC 7: Established a four-stage test for patentability.
• AT&T Knowledge Ventures v Comptroller [2009] EWHC 343 (Pat): Provided criteria for determining whether a computer program makes a technical contribution.

In the EPL case, the Hearing Officer applied these criteria, concluding that the ANN's training process was too closely related to programming to be considered technical. The High Court, however, found that the ANN’s output—music recommendations based on semantic similarity—constituted a technical effect, thus overcoming the exclusion (to find out more about the High Court Ruling read the article here).

The Court of Appeal

The Comptroller challenged the High Court's decision on several grounds, arguing that:

1. The ANN constitutes a computer program and should fall under the exclusion.
2. The Judge incorrectly considered the hardware implementation of the ANN.
3. The Judge overlooked the mathematical method exclusion.
4. The Judge misinterpreted the technical contribution.

Assessment of Ground 1 and Ground 2

The term “program for a computer” under Section 1(2) aligns with Art 52 EPC and focuses on whether an invention involves a set of instructions for a computer. Historical context and definitions suggest that ANNs, regardless of their implementation in hardware or software, fall under this exclusion. The Comptroller argued that ANNs, including their weights and biases, should be classified as computer programs based on traditional definitions. Conversely, EPL contended that ANNs represent an evolving system that does not fit traditional programming definitions due to their self-structuring nature.

The Court of Appeal found that regardless of how ANNs are implemented or their internal workings, their weights are still a set of instructions for a computer and thus fall within the exclusion from patentability for computer programs.

Assessment of Ground 4

It is important to note that while Section 1(2) of the Act applies to cases involving ANN implementations just as it does to any computer-implemented invention, this does not automatically render such inventions unpatentable. Various technologies, even those involving complex computer systems like ANNs, have been patentable when they involve technical contributions. Analysis of the technical contribution included looking at training and music recommendation output. The training an ANN is part of creating the program and was considered not central to the patentability assessment. The value of the music recommendation output was considered to be subjective and did not represent a technical contribution.

The appeal on Ground 4 was upheld. The Hearing Officer's decision was determined to be correct, with the music recommendation output being of a non-technical nature, thereby excluding it from patentability.

Ground 3: Mathematical Method Exclusion

The court did not need to address the mathematical method exclusion as Ground 4’s resolution was sufficient. However, if the ANN had not been considered a computer program, this exclusion might have been relevant.

Conclusion

The Court of Appeal examined whether ANNs, implemented either in hardware or software, fall within the exclusion for computer programs "as such." Despite the technical nature of the ANN's operation, the music recommendation's subjective nature (emotional similarity) did not meet the criteria for a technical contribution necessary for patentability.

The case underscores the nuanced interpretation of patentability criteria for AI-based inventions. While the High Court's ruling acknowledged the technical nature of ANN operations, it highlights the importance of distinguishing between technical processes and the subjective outcomes they produce. The Court of Appeals decision indicates that the technical nature of the ANN’s output can render it patentable, provided it meets the criteria for technical contribution and is not merely a computer program "as such."

If you’re interested in obtaining patent protection for your software or AI innovations, reach out to ip21 today to discuss your needs.

 Written by Lauren Mills for ip21 Ltd.