Examination Guidelines of AI Inventions (Exam. Practice Guidance)

(Subject) Patent applications of inventions requiring machine learning based AI technology for implementing inventions
(Established in 2020) Description requirements of a specification and criteria on patentability examination are clarified, and specific cases of examination of AI-implemented technologies are presented.
Established in 2020
1 Description requirements of a specification
  • A specific means for implementing AI inventions (learning data/pre-processing/learning model, etc.) should be presented and described in a specification.
  • A correlation between input data and output of a learning model should be described.
  • Where data pre-processing technology is featured, a correlation between raw data and learning data should be specifically described.
  • Reinforcement learning based AI technology should have an agent, environment, condition, actions and rewards as required features.
  • AI-applied inventions may describe a general training model name (Description method).
2 Patent eligibility
  • Current examination criteria on computer-SW inventions are maintained.
  • Software information processing is specifically implemented in use of a hardware.
3 Novelty/ Inventive step
  • Where AI technology is not specifically described, it is regarded as simple use of a disclosed AI technology.
  • Specific means, such as pre-processing, learning model, learning data, etc., is specified, and if it is recognized that the means brings about ‘unexpected advantageous effect’, the inventive step is acknowledged.
  • Where BM technology is simply systemized with a disclosed AI technology, the inventive step is denied.
  • Where output data of an AI invention is specifically utilized and unexpected advantageous effect are produced accordingly, the inventive step is acknowledged.
  • Where the industrial field is different, such aspects or effect is considered (e.g., the change of the industrial field or overcoming of technological difficulties)

top button

(Revised in 2021) Definition of a learning model is provided, and description requirements of claims are clearly presented where a learning model is described in the claimed subject.
Revised in 2021
1 Definition/requirements of learning model
  • Description requirements of a specific means and definition of a learning model are provide
  • Learning data, date pre-processing, loss function, etc. are presented as a specific means to be described for embodying AI inventions.
2 Example cases
  • Examples of a claim of a learning model having a neural network as a basic structure are presented.
  • Examples of a claim of a product using a learning model are presented.
3 Deficiency in description
  • Provision(s) regarding deficiency in description and its explanation are presented based on a real case.

top button

Examination Practice Guidance in the Artificial Intelligence field


IP5 Offices’ Examination Practice Guidances in the Artificial Intelligence field

  • Last updated 23 May 2024
  • Artificial Intelligence & Big Data Examination Division