A short primer on patenting AI and machine learning inventions at the European Patent Office

A short primer on patenting AI and machine learning inventions at the European Patent Office

AI is booming. From banking to medical devices to self-driving vehicles, it's ubiquitous. But what can be patented in this area? This is a short guide on how the EPO treats AI inventions. To keep it concise, we'll look at two types of AI invention that the EPO accepts and then two it won't.

Safe territory

The EPO will grant patents to applied AI. Put more correctly, it will grant patents to AI applied to a particular technical application. Good examples include machine learning for image processing, computer vision or in medical devices like pacemakers or insulin pumps. This is the first of two safe harbours of patentability for AI in Europe, and is the easier of the two to understand: European patents have been granted for real-world technological improvements since the EPO was founded.

The second, more abstruse allowable safe harbour is when the AI has been adapted to a specific technical implementation. What does that mean? It means the design of the algorithm has been adapted to the particular internal functioning of the computer as opposed to just being maths. To illustrate the point, a method of training a neural network which defines a balance of when to cache information in a memory and when to recompute it should be allowable. This is because it considers features of the computer and not just the algorithm. It's a subtle, but important distinction. And one that moves an otherwise unallowable idea into the second safe harbour of patentability for AI.

Dicey ground

There are two danger zones that would-be patentees should be wary of when trying to patent AI in Europe. The first of these involves trying to patent the fundamental AI or ML algorithm on its own. This isn't possible in Europe. This is because AI and ML are based on computational models and algorithms for classification, clustering, regression and dimensionality reduction, such as neural networks, genetic algorithms, support vector machines, k-means, kernel regression and discriminant analysis. These in themselves - without any consideration of a technical application or technical implementation - are considered mathematical methods. And you can't patent maths in Europe.

The second danger zone involves the non-technical application of AI, most notably in areas such as finance, administrative methods, linguistics and even some classification algorithms. Somewhat counter-intuitively, the EPO will consider many AI innovations in these areas unallowable. The rationale behind this is simply that the applications themselves are not considered technical. Maths combined with a non-technical application isn't patentable in Europe.

Navigating the way

For AI cases, you need to be particularly careful to ensure your case doesn't blunder blindly into unpatentable territory. We can help you. This might simply involve helping you understand whether the European system is suitable for your AI invention. But in many instances, with the right care and attention applied when drafting the case, the danger zones can be avoided and one of the safe harbours reached.

Gareth Fennell is a partner in our Tech group. He advises clients on patent strategy in the fintech, software and medical device sectors

For more information please contact Gareth or your usual Kilburn & Strode advisor.

Let us keep you up to date. If you’d like to receive communications from us, ranging from breaking news to technical updates, thought leadership to event invitations, please let us know.

Connect with us

Cookies improve the way our website works. By using this website you are agreeing to our use of cookies. For more information see our cookie policy I accept