How to Get a Patent for AI-Related Inventions in Korea
Young Cheol Byun, Yoon & Yang (IP) LLC (Korea)
Artificial intelligence (AI), a major component of the wave of technological innovation known as the Fourth Industrial Revolution, made rapid progress in the 2010s. Countries including South Korea are devoting so much effort to developing AI technology because of the huge effect it will have on human lives. AI will have a sweeping impact not only on human occupations and working hours and methods, but on all areas of our daily lives.
In order to adapt to developments in the IP environment and provide sufficient IP services, the Korean Intellectual Property Office (KIPO) implemented appropriate measures to improve its patent examination.
First, in light of hardware, an organizational restructuring led to the establishment of a new “Convergence Technology Examination Bureau” dedicated to the examination of technologies related to the Fourth Industrial Revolution, such as artificial intelligence (AI), big data, and bio-health.
Along with this, in light of software, KIPO announced “Examination Guidelines of AI-related Inventions” in January 2021. In the guidelines, KIPO outlines specific guidelines on description and novelty/inventive step requirements for different categories of AI inventions such as AI program (software) invention and AI application invention.
1. Description Requirements
The specific technical configuration to enable the AI invention should be included in the specification. The technical configuration can be, e.g., training data, data preprocessing, trained model, and loss function, which can be more detailed depending on the different categories of AI inventions as follows:
(i) AI program invention
A. How to perform and embody the steps and functions of data preprocessing
B. Correlation between the raw data and the training data
(ii) AI application invention
A. Correlation between the input data and output data of a trained model. For the description of the correlation, ① training data is specified, ② the correlation among characteristics of the training data for solving a technical problem is described, ③ the learning model to be trained using the training data is specified, and ④ a trained model for solving the technical problems is generated by the learning data and the learning model.
2. Novelty/Inventive Step Requirements
Even for an AI invention, KIPO maintains its basic standards for assessment of novelty/inventive step, i.e., KIPO mainly takes into account a difference in the technical configuration (e.g., training data, data preprocessing, trained model, loss function, etc.) and the technically superior effects (beyond those from conventional AI technologies) accruing from the technical configuration. KIPO provides some examination cases as follows:
(i) Case 1: When the claimed invention is used in an entirely different industrial field and the application to the different industrial field contributes to overcome a problem which had been unsolved for a long time in a specific industrial field, or overcome a technical difficulty in a specific industrial field, or achieve superior effect by changing the industrial field, then inventive step can be recognized.
(ii) Case 2: When the claimed invention is identical to the cited invention in terms of the technical field and training data, but there is a difference in the learning model which results in superior effect, then inventive step can be recognized.
(iii) Case 3: When the claimed invention is identical to the cited inventions in terms of the technical field and learning model, and there is a difference between the claimed invention and prior art 1 in the training data, but prior art 2 discloses the corresponding feature to the training data, and there is no difficulty in combining the prior arts and there is no difference in effect, then inventive step may not be recognized.
(iv) Case 4: When the claimed invention is identical to the cited invention in terms of the technical field and training data, and a difference in the learning model can be derived using a simple design change, the inventive step may not be recognized. However, where specific configurations other than the training data and learning model are added to the AI invention, and thereby the claimed invention has a unique effect distinguished from the prior art, the inventive step can be recognized.
You may think that there remains some doubt as to whether the Examination Guidelines sufficiently reflect the characteristics of AI inventions, however, such details or characteristics of AI invention are gradually being reflected more clearly. Accordingly, while complying with the current guidelines, it will be necessary to keep an eye on accumulated examination data to which the guidelines are applied within KIPO.