The Court of Appeal recently ruled that Emotional Perception AI’s neural-network-based music recommendation tool should be treated as any other computer program under patent law, overturning a previous High Court decision. The judgment, delivered by Birss LJ with the agreement of Davies LJ and Arnold LJ, dismissed the High Court’s assertion that the unique features of artificial neural networks (ANNs) are sufficient to exempt them from the default patent ban on “a program for a computer … as such.”
The Court of Appeal delved into whether the tool, despite being a computer program, could still be patentable by making a “technical contribution” beyond merely being a program. They determined that the distinctive attributes of ANNs do not impact this consideration. Instead, the core function of the program—in this case, recommending a music track—remains the primary criterion for this evaluation.
Emotional Perception AI’s patent application involves an ANN-based tool designed to recommend files based on semantic similarity, associating musical tracks by their emotional and musical likeness. Tracks of the same genre and mood are considered closely related, facilitating the selection of a new track similar to a given input track. This ANN process operates through a set of weights developed autonomously via a backpropagation process, refining its analysis of training data to match human-determined outcomes. These weights become immutable during the ANN’s further operation.
The application claimed both a method of making this association and recommendation, and a system implementing it. However, the Court’s decision did not hinge on the distinction between these claims.
In the first part of the appeal, the Court examined whether Emotional Perception AI’s claimed ANN qualified as a computer program. Birss LJ defined a computer program as “a set of instructions for a computer to do something,” with a computer being “a machine which processes information.” The Court had to decide whether Emotional Perception AI’s process fell under this definition.
Emotional Perception AI contended that its ANN was not a computer program, citing the unique features of ANNs compared to standard programs. However, Birss LJ rejected this argument for several reasons. Firstly, the final form of the process, defined through training rather than by a human programmer, was considered not a distinction, given that the final form of computer programs is typically computer-generated through the compilation of human-written code. Secondly, the fact that ANNs solve problems difficult for human programmers to fully formulate was seen as an extension of the nature of current programs, which routinely address complex problems. Finally, the permanence of the learning process outcome (with weights embedded in the product) was deemed irrelevant, likened to previous case law which found both permanent and alterable code equally capable of being computer programs.
The Court thus concluded that the ANN-based process constituted a set of instructions for a computer to perform a task, and therefore, it was a computer program. The next question was whether these claims could still be patentable by providing a technical contribution. The Court concluded that the features of how the ANN was created, like the training process, were merely the means of creating the program and not part of the technical contribution. Furthermore, the output—recommending a particular file—was simply the presentation of information and, without more, unpatentable.
Birss LJ concluded that the patent’s key contribution was the provision of improved file recommendations. The patentability hinged on whether this alone constituted a sufficient technical contribution. The improvement was based on a semantic comparison between files, deemed superficial or aesthetic, and thus not patentable. Consequently, the patent application was rejected.
Prior to this judgment, the primary approach to patenting AI inventions in the UK involved navigating the “technical contribution” test, seeking protection for AI used for specific technical purposes or for technical means of implementing AI. The High Court’s prior judgment would have broadened the scope of patentable AI by allowing the AI implementation itself to be patented. This change was promptly adopted by the Intellectual Property Office (IPO) through revised patent guidelines issued on November 29, 2023.
The present judgment reverts this change. While Emotional Perception AI may still appeal to the Supreme Court, the IPO is expected to rescind the changes to their patent guidelines, removing special treatment for ANN-related patent applications. As a result, ANN-based inventions will revert to the established requirement of demonstrating a “technical contribution” for patentability. Although this is more challenging for computer programs, the significant advances ANNs bring to AI development and their broader contributions ensure that such innovations remain within the realm of potential patentability.
