Patentability Of Inventions Implementing Artificial Intelligence In Europe And France, The Question Of Inventive Step

Nowadays, artificial intelligence (AI) is at the heart of numerous technological innovations influencing major sectors such as health, transport and agri-food. In France and Europe, patent applications involving AI are multiplying. However, despite this enthusiasm, patenting an invention involving AI remains complex. The main problem is not so much the novelty of these technologies, but their abstract nature, which leads them to be often considered non-technical because they belong to the exceptions to patentability (see article Exceptions To The Patentability Of Mathematical Methods And Computer Programs In France And Europe, With A Focus On The Special Case Of Artificial Intelligence In Terms Of Patents.) as defined in the law.

In addition to the outright exclusion from the field of patentability discussed in the aforementioned article, this abstract nature also plays a central role in the assessment of inventive step, one of the fundamental criteria for obtaining a patent. In Europe, two major decisions structure the analysis of AI-based inventions: the COMVIK decision (T641/00), which defines the basis for the evaluation of so-called ‘mixed-type’ inventions, i.e. those that include technical and non-technical characteristics, and the G1/19 decision, which specifies its application to computer simulations - often similar in their functioning to AI systems.

In general, the assessment of the inventive step of an invention consists in determining whether the invention does not obviously derive from the state of the art for a person skilled in the art. At first glance, this criterion seems conventional. But as soon as the invention relates to AI, its assessment becomes more complicated.

Indeed, neither European nor French law authorizes the patentability of certain categories of inventions including computer programs and mathematical methods, unless they produce a technical effect. However, AI seems to belong to both these categories. The difficulty is therefore to demonstrate that AI does not just manipulate data but that it contributes to solving a technical problem.

The COMVIK decision laid the foundations for the examination of inventions containing both technical (e.g. a sensor, a processing system, a processor, etc.) and non-technical (e.g. an algorithm, an optimization model, etc.) elements. It establishes a simple rule: only technical features can be taken into account when assessing inventive step. In other words, if the innovation is based on an AI model that is unrelated to a technical system, or has no technical effect, it will not be patentable.

Decision T 1543/06 (GAMEACCOUNT) subsequently supported this principle by stating that the COMVIK decision could be rephrased as follows: ‘an invention which, as a whole, does not fall under the exclusion referred to in Art. 52(2) EPC 1973 (i.e. having a technical character) cannot rely solely on excluded elements, even if it is new and non-obvious (in the usual sense of the term), to be considered as satisfying the requirement of inventive step (see also T 336/07). In other words, the inventiveness of an innovation cannot be based solely on an object excluded from patentability, however original it may be.

Decision 154/04 (DUNS LICENSING ASSOCIATES) also validated this approach by confirming that a non-technical feature can produce a technical effect, for example if it contributes to the solution to a technical problem.

The COMVIK decision and subsequent decisions therefore imply that, when characteristics that can be considered non-technical as such interact with technical characteristics to provide a technical solution to a technical problem, and these characteristics do not clearly derive from the state of the art, the inventive step of a claim including such mixed characteristics is recognized.

G1/19, a decision handed down by the Enlarged Board of Appeal of the European Patent Office in 2021, complements this approach. This decision relates more specifically to a computer simulation of the movement of pedestrians in a building. The question that G1/19 sought to answer was whether a purely digital simulation could be patented if it does not physically interact with the real world.

The answer, somewhat nuanced, reiterated that, as with COMVIK, the technical effect is the central condition. A simulation can be patented if it contributes to an identifiable technical solution, even in a digital environment. It is therefore not compulsory for an AI to control a machine or produce a physical output to be patentable. On the other hand, the model must do more than simply calculate or model. In particular, a digital improvement to a computer system may be sufficient. For example, an AI that improves the functioning of a computer system, reduces the consumption of resources or provides data that can be used by an industrial process can be patented. The inventive step is still assessed according to the COMVIK approach; only the technical characteristics are taken into account.

The interweaving of COMVIK and G1/19 decisions produces a strict but coherent framework. COMVIK defined the method for evaluating mixed-type inventions, while G1/19 clarifies that certain methods, even purely numerical ones, can produce a technical effect if they have a clear and concrete purpose.

Thus, when drafting patent applications relating to inventions involving AI, it is not enough to simply state that a system or method involves a processor implementing an AI model. It is necessary to describe precisely how this model interacts with a technical environment and what improvements it enables in order to solve a well-defined technical problem.

Internationally, the assessment of the inventiveness of an invention implementing AI is more permissive, particularly in the United States and China.

In the United States, it is theoretically forbidden to patent abstract ideas such as intellectual or purely mathematical methods. However, if these are applied to a concrete problem, and their functioning is sufficiently described, a patent may be granted. In China, the conditions are even more reduced, since it is sufficient that inventions using AI can have a concrete industrial application.

The inventive step of innovations using AI in France and in Europe is therefore assessed according to a stricter framework than internationally. It is therefore advisable to adapt drafting practices when filing a patent application in these territories. In particular, when drafting such applications, it is advisable to emphasize the technical contribution of the invention by focusing on the functional and technical aspects of AI.

 Credit: Brunhilde GREIN