On 21 November 2023, the High Court of Justice of England and Wales, decided that a claim directed to a trained ANN (Artificial Neural Network) was not merely a “program for a computer.” In fact, here, the judge concluded that the claimed subject-matter is not directed to a computer program at all. Therefore, the computer-program exclusion is not invoked. Of course, this is an important step forward for AI patenting in the UK, and perhaps also in other countries.
Appellant: Emotional Perception AI Ltd.
Respondent: Comptroller-General of Patents, Designs and Trade Marks
Judge: Sir Anthony Mann
The patent application in suit concerns a system for providing improved media file recommendations to an end user, for example in the form music tracks from a website. The advantage of the proposed system is that it offers suggestions of similar music in terms of human perception and emotion irrespective of the genre of music and the apparently similar tastes of other humans. The system arrives at its suggestions by passing music through a trained ANN.
Like the patent laws in most countries, the UK Patents Act 1977 excludes from patent protection “a program for a computer … as such.” The appeal in question concerns the exclusion only and does not for example deal with sufficiency of disclosure, or any other questions that may determine whether a patent can be granted based on the application filed by the applicant.
The claimed invention is applicable to various types of media, including music, video, still images and text. The claims define a system and a method respectively, which rely on a trained ANN. In short, the invention is based on classification of each file, e.g., a music track, in two different domains, namely a semantic space and a property space. A pair of music tracks is taken, which each is accompanied by a natural language description of how that music is perceived by a human. At its simplest the music may be described as happy, sad, or relaxing. These descriptions are in word form i.e., “semantic” and its derivatives which are used to describe this sort of feature of the music and are to be analysed by an ANN via natural language processing software.
A first ANN is given instructions that enable it to assimilate the characterisation of each music track and produce a respective vector in a semantic space based on the type of music for each of the tracks in the pair. The similarity or difference between the semantic types of music is reflected by the distance between those two vectors in the semantic space. Two tracks of music which are semantically similar will have their vectors closer to¬gether; and the farther apart they are in similarity, the farther apart their vectors will be.
The same two tracks are also analyzed in second ANN for what are described as its physical properties, e.g., tone, timbre, speed, loudness and a lot of other characteristics set by the human. That analysis produces vectors in a “property space.” Here, the differences or similarities between the music tracks is also reflected by the pro-ximity of respective vectors. The second ANN is the one that will be the final operative ANN in the system.
The second ANN is trained to make the distances between pairs of the property coordinates (i.e., the vectors) converge or diverge in alignment with the distancing between them in the semantic space. If the property space co-ordinates are farther apart than those in the semantic space, they are moved closer together in the property space, and conversely if the distancing is too close together in the property space to reflect semantic dissimilarity. This training is carried out by so-called backpropagation, where an “error” in the property space is corrected to make the results coincide with the training objectives. The back-propagation algorithm is provided by a human and the correction is achieved by the ANN’s adjusting its own internal wor-kings in such ways as adjusting weighting and bias in its nodes and levels of assessment. It learns from the experience without being told how to do it by a human being.
The training process is repeated many times with many pairs of tracks and the ANN learns, by repetitive correction, how to produce property vectors whose relative distances reflect semantic similarity or dissimilarity. The training continues until the ANN is “getting it right,” at which point it is “frozen” and ready to perform its intended function.
When the ANN is ready, it may take any given music track provided/proposed by a user, determine its physical properties and attribute a property vector to it. The ANN then relates a property vector to the vectors in an overall database and ascertains music that is semantically similar by looking for music tracks with proximate physical vectors and makes a recommendation of a similar track from those nearby vectors.
In the present case, to determine whether the computer-program exclusion is engaged, the judge asked the most fundamental questions:
“Where is the computer?”
“Where is the program?”
In this process, various dictionary definitions of a computer program were investigated, which were all consistent with respect to the fact that a computer program contains: “a set of instructions that makes a computer do a particular thing.”
However, ML (Machine Learning)/AI (Artificial Intelligence) eliminates the need to define hand-crafted rules that strictly follow a defined specification written by a programmer. Namely, the ML/AI technology is not processing data on a step-by-step instructional basis. Instead, this technology uses training data to learn the logic to solve a specific problem. Thus, ML/AI does not follow an ‘if-then’ statement approach.
From this starting point, the judge discussed the circumstances under which the claimed invention operates. The judge reached the conclusion that the training stage involves a computer program because a program is necessary to provide the training. The programming involves setting the training objectives in terms of the structure of the ANN and the training objectives. It is not possible to define the programming any further than that.
However, the internal training and the subsequent operation of the trained emulated ANN do not involve a computer program. At that point, there is no program because no person had given a set of instructions to the computer to do what it does – the ANN had trained itself.
Consequently, as a matter of construction the claim is not to a computer program at all. The exclusion is not invoked.
Indeed, this is a new take on how AI related inventions may be regarded, which appears to open for new and interesting ways to protect such subject-matters.
Following the judgment, the UKIPO made an immediate change to practice for the examination of ANNs for excluded subject matter. Patent Examiners should no longer object to inventions involving an ANN under the “program for a computer” exclusion of section 1(2)(c) Of the Patents Act.
We will continue to monitor the development of the AI patenting practice in the UK as well as in the rest of the World and will revert with an update as soon as there is something new to report.
Text: Joakim Wihlsson
Chancery Division of the High Court
Examination of patent applications involving artificial neural networks