It is well established that the EPO will grant a patent for an invention involving machine learning. Often, high quality training data is key to the ability of the trained algorithm to act as an effective classifier and as such can give rise to technical character. Therefore, it is unsurprising that a failure to disclose an adequate description of input data, or how to obtain such data, for training the algorithm can led to a finding of a lack of sufficiency and inventive step.
A recent EPO Board of Appeal decision (T0161/18) highlights the importance of drafting such applications carefully so that sufficient details of the training dataset are included and provides useful guidance on the level of disclosure required.
The patent application at issue in T 0161/18 related to a method for determining cardiac output from an arterial blood pressure curve. The purported invention was to use a peripheral blood pressure curve to estimate an equivalent aortic pressure, where weighting values used in the estimation were calculated with the help of an artificial neural network. However, it was decided that the training of the neural network could not be carried out for lack of disclosure in the application. The net result was that the application was refused on grounds of insufficient disclosure.
The requirements for sufficiency are set out in Article 83 EPC and states that an invention be disclosed so clearly and completely that a person skilled in the art can carry it out. The board stated that, for this purpose, "the disclosure of the invention in the application must enable the person skilled in the art to reproduce the technical teaching inherent in the claimed invention on the basis of his general specialist knowledge".
With regard to the training of the neural network, the description stated that the input data should cover a wide range of patients of different ages, genders, constitutional types, health status and the like so that the network did not become specialised. However, the description did not disclose which input data were suitable for training the artificial neural network according to the invention or at least one data set suitable for solving the technical problem of the invention.
The board stated that, "the training of the artificial neural network cannot therefore be reworked by the person skilled in the art, and the person skilled in the art cannot therefore carry out the invention". This is because the dynamic and unpredictable behaviour and blackbox character of an AI tool is influenced heavily by the data used to train it, opening up the question of the reproducibility and reliability of the purported effect. The board therefore concluded that the invention was not sufficiently disclosed and thus did not meet the requirements of Article 83 EPC.
Interestingly, the board also considered the invention to lack an inventive step for the same reason, i.e. the invention was not sufficiently disclosed. Although not explicitly stated as such, it appears that the lack of disclosure was such that the skilled person would have to supply the missing teaching from their own knowledge in order to carry out the purported invention and hence it was obvious and lacking in inventive step.
Given that AI algorithms per se are treated as mathematical methods by the EPO – a category excluded from patentability – relying upon an AI algorithm for inventive step requires claims to the use of the AI algorithm for a particular technical purpose. Transformation of a blood pressure curve can in principle provide this technical purpose but only in the case where the AI algorithm can be said to credibly achieve this transformation. The breadth at which the AI algorithm was claimed in this case led to embodiments being encompassed by the claims that did not credibly enable this transformation, leading to an inability to rely upon the AI algorithm for inventive step due to the further reason that it was (at least partially) excluded from patentability.
The board therefore considered that the claimed neural network was not adapted for the specific, claimed application and stated that there was therefore, "only an unspecified adaptation of the weight values, which is the nature of every artificial neural network".
Therefore, it appears that the Board considered the lack of specific detail on training datasets to be a deficiency in the application for both sufficiency and inventive step, leading to a dismissal of the appeal against the refusal of the application.
However, a note of caution: to minimise the risk of such objections on applications employing machine learning one may be tempted to include the data used to train the algorithm in full. While the inclusion of such data has been seen as a way to satisfy the requirement of Article 83 EPC, providing public access to datasets may be impractical and/or undesirable given their likely size and the intrinsic value they hold. Therefore, Applicants may instead choose to disclose only those parameters that govern the generation of the training data set that go to produce results consistent with the invention. That is, in relation to training data for AI, the sufficiency requirement may be considered met if, as particularly succinctly and well put by Francis Hagel in an article entitled "T 011/18 brings to the fore Art 83 for training data", published in epi information 4-2020, "the application discloses the methodologies for the selection of data sources and processing of data which are specifically adapted to enable the skilled person to prepare training data relevant to the objective".
If you have any questions about the drafting requirements for machine learning inventions or require advice in relation to intellectual property in general please contact your usual WP Thompson contact or lmu@wpt.co.uk or jmp@wpt.co.uk.
© WP Thompson Limited. Authors: Lewis Mulholland & Dr. Julian M Potter