News & Events about IP just for you
possessed photography rDxP1tF3CmA unsplash

Artificial Intelligence Patents at the EPO and UKIPO

The number of patent applications for inventions relating to Artificial Intelligence  “AI” is growing fast! But, can artificial intelligence be patented at the European Patent Office? The answer is both yes and no. The application of AI (including Machine Learning “ML”) and specific technical implementations of AI can be patented just as any other software. However, fundamental algorithmic/mathematical level artificial intelligence innovations are commonly held to be not patentable.

The EPO considers that AI and ML are based on computational models and algorithms which solely relate to mathematical methods. However, mathematical methods are excluded from patentability if they are claimed “as such” according to Art. 52(3) EPC, as these methods are not considered to have a technical effect. The EPO has adopted a “two hurdle” approach for distinguishing "inventions" from "non-inventions", based on a decision established in the appeal case T 641/00 (known as the COMVIK decision). The first hurdle relates to the requirement for the subject matter to be eligible for patentability and the second hurdle relates to the inventive step requirement.

The first hurdle is relatively low and can be overcome simply by identifying a single technical feature in the patent claim. For example, in apparatus claims a hardware element can be included, such as a computer; or a method claim can be defined as a “computer implemented method”.

The second hurdle is more complicated; the assessment of inventive step only takes account of the technical features that contribute to the solution of a technical problem. Therefore, the technical features embodying the “mathematical method” must contribute to the technical character of the invention or have a technical effect.

A mathematical method may contribute to the technical character of an invention, namely contribute to producing a technical effect, and thus be eligible for patentability if:

  • it serves a technical purpose by its application to a field of technology, and/or
  • by being adapted to a specific technical implementation (T 2330/13).

When assessing the contribution made by a mathematical method to the technical character of an invention, it must be taken into account whether the method, in the context of the invention, serves a specific technical purpose. A generic purpose such as "controlling a technical system" is not sufficient to confer a technical character to the mathematical method.

Moreover, the patent claim must be functionally limited to the specific technical purpose, either explicitly or implicitly. This can be achieved by establishing a sufficient link between the technical purpose and the mathematical method steps, for example, by specifying how the input and the output of the sequence of mathematical steps relate to the specific technical purpose so that the mathematical method is causally linked to a technical effect.

Alternatively, rather than a patent claim being directed to the specific technical purpose, the claim may be directed to the specific technical implementation of the mathematical method. The claim may contribute to the technical character of the invention when the mathematical method is particularly adapted for that implementation.

AI patent applications must be considered on a case-by-case basis and the answer to whether an application is patentable or not is developing in the case law of the Boards of Appeal. Practically, we can also look at the Guidelines for Examination in the European Patent Office specifically related to AI and Machine Learning.

The Guidelines give several examples of claimable inventions including a technical purpose/implementation:

  • controlling a specific technical system or process, e.g. an X-ray apparatus or a steel cooling process; 
  • digital audio, image or video enhancement or analysis, e.g. the classification of digital images, videos, audio or speech signals based on low-level features such as edges or pixel attributes for images are further typical technical applications of classification algorithms; or,
  • providing a medical diagnosis by an automated system processing physiological measurements, e.g.  the use of a neural network in a heart-monitoring apparatus for the purpose of identifying irregular heartbeats makes a technical contribution.

A separate criterion (albeit linked to the consideration of whether a technical feature is present) is whether the technical features are sufficiently disclosed. The question of how much detail to include is a difficult decision for the applicant, inventor and patent attorney alike. AI/ML solutions are often specific or sensitive to the choice of network architecture, input representation and/or training data. However, as discussed above, a specific technical purpose or implementation must be shown. This begins to rule out fundamental AI/ML improvements being patentable.

Examples of  AI inventions found to be insufficiently disclosed by the EPO are T0509/18, which relates to machine learning-based driver alertness detection; and T0161/18, which relates to training of an artificial neural network.

Therefore, AI as such is not considered technical, and merely providing a generic purpose for AI is not sufficient to confer the necessary technical character. Also, a generic AI with algorithmic efficiency is not considered to be technical. However, AI with a specific technical purpose can be considered technical for the purposes of the EPO’s second hurdle. When drafting a claim, the invention must be functionally limited to the specific technical purpose, either explicitly or implicitly, and the direct technical relevance of the AI derived results must be indicated.

A recent decision of the Enlarged Board of Appeal, namely the G 1/19 decision, tested the EPO’s approach to assessing the requirements for establishing eligibility for mathematical related inventions. The patent which was subject to the appeal, related to the modelling and simulation of the movements of a pedestrian in an environment, such as for evacuation of a building or stadium in an emergency. The decision confirmed the “COMVIK” approach is to be used in deciding the patentability of computer implemented simulations in the same way as any other computer-implemented invention. No definition of “technical” was given, meaning technicality will also continue to be decided on individual facts of the case. Importantly, the decision implied that computer implemented inventions, including AI/ML inventions would be assessed in the same way. Even if an AI invention does not include a computer simulation, it will still have to be demonstrated to be more than a computer simply running instructions in order to solve a technical problem in order to be eligible for patentability.

The UKIPO approach is very similar to the that of the EPO. The Manual of Patent Practice notes that AI inventions, particularly inventions involving computer simulation, are generally computer-implemented and relate in some way to a mathematical method. If an AI invention is directed to a specific technical process outside of a computer or if it forms part of the internal workings of a computer, then it will likely not be excluded. If an AI invention is directed to a non-specific purpose, it may also fail by way of encompassing excluded processes. AI inventions or algorithms that provide an improvement in programming will likely be excluded. Currently, there are no UK court decisions directly relating to the patentability of AI subject matter.

The World’s five largest patent offices* have joined together to advance their co‑operation in the area of new emerging technologies such as AI by setting up a special task force to co‑ordinate their initiatives, including examination practices. The similarities between examination guidance at these Offices is promising at least in terms of certainty for applicants. A common theme is the need to specifically define the technical implementation or purpose of the AI based invention. Therefore, to be successful, the drafting of AI/ML patents may require more focus on examples and data than for other types of software invention.

*(Korean Intellectual Property Office (KIPO), European Patent Office (EPO), Japan Patent Office (JPO), China National Intellectual Property Administration (CNIPA) and United States Patent and Trademark Office (USPTO)).

Matthew Veale, Patent specialist in AI and computer related inventions

Dr Elliott Davies, Partner and Patent Attorney

Additional Reading$File/comparative_study_on_computer_implemented_inventions_software_related_inventions_EPO_CNIPA_en.pdf


Related News

How can we help