A recent decision by the UK High Court has delivered a highly favourable verdict for AI innovators, which could change the way that the patentability of AI-related inventions is assessed in the UK.
In Emotional Perception, AI Ltd v Comptroller- General of Patents, Designs, and Trade Marks,  EWHC 2948 (Ch) the High Court decided that the UKIPO was wrong to refuse Emotional Perception's patent application as being excluded from patentability.
If the decision is followed, then it would suggest that the UK Intellectual Property Office (UKIPO) stands to be a more favourable destination for obtaining patent protection for AI-related inventions than the European Patent Office (EPO) and could open the door for patent protection for subject matter which might previously have been excluded from patentability.
Recent years have seen an enormous increase in the use and development of AI. As well as news headlines, this increase has also been reflected in large increases in the number of patent applications filed for inventions directed to, or utilising, AI, and indeed many AI-related inventions are eminently patentable. However, obtaining patent protection for AI-related inventions at the UKIPO and the EPO often requires careful consideration due to some of the exclusions to patentability that apply in Europe. Both the EPO and the UKIPO have issued patent examination guidance which treats the training and implementation involved in AI and machine learning as mathematical methods implemented by computer programs. This is significant because computer programs and mathematical methods "as such" are excluded from patentability by statute. In practice, this doesn't render all AI inventions as unpatentable. However, it does mean that in order to benefit from patent protection, an AI invention must be shown to provide a further technical effect either through contributing to the solution of a technical problem which exists outside of the computer or through special consideration of, or interaction with, the underlying hardware of the computer on which it is implemented. As a result, many of the headline-grabbing AI innovations that are the topic of the news might be difficult to obtain patent protection for at the UKIPO and EPO.
Emotional Perception's patent application relates to the training and use of Artificial Neural Networks (ANNs) for identifying semantically similar content. A particular example, which is the focus of the decision, is a music file to which subjective semantic descriptions such as "happy," "sad," or "relaxing" might be attached. An ANN is trained based on a series of pairs of files, between which two distances are derived. A first (semantic) distance is derived as a measure of the similarity of the semantic descriptions associated with the files. A second (property) distance is derived as a measure of the similarity of measurable properties of the files (e.g., tone, timbre, speed, loudness, etc. in the case of a music file). The ANN is specifically trained to output second (property) distances (related to the similarity of measurable properties) which converge with the first (semantic) distances. Once trained, the ANN then receives a new file (with no semantic description but with measurable properties), and generates output distances with respect to a database of reference files, where the output distances represent a semantic similarity (or otherwise) between the new file and each of the reference files. In this way, reference files in the database are identified which are semantically similar to the new file and sent to a user device (e.g., to recommend similar music to the user).
The application was refused by the UKIPO on the grounds that it was excluded from patentability by virtue of being directed to a computer program "as such." In particular, it was found that: i) training or implementing an ANN does not in itself produce a technical effect, and ii) an improved recommendation of a file which has semantic similarity with an input file does not produce a technical effect (because its improvement is subjective and of a cognitive nature).
Notably, the High Court disagreed with both of these conclusions. In particular, it was decided that an ANN is not a program for a computer, even when implemented in software, and was seen to operate at a different level from the underlying software on the computer. This conclusion appears to have been reached, at least in part, because an ANN was seen to be operating according to something that it has learned itself rather than implementing code which has been given to it by a human. It was also found to be persuasive that a software emulation of an ANN is equivalent to an ANN implemented in dedicated hardware, which it was argued, is not a programmable computer at all and thus would not fall within the statutory exclusion of computer programs.
In addition to considering whether or not the invention is a computer program at all (it was decided that it was not), the judgement also considered whether the invention could be seen to provide a technical contribution (which would have been a relevant question if it was decided that the invention was solely a computer program). On this question the High Court also disagreed with the UKIPO noting that "the ANN has certainly gone about its analysis and selection in a technical way... [i]t is not just any old file; it is a file identified as being semantically similar by the application of technical criteria which the system has worked out for itself". It was therefore held that the sending of a file selected by the trained ANN as being semantically similar to a target file did provide a technical effect outside of the computer irrespective of whether a user receiving the file ever listened to the file.
The judgement could yet be appealed by the UKIPO. However, if the decision is followed then it appears that the examination practice of AI-related inventions in the UK will need to change. Indeed, at the time of writing the UKIPO's website indicates that their examination guidelines for AI-related inventions have been temporarily suspended pending consideration of the judgement.
The finding in particular, that an ANN trained through machine learning, is seen to operate at a different level to the underlying software of the computer, would appear to open up patentability for a large range of subject matter that otherwise would not have been considered to be patentable. For example, the argument (which was found to be persuasive) that a trained ANN is a technical entity because it operates according to weights and biases that it has determined itself through training (rather than being dictated by a human programmer) is equally applicable to any model trained through machine learning. It is notable also, that whilst the judgement is based largely on discussion of application of the invention to media files, the patent claim more broadly encompasses applications to video, audio, image, or text files. The semantic analysis of text files would historically have been treated as a non-technical purpose by the UKIPO, but under this decision appears to have been deemed to be patentable, which could have wide-reaching consequences for the patentability of AI-based language processing.
The Emotional Perception patent application on which the judgement is based has also been filed in other jurisdictions, including in front of the EPO, where it has faced significant objections. It will be interesting to see how prosecution of the application plays out in Europe but given the current examination practice of the EPO, it is likely that it will continue to face objections. If the decision is followed rather than appealed, the UKIPO could therefore become a more attractive forum than the EPO for patenting AI-based innovations.
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.