As of March 6, 3, the Japan Patent Office has added new patent examination cases related to AI-related technologies to the "Patent and Utility Model Examination Handbook."
In addition, on the Japan Patent Office website “About explanatory materials for cases related to AI-related technologies”,
-Material explaining the 10 newly added cases in slide format
-Full text of all cases related to AI-related technologies
We publish various information such as.
table of contents
Introducing new patent examination cases related to AI-related technology
Here, we will introduce the points indicated by the Japan Patent Office regarding the 10 newly added cases.
The full text of each case in the "Patent/Utility Model Examination Handbook" can be viewed from the link in parentheses.
Additional cases 1 to 4: Cases of determining inventive step requirements
Additional example 1: Automatic answer generation device for customer center (Annex A 5. Case 37）
Regarding the attempt to apply generation AI such as large-scale language models to business operations, we believe that it is a "simple systemization using artificial intelligence of work performed by humans" for the following reasons (1) and (2). This is a case in which the inventive step is denied because it is judged to be a demonstration of creative ability.
① Improving efficiency by systematizing tasks performed by humans and implementing them using computers is an obvious issue that those skilled in the art would normally consider.
② In the field of information processing technology, "replacing human judgment with machine-learned models" is a commonly used technology.
Additional example 2: Prompt sentence generation method for input to a large-scale language model (Annex A 5. Case 38）
Regarding the generation of prompts to be input to generation AI for large-scale language models, etc., those skilled in the art can understand the effects obtained by the specific method specified in the claims from the detailed description of the invention. This is a case where the inventive step was affirmed.
Additional example 3: Learning method for a trained model used to adjust the brightness of radiographic images (Annex A 5. Case 39）
While it was held that changing the configuration of the loss function for a trained model that estimates output data from input data is simply a design change or adoption of a design matter,
This is a case in which inventive step is affirmed because the cited invention has a structure that produces an effect that is not the focus of the cited invention, and an effect that is difficult to predict from the cited invention.
Additional example 4: Laser processing equipment (Annex A 5. Case 40）
An invention related to "systematization of tasks performed by humans using artificial intelligence" is recognized as an inventive step because a new feature in the training data used for learning has an advantageous effect compared to the cited invention. This is an example.
Additional cases 5 to 7: Examples of determining feasibility and support requirements
Additional case 5: Fluorescent compound (Annex A 1. Case 52）
The following ① to ③ are examples that meet the enablement and support requirements for "invention of a product estimated to have a certain function by AI."
① The evaluation of the actually manufactured product is stated in the specifications, etc.
②The prediction accuracy of the predicted value shown by AI is verified in the specifications, etc.
③At the time of filing, there was common technical knowledge that AI prediction results could replace the evaluation of actually manufactured products.
Additional example 6: Image generation method for training data (Annex A 1. Case 53）
The following ① and ② are examples that meet the support requirements for inventions related to "methods for creating teaching data."
① Regarding the training data to be created, the content of the AI that is the subject of machine learning or the training data related to machine learning is sufficiently specified in the claim.
② The means for solving the problem of the invention described in the detailed description of the invention are reflected.
Additional case study 7: Machine learning device for screw tightening quality (Annex A 1. Case 54）
The following ① and ② are examples that satisfy the support requirements for inventions related to "correlation between multiple types of data included in training data."
① The claim specifies the input/output relationship of each data included in the training data.
② It is common general technical knowledge at the time of filing that there is a certain correlation between each data.
Additional cases 8 to 9: Examples of determining eligibility requirements for inventions
Additional example 8: Teacher data and image generation method for teacher data (Annex A 3. Case 5）
Regarding the eligibility of inventions related to "teacher data" in "machine learning",
The "teacher data" according to claim 1 does not have technical characteristics in the information presentation means or presentation method, but has characteristics only in the content of the information presented, and is mainly used for information presentation. Because it is intended for the purpose of the invention, it is a mere presentation of information and does not satisfy the requirements for cooperation between software and hardware, so it is said that it does not meet the requirements for eligibility as an invention.
The "method for generating images for teaching data" according to claim 2 is a case in which the eligibility of the invention is recognized as it satisfies the requirements for cooperation between software and hardware.
Additional example 9: A trained model for analyzing the reputation of accommodation facilities (concerning a trained model configured as a parameter set) (Annex B Chapter 1 3.2 Case 2-14'）
Because a "trained model" does not specify computer processing and is merely a parameter set with characteristics only in the content of information obtained through machine learning, it is a mere presentation of information and is not a "program" or This is a case in which it is determined that the "conception based on the software perspective" (Examination Handbook Annex B Chapter 1 2.1.1.2) does not apply and does not meet the requirements for eligibility as an invention because it cannot be said to be equivalent to that. .
Additional example 10: Example of determining clarity requirements
Additional example 10: Learned model to output work details to be performed for abnormalities (Annex A 1. Case 55）
The following ① and ② are examples of cases where the invention of a “trained model” violates the clarity requirement.
① Since the trained model includes a "program" which is an "invention of a product", it cannot be said to be an "invention of a method"; however,
・There is no mention of making the computer realize multiple functions.
- It is not required that the trained model be a program.
Therefore, it cannot be said that the invention is a "program", so it cannot be determined whether it is an "invention of a product" or "invention of a method", and the category to which the invention belongs is unclear.
② Cases where the invention is unclear because a "trained model" which is a program is described as having a "means" even though the "program" itself does not function as a "means"