Patenting Artificial Intelligence – what do you need to disclose?

AI Data Net

One of the challenges in filing a patent application is the level of detail of the invention in the description. Patent monopolies are only granted if the invention is completely described in a manner that is “sufficiently clear for it to be carried out by the person skilled in the art”. This provides a guideline, but no definite instructions, and until now the level of detail required for artificial intelligence based patent applications has not been clear. Earlier this year, however, the European Patent Office’s Board of Appeal provided some helpful insights.

The case involved a method and apparatus for the determination of a cardiac output of a patient’s heart using a measuring device and a computing unit. The applicant wrote in the description that the invention used an artificial neural network to calculate the cardiac output of a patient’s heart only from a peripherally measured arterial blood pressure. However, there was little detail given about the neural network and how it was programmed.

The patent examiner searched the prior art and found solutions for measuring the cardiac output. These solutions  required surgical intervention (insertion of a device or of a fluid in the patient’s blood stream or heart) or involved measurements and subsequent complex calculations that were prone to error or misreading. The applicant claimed that it had developed an improved solution based on the use of an artificial neural network to map the arterial blood pressure to the aortic pressure and thus enabling a reliable determination of the current cardiac output. The artificial neural network was trained with patient data (e.g. age, gender, overall health, etc.) but the application included no details of the actual data used for the training of the artificial neural network.

The Board of Appeal concluded that a mere description of the use of an artificial neural network and a vague indication of the training data used is not sufficient. It is necessary to describe actual examples of the training data to justify grant of the patent and the inventiveness compared to the prior art.

The Patent Office‘s message was clear – without examples of the training data and the explanation of how the training was carried out, then an applicant cannot expect its patent to be granted. A good patent description will discuss the prior art and how the invention solved the known problems. In patent applications around machine learning and artificial intelligence, the discussion will need to include details on how the training was carried out, and how the trained networks contributed to the invention.

We’ve been able to support a number of applicants in moving their AI-based inventions through the patent office. Feel free to contact us if you have any questions.

Dr. Robert Harrison / European Patent Attorney
Beat Wüst / Trainee Patent Attorney

Leave a Reply

Your email address will not be published. Required fields are marked *