AI Is Just a Computer Program After All – UK Court of Appeal rules

The Court of Appeal has just ruled that Emotional Perception AI’s neural-network based music recommendation tool should be treated the same as any other computer program under patent law, overturning the High Court’s decision. The judgment was given by Birss LJ, with whom Davies LJ and Arnold LJ agreed.

In doing so they dismissed the High Court’s finding (covered by our previous article here) that unique features of artificial neural networks (“ANNs”) differentiate them sufficiently to allow them to fall outside the default ban on patents for "a program for a computer ... as such".

After finding that the tool was a computer program, they turned to whether it is nevertheless patentable by virtue of making a “technical contribution” beyond just being such a program. They found this not to be impacted by the features of ANNs, and instead the function of the program itself (in this case to recommend a music track) is all that need be considered for that test.

Emotional Perception’s patent

Emotional Perception AI’s application concerns an ANN-based tool for associating (and thereby recommending) files based on semantic similarity. For example, it would associate musical tracks together based on both their emotional and musical similarity, with songs of the same genre and mood being considered closely related. This is then to be used to select a new track that is sufficiently similar to any given input track.

This ANN based process functions through having a set of weights developed through training. These weights are adapted autonomously through a backpropagation process until their analysis of training data provides the same outcome as the human-determined “right” outcome for that data, and are then immutable in further operation of the ANN.

The application claimed a method of making this association and recommendation, and a system which does this. However, nothing in the decision turned on the distinction between these claims to the ANN. 

(For further details, and a summary on how ANNs work, see our previous article here)

ANNs as Computer Programs

The first part of the appeal covered whether Emotional Perception AI’s claimed ANN was a computer program. 

Birss LJ noted that a computer program could be defined simply as “a set of instructions for a computer to do something”, with a computer simply defined as “a machine which processes information”. The question was whether Emotional Perception AI’s process fell under this definition.

Emotional Perception AI argued that its claimed ANN was not a computer program, based on the unique features of ANNs compared to standard computer programs. However, applying the definition of a computer program, Birss LJ rejected this analysis for the following reasons:

• Firstly, the fact that the final form of the process was not defined by a human programmer (e.g. through the weights being developed through training) was deemed to not be a distinction, noting that the final form of computer programs in general is normally computer-generated due to the compilation process applied to the human-written code.
  
• Secondly, the fact that ANNs solve problems which are hard to be fully formulated by a human programmer was deemed just an extension of the nature of current programs, which routinely solve problems that humans would find hard to solve unaided.
  
• Finally, the permanence of the outcome of the learning process (with the weights etc being “embedded” in the product) was deemed inconsequential with regards to whether something is a program. This was by comparison to previous case law, which found that permanent code stored in Read-Only Memory (ROM) and alterable code in other media were equally able to be computer programs.

For these reasons, Birss LJ found that the proposed ANN-based process was still a set of instructions for a computer to do something, and therefore a computer program. 

The Technical Contribution of an ANN

As Emotional Perception AI’s claims were found to be for a computer program, the Court of Appeal then turned to whether they were patentable (despite being a program) by virtue of offering a technical contribution.

In considering this, the Court found the features of how the claimed ANN was created, for example the process of training the ANN, was simply the means of creating the program and not part of the technical contribution. Furthermore, the output, that a particular file is recommended, is simply “the presentation of information” and similarly unpatentable without more.

On this basis, Birss LJ found the patent’s key contribution is the provision of improved file recommendations, and therefore the patentability turned on whether that alone was a sufficient technical contribution.

On reaching this conclusion, Birss LJ noted that the improvement of the recommendation was based on a semantic, rather than technical, comparison between the files, and therefore based only on superficial or aesthetic matters. Such a comparison is not patentable, and accordingly the patent application is to be rejected.

Comment

Prior to the High Court’s decision, the primary approach to patenting AI inventions in the UK was to navigate the “technical contribution” test, e.g. through seeking protection for the use of AI for a specific technical purpose or for a technical means of implementing AI. If it had stood, the High Court’s judgment would have fundamentally changed that, broadening the scope of when AI can be patented by allowing the AI implementation itself to be patented. This change was implemented promptly by the IPO, through revised patent guidelines which were issued on 29 November 2023.

This present judgment, therefore, will reset that change. Pending news on whether Emotional Perception AI will appeal to the Supreme Court (and, if so, the outcome of such an appeal), we expect that in the meantime the IPO will remove the changes made to their patent guidelines, removing the special treatment for patent applications involving ANNs.

Consequently,  ANN-based inventions will return to the familiar, if intricate, state of only being patentable if they make a “technical contribution”. This is harder to show for computer programs, but the great advances ANNs enable (both in terms of development of AI systems and the contributions AI can provide elsewhere) means this Court of Appeal’s judgment by no means rules out such developments from patentability.

Written by Toby Bond and Fred Cascarini