Patentability of AI: Sections on AI and Machine Learning Added to the Guidelines for Examination of the EPO

The European Patent Office (“EPO”) has recently updated its Guidelines for Examination (“Guidelines”) by adding a section on AI and machine learning (Section 3.3.1 of Part G of the Guidelines). The updated Guidelines became effective on 1 November 2018.

The section on AI has been inserted under the part which concerns mathematical methods since the Guidelines state that the computational models and algorithms, which AI and machine learning are based on, are mathematical by nature. Mathematical methods as such are excluded from patentability under article 52 (2) (a) of the European Patent Convention (“EPC”). The previous Guidelines mentioned mathematical methods as an example of the principle according to which purely abstract or intellectual methods are not patentable.

In order for mathematical methods to be patentable, they must have a technical character. The technical character of an invention can be recognized only if it serves a technical purpose (T 1784/06). As examples of technical purposes which may be served by a mathematical method, the Guidelines mention separation of sources in speech signals and optimizing load distribution in a computer network.

When it comes to AI and machine learning, the examination of whether the claimed subject-matter has a technical character as a whole, which is required in article 52 of the EPC, must be performed diligently because expressions such as “support vector machine”, “reasoning engine” or “neural network” generally refer to abstract models without technical character. The requirement of technical character means that the subject matter must solve a particular technical problem using particular technical means (T 154/04). Organizational, administrative, commercial or mathematical algorithms don’t necessarily meet the said requirement which can lead to the dismissal of a patent application for example when protection is sought for classification algorithms (T 1784/06).

As an example of a technical application a heart-monitoring apparatus which identifies irregular heartbeats by using a neural network is mentioned in the Guidelines. Further, classification of digital images, videos, audio or speech signals based on low-level features (e.g. edges or pixel attributes for images) are further typical technical applications of classification algorithms. However, classifying text documents solely in respect of their textual content is however not regarded as a technical purpose per se (but rather a linguistic one).

The updated Guidelines can be seen as an affirmation that the EPO in principle acknowledges the patentability of AI and machine learning applications as long as they fulfill the prerequisite of having a technical character. Now that a specific section on AI and machine learning has been added to the Guidelines, it will be interesting to see if this will have an impact on patentability and if so, what the impact will be in practice.