How the JPO handles AI-related applications
New case examples of the ways in which the JPO examines AI-related patent applications are designed to demystify the process for applicants, explain the office’s Toru Matsuoka and Masataka Saito
On 30th January 2019, the Japan Patent Office (JPO) published 10 new case examples involving AI (artificial intelligence) inventions for patent examination. The addition of these latest case examples stems from the increase in the number of applications related to the AI field due to the recent rapid development of AI technologies such as deep learning, even though AI itself is not a new technology.
The potential varieties of AI inventions include inventions of AI itself such as algorithms, and the applications or uses of inventions based on AI technologies. The new case examples feature the latter technological field and show the examination practices that the JPO conducts to applicants who are not familiar with AI technologies and computer software inventions. To achieve easy-to-understand case examples, we created ones covering various technological fields and industries. We also provided both patentable examples and non-patentable examples to help users gain a clear understanding of the points the JPO considers in determining whether to grant patents to AI inventions.
In the case examples, the JPO focuses on the disclosure requirement and inventive step among patentability requirements, things that users are keenly concerned about. The examples deal with the extent to which disclosure in the description is required on AI inventions and cases in which the existence of the inventive step is approved for AI inventions. This is because patent subject matter eligibility of AI inventions is not a hurdle to obtain patents based on the JPO’s practices.
Overview of case examples for the description requirement
For the description requirement of AI inventions, it is generally required that there is some correlation between input and output data. We show that directly indicating the correlation in the description is not necessarily required, and detailed description of the correlation might be omitted if the correlation is apparent to a person skilled in the art. We also explain that it is enough to show the correlation indirectly by using the experimental results of an AI system for proving that the AI system has learned the correlation through machine learning, even if an applicant cannot analyze the correlation specifically.
We not only provide case examples of apparatuses or systems to which AI is applied, but also add a case example in which a substance itself is claimed, whose nature or characteristics are estimated by AI. This is connected to the development of material informatics, as to whether actual results from experiments on chemical substances are required.
Figure One
Overview of case examples for the inventive step
For the inventive step of AI inventions, we provide examples showing that the inventive step can be recognised when an advantageous effect is accomplished based on the choice of training data or preprocessing of training data. We also show that the inventive step might be denied when the invention is merely a systemisation by AI of known procedures or methods
Detailed information of case examples on AI inventions are available from our website.
The JPO hopes that these examples help users to properly and effectively protect their intellectual properties on AI inventions.