T 0161/18 – A reminder to make sure AI inventions are disclosed in sufficient detail at the EPO
The algorithm used by AlphaGo to defeat 9-dan professional Go player Lee Sedol in 2016 was a major milestone in artificial intelligence research. However, while filing numbers for patent applications in the field of artificial intelligence continue to skyrocket, many ‘AI inventions’ aren’t really about the AI: often, the AI component is just a clever tool that enables the true innovation. For example, Google’s patent US 9,678,664 relates to a method of correcting typos linked to the physical layout of the keyboard used. AI is used to identify and correct commonly mistyped patterns, but it could be argued that the invention is application of AI to this age-old problem, and not the AI itself.
So when AI isn’t the star of the show, only giving it a ‘superficial’ treatment when drafting a patent application is an understandable pitfall that can lead to uncurable objections at the EPO. Recent decision T 0161/18 of the EPO Boards of Appeal is just such an example.
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. The net result was that the application was refused on grounds of insufficient disclosure.
Neural network sufficiency requirements
Article 83 EPC requires that an invention be disclosed so clearly and completely that a person skilled in the art can carry it out. With regard to the training of the neural network according to the invention at hand, it was stated in the application 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 does not become specialized. However, the application did not disclose which input data were suitable for training the neural network, or at least provide one data set suitable for training, such that the invention could be carried out. The Board of Appeal therefore decided that the training of the artificial neural network could not be reworked by the person skilled in the art, and so the application was refused.
This case is a compelling reminder that it is best practice to always provide an in-depth example of how your invention can be carried out, no matter how trivial the details may seem to you or the inventors. If this application had simply included a set of training data on filing, there would most likely have been no sufficiency issue. Had this data not been available, it would have been prudent to delay filing of the application.
To reduce the risk of sufficiency issues in your AI applications, we suggest making sure you provide at least one detailed example of the input and output of your AI algorithm/neural network.
Where there is a machine learning/deep learning/big data aspect, be sure to include sample training data if possible, or at the very least unambiguously state which data are suitable for training.
If you would like to know more, please contact Nico Cousens or another member of the Kilburn & Strode patents team.