In decision BPatG 19 W (pat) 32/20, the German Federal Patent Court affirmed a decision refusing a German patent application for lack of enabling disclosure. According to the reasons of the decision (in particular section II.5.2(3)(iii)), the Senate considered the structure of the AI model to be insufficiently disclosed. The Senate also considered the information on the training data to be insufficient, in particular when it comes to how the training data are used to train the ANN structure to provide the desired output quantity (relating to wear of a technical system).
By way of background, it is rare that an application or patent is considered insufficiently disclosed by the German Patent and Trademark Office or by the German Federal Patent Court. This applies in particular in the arts that are considered “predictable arts” (physics, mechanics, etc.). The decision reflects a strict practice with regard to AI-based inventions, as also evidenced by the EPO Boards of Appeal.
There appears to be a tendency in the office practice (in particular in the EPO Boards of Appeal) to affirm refusals of applications or revocations of patents for AI-based inventions, based on lack of inventive step, when the invention relates to using AI in a novel, specific field of use. The office’s rationale is often that using AI is a general trend and the skilled person would consider the use of AI in a novel field without exercise of inventive skill. Little consideration appears to be given on the burden imposed on the skilled person for training an AI model to make it amenable for the novel use. There appears to be a discrepancy between (a) such a lack of inventive step rationale and (b) the strict criteria imposed by the office practice and the EPO Boards of Appeal / the German Federal Patent Court (see cited decision) on the disclosure of an application/patent. One can only hope that the EPO, GPTO, and BPatG will find a more balanced view in the future.