The Patentability of Neural Networks in Europe

The UK High Court has just ruled that neural networks are not excluded subject matter under the UK’s computer programs exclusion – has the UK just become the most patent-holder friendly jurisdiction in Europe for AI? By Kimberley Bayliss, Mark Richardson & Colin Paterson.
 
Before the Emotional Perception AI Ltd v Comptroller-General of Patents, Designs, and Trade Marks [2023] EWHC 2948 (Ch) (“Emotional Perception”) ruling by the UK High Court, the outlook for AI inventions in Europe had been stable for some time. Whether before the European Patent Office (EPO), or the UK Intellectual Property Office (UKIPO), the requirement for patent eligibility has essentially boiled down to whether the AI is used in a system that produces a technical effect. From this perspective, AI algorithms embedded in technical systems, like medical imaging systems, telecoms systems, and drug discovery systems have tended to constitute patentable subject matter (subject to the usual novelty and inventive step requirements). Conversely, AI systems used to process business-related data, or to produce improved data visualisations have tended to be rejected as non-technical. This is on the grounds of such systems being considered computer programs at the UKIPO, and business methods or presentations of information at the EPO. It is fair to say that the UKIPO has historically been the stricter of the two offices, to the point where, with some embarrassment, UK patent attorneys have for many years now gently tried to steer their clients away from making UK filings for AI-related inventions.
 
With this backdrop, the Emotional Perception  ruling therefore came as a shock to many of us in the UK patent profession. The case relates to a method of training a neural network to generate a score describing the similarity between two media files. The scores can be used to make recommendations for users in content streaming platforms. The invention had been summarily dismissed at the UKIPO as a mere computer program, but prevailed at the High Court which ruled that neural networks are not computer programs as such.
 
This is the first time that a court in the UK has made a ruling on the patentability of any machine learning system, and seems to suggest that at least one UK judge considers the current statute to have been routinely interpreted too narrowly.
 
So, should we all now be recommending that AI inventions be filed in the UK? We would tread a little more cautiously than that.  Although the UKIPO immediately issued new guidance stating that examiners cannot now raise computer program objections to neural networks, the UKIPO have also filed an appeal against the ruling. The fact that an appeal has been raised is noteworthy in itself, given that it comes against the backdrop of the UK Government’s National AI Strategy, and it will be interesting to see how this plays out.
 
The main reason for caution however is that there are other exceptions in the statute that the Judge didn’t rule upon that could still cause problems to inventions involving neural networks. As an example, a neural network that processes shopping data might still be considered a business method.
The most significant unknown at this stage is how the UKIPO will deal with inventions in the field of machine learning itself. New neural network architectures or improved learning algorithms have tended to fall under the mathematical method exclusions, and this ruling doesn’t directly address this. Thus, the frustrations in the European patent community around the lack of protection for new more fundamental improvements to machine learning are likely, for the time being at least, to continue.
 
Nonetheless, with respect to the titular question at hand: Has the UK just become the most patent-holder friendly jurisdiction in Europe for AI? The answer has to be yes. At the very least, a UK judge has just ruled that the technical effect provisions have been interpreted too narrowly. In Emotional Perception, the AI system selected files to serve to a user primarily on the basis of cognitive content, e.g. the aim was to serve up files that a user would enjoy. The judge ruled that serving up a file, for any purpose, is technical, even if the choice of file was ultimately made for the user’s enjoyment. This is a significant departure from EPO practice where systems based on cognitive considerations are likely to be considered non-technical, and this in itself could trigger a significant shift in practice for the UKIPO. Given the uncertainties and tendency to tread cautiously, there is of course, the risk that this could come to very little in practice, however, in view of the low filing costs at the UKIPO, and particularly for cases that are likely to encounter issues at the EPO, the UK is now looking attractive indeed.
 
Kimberley Bayliss is a partner at Keltie LLP, based in the Cotswold Office in the UK. She has spent the majority of her career helping multinationals develop their AI portfolios and is the co-chair of Keltie’s AI Group of specialist practitioners. She has a PhD in Astrophysics and specialises in the application of AI to Imaging, MedTech and Telecoms. 
 
Mark Richardson is a partner at Keltie LLP, based in the London Office. He is a member of Keltie’s AI Group of specialist practitioners with experience of cryptographic, automotive and telecommunications technology areas.
 
Colin Paterson is a partner at Keltie LLP, based at their London office. He is a member of Keltie’s AI Group with experience in AI applications in drug discovery and space technology, in control systems for wind energy and automotive applications, and in electronics and electrical engineering.