Protecting AI inventions in the UK

The supposed ‘dangers’ of artificial intelligence (AI) seem to be a regular feature in the news. Such scare stories should usually be taken with a pinch of salt. However, there is no denying that AI is an enormous and continually growing area, and this is supported by the number of new patent filings for AI-related inventions.

As reported by GlobalData, global patent filings in the field of AI recorded the highest average annual growth rate amongst all technology areas between 2018 and 2022 (a 28% average annual increase). Inventions in this field might relate to machine learning (ML) models, speech recognition, image analysis, natural language processing systems, etc.

But just how patentable are AI-related inventions in the UK?

Like other computer-implemented inventions, AI-related inventions are not necessarily excluded from patentability at the UK Intellectual Property Office (UKIPO). However, they do have to be shown to solve a technical problem in a technical way, and whether this can be successfully argued before the UKIPO can be difficult to predict. The UKIPO is often considered to be relatively strict when it comes to computer-implemented inventions compared to many other jurisdictions.

According to their recent guidelines for examining patent applications relating to AI inventions (available here), the UKIPO seems to divide most AI-related inventions into one of two categories: ‘Applied AI’ inventions and ‘Core AI’ inventions. ‘Applied AI’ inventions mostly apply AI techniques to a field other than AI itself, e.g. to solve a specific problem outside the computer on which the AI program runs, or to make the computer itself work better. ‘Core AI’ inventions, on the other hand, define advances in the field of AI itself, e.g. an improved ML model or algorithm.

‘Applied AI’ inventions are in general much easier to protect in the UK than ‘Core AI’ inventions. This is because the field to which the AI is being applied often provides a sufficiently technical purpose. For example, the use of AI techniques to monitor a gas supply system for faults, to analyse movement from motion sensor data, or to control a fuel injector in a combustion engine, is likely to be considered sufficiently technical by the UKIPO. Therefore, one way of improving the prospects for grant in the UK for an AI-related invention is to limit the claims to a particular ‘technical’ or real-world application of the AI method. Not all applications of AI are likely to be helpful for patentability, however. Applications relating to business methods or purely administrative tasks, such as the use of AI to analyse patient health records or to process financial information, are unlikely to be considered sufficiently technical by the UKIPO.

In contrast to ‘Applied AI’ inventions, a ‘Core AI’ invention is considered to be limited to the field of AI itself. This means that, unlike ‘Applied AI’ inventions, ‘Core AI’ inventions are unlikely to be directly concerned with the real-world application of AI models and algorithms to technical problems external to, or within, a computer system. This makes such inventions more difficult to protect in the UK. It appears that, to be considered patentable, the effect produced by a ‘Core AI’ invention must be something more than an effect derived by the invention being just a ‘better’ algorithm running on a conventional computer. Rather, the ‘Core AI’ invention must bring about a change to the technical operation of a conventional computer, not merely a change to the way a program or algorithm works.

In an effort to provide some guidance on this issue, the UKIPO has provided a set of scenarios defining examples of AI-related inventions and reflecting how the patentability of such inventions would likely be assessed. Examples of ‘Core AI’ inventions that are considered not to be excluded from patentability include: processing a neural network on a heterogeneous computing platform; a special purpose processing unit for machine learning computations; and a multiprocessor topology adapted for machine learning. In each of these cases, the UKIPO emphasises that what is important is that the computer on which the AI program runs is made to operate in a new way in a technical sense.

Examples of ‘Core AI’ inventions that are considered (hypothetically) to be excluded from patentability include: optimising a neural network; avoiding unnecessary processing using a neural network; and active training of a neural network. In these cases, the UKIPO states that there would be no technical effect beyond the mere running of a ‘better’ program, and no solving of a technical problem to do with the workings of the computer itself, and therefore these inventions would not be considered patentable.

The UKIPO’s comments on some of these scenarios seem questionable, for example it could be argued that ‘avoiding unnecessary processing’ does indeed improve the performance of the computer itself, not just the neural network running on the computer. However, it is worth noting that the UKIPO stresses that the example scenarios it provides are not a source of law and are non-binding. That being said, some recent decisions by the UKIPO have cited these scenarios (e.g. as comparisons) when assessing the patentability of actual inventions. This suggests that some weight may indeed be given to these recent guidelines, although how much weight remains to be seen.

We are a European firm and assist our clients to protect their IP rights in the UK, Europe and worldwide from our offices in the UK and The Netherlands and through our international network of trusted local attorneys.

Get in touch if you would like to discuss your innovations and brand protection further.