1 June 2021

G1/19 – what does it mean for AI?

On 10 March 2021, the Enlarged Board of Appeal of the European Patent Office (EPO) handed down its landmark decision in G1/19 (Pedestrian Simulation). This is the first time since the European Patent Office (EPO) was founded that the EPO’s highest court has considered questions on the patentability of computer-implemented inventions in an admissible referral.

Venner Shipley has been involved with the case from the very start: from preparing and filing the application, handling examination, to the appeal before the Technical Board of Appeal and the proceedings before the Enlarged Board.

The patent application in G1/19 relates to a method of simulating pedestrian crowd movement which can be used to help design or modify a venue, such as a railway station or stadium. It seeks to provide a realistic simulation, in real-world situations, which cannot be adequately modelled by conventional simulators.

The application was refused during examination on the grounds that it lacked an inventive step: the examining division argued that a simulation model was non-technical and that its implementation on a computer was obvious.

On appeal, the Technical Board of Appeal (TBoA) accepted that the simulated environments, when they exist in physical reality, are technical and that an environment’s “behaviour” when a crowd of pedestrians moves through it, for example the rate at which pedestrians can pass through the environment, is a technical property of the environment, not unlike the ability of a roof to drain rainwater. The TBoA further accepted the appellant’s case that there was an evident analogy with case T1227/05 (Infineon) in which it was found that simulation of a circuit subject to 1/f noise constituted an adequately defined technical purpose. However, the TBoA questioned the reasoning in the Infineon decision and considered that a referral to the Enlarged Board of Appeal was needed in order to resolve questions on the patentability of simulation methods.

In a carefully reasoned decision, the Enlarged Board has found that computer-implemented simulations can be patentable but must, like any other computer-implemented invention, be examined on a case-by-case basis to determine whether the standard “technicality” criteria are met using the well-established “COMVIK” approach. Importantly, the Enlarged Board found that in this assessment it is not sufficient that the simulation is based on technical principles underlying the simulated system or process. That is, it is not decisive whether the system or process being simulated is technical or not. Instead, what is relevant is whether the simulation of the system or process contributes to the solution of a technical problem.

However G1/19 is about more than simulations. It is now the most authoritative summary available on the established EPO case law relating to the technicality of computer-implemented inventions and the approach that should be applied to their examination. It tackles a number of aspects of this approach which are of very general relevance for computer-implemented inventions, such as the question of how and when “technical effects” or “technical interactions” may occur in the context of a computer-implemented process, the question of whether “potential technical effects” (i.e. effects achieved only in combination with non-claimed features) should be taken into account, and the question of whether a technical effect requires a “direct link to physical reality” (fortunately for patent applicants the answer to the last question is “no”).

What does G1/19 mean for inventions relating to artificial intelligence (AI) and machine learning (ML)? There are evident parallels between AI/ML and simulations, not least that both employ mathematical models. Indeed, in popular usage AI is often seen as an effort to simulate aspects of human intelligence. Prior to the decision in G1/19, concerns were raised that depending on the answers given by the Enlarged Board, the decision could potentially have negative consequences for the patentability of AI.

These concerns have not been realised. The decision confirms that no group of computer-implemented inventions can be excluded from patent protection as a matter of principle and that all computer-implemented inventions should be treated in the same way when examined at the EPO. It largely affirms the existing case law and maintains the legal status quo, with the clarification that it is not decisive for technicality of a simulation whether the system or process being simulated is technical or not. However, this clarification is not expected to have a significant impact on practice for AI/ML. The prosecution case for technicality of AI/ML inventions is not normally based on a contention that the machine learning model somehow reflects an underlying technical system or process. Instead, in view of current EPO examination guidelines, arguments normally focus on the application of algorithms or models to a technical use case, or on adaptations “motivated by technical considerations relating to the internal functioning of the computer itself”. Such arguments are clearly still accommodated within the framework set out in G1/19.

It is possible to go further and draw a degree of analogy between the current EPO examination guidelines for AI/ML and the conclusions of G1/19 for simulations. The current EPO guidelines (section G-II, 3.3) allow two types of argument for technical effect of AI/ML invention: 1) by application to a field of technology or 2) by being adapted to a “specific technical implementation”. Analogy can be drawn to the comments in G1/19 regarding the technicality of simulation models: G1/19 suggests that the constraints defined by such models can be technical if they for example 1) “form the basis for a further technical use” or 2) if they are a “reason for adapting the computer or its functioning”.

However, in contrast to the current guidelines, G1/19 makes it clear that these two possibilities are just examples. Applications should be examined on a case-by-case basis under the COMVIK approach, and “any technical effect going beyond the normal electrical interactions within the computer” may be considered for inventive step. Thus, the decision contemplates a broad base for technicality arguments which is not limited to the two situations set out in section G-II, 3.3 of the current guidelines.

At least some revisions to the guidelines will be needed in view of G1/19 and changes can be expected in at least section G-II 3.3.2 to reflect the Enlarged Board’s comments on T1227/05. Although more wholesale changes to reflect the general guidance that G1/19 provides would be welcome, it is considered more likely that the EPO will adopt a conservative approach and that only limited changes will be made. Even so, the influential nature of G1/19 may help to bring further clarity and consistency to AI examination practice by bringing the practice of the examining divisions more closely into line with that of the Boards of Appeal. For example, the current examination guidelines provide limited guidance on the criteria for assessing algorithmic inventions which are adapted to the internal functioning of the computer, but G1/19 provides additional assistance by referring to case T697/17 and the list of additional examples that it cites. Indeed, T697/17, which is cited several times in G1/19, provides much useful guidance itself, e.g. on the role of program performance improvements such as execution time, processing speed, latency and amount of memory required in distinguishing between technical and non-technical features.

While G1/19 aims primarily at ensuring the uniform application of the law, it also has an eye to the future. To this end the Enlarged Board deliberately refrained from defining the term ‘technical’, acknowledging that it must remain open, not least in anticipation of future developments, since “any definition of a technical invention may have to be extended in the course of time in order to accommodate new technical or scientific developments or to reflect societal changes”. This is ultimately a trade-off and reflects an inherent tension in the patent system. While a more concrete definition of “technical” might well have improved legal certainty and consistency of practice, the system cannot be too rigid if it is to be fit for purpose. By leaving the definition of “technical” open, the Enlarged Board has ensured a degree of flexibility to allow the patent system to adapt to cater for the results of future developments in new and emerging fields.

For more updates on the G1/19 case, click here.