On 1 March 2024, the European Patent Office (EPO) updated its Guidelines for Examination regarding AI inventions. This included new wording relating to the sufficient disclosure of training data.
Specifically, the new Guidelines state in F-III-3 that there may be insufficiency in the invention where "the mathematical methods and the training datasets are disclosed in insufficient detail to reproduce the technical effect over the whole range claimed. Such a lack of detail may result in a disclosure that is more like an invitation to a research programme"; and in G-II-3.3.1 that "the technical effect that a machine learning algorithm achieves may be readily apparent or established by explanations, mathematical proof, experimental data or the like. While mere allegations are not enough, comprehensive proof is not required either. If the technical effect is dependent on particular characteristics of the training dataset used, those characteristics that are required to reproduce the technical effect must be disclosed unless the skilled person can determine them without undue burden using common general knowledge. However, in general, there is no need to disclose the specific training dataset itself".
Since then, we have become aware of commentary from US practitioners suggesting that these updates introduce "ambiguity" to the disclosure requirements that will make US applicants unsure whether it is even worth filing AI related patent applications at the EPO.
However, from a European patent attorney's perspective, these new sections represent a relatively straightforward application of established EPO principles of sufficiency to AI inventions. For sufficiency, what matters is whether the skilled person, with their common general knowledge, has enough information to carry out the invention without undue burden (see e.g. F-III-1 of the Guidelines, as well as II-4.1 of the Case Law of the Boards of Appeal); this guidance has existed in the Guidelines since at least 2001 (see C-II-4.9 of the 2001 Guidelines). As applied to AI, the skilled person needs to know, or be able to determine without undue burden, the characteristics of the training data required to produce the technical effect. If the training data is standard, which it very often is, then not much will be needed, whereas if there is something special about the training data or how it is collected, then more detailed steps of data collection, or alternatively a publicly available dataset, should be provided.
In our opinion, it therefore seems at least some of the suggested "ambiguity" in the new Guidelines may have more to do with a lack of familiarity with longstanding European patent practice, than with an inherent ambiguity in the update itself.
Our conclusion is not that US companies should avoid filing in Europe, but that close collaboration between their European and US patent attorneys is to be encouraged to optimise the specification for both jurisdictions. With our experienced team of European and US patent attorneys working together, EIP is particularly well placed to help clients get the best protection possible for their AI inventions, globally.