📌 For those reading this article: If you are drafting your statement using generated AI such as ChatGPT, please also use the AI draft x patent attorney check page.
With the rapid development of AI technology, the number of AI-related patent applications is also increasing. However, when determining the patentability of AI-related inventions, there are many considerations that are different from those in conventional technical fields, and specialized knowledge is required to formulate a filing strategy. In this article, we will explain the examination standards and practical points to keep in mind for AI-related patents, with reference to the ``Cases related to AI-related technologies'' in the examination handbook published by the Japan Patent Office.
The Patent/Utility Model Examination Handbook contains 25 cases to illustrate how the examination standards are applied to patent applications related to AI-related technologies. These are mainly organized from the following three perspectives.
These cases cover various aspects of AI-related technologies and serve as important guidelines for understanding the criteria for examining AI-related patents.
In inventions of AI-related technologies, especially inventions that apply AI, correlation between multiple pieces of data included in training data is particularly important. The examination standards require that it be recognized that a certain relationship, such as a correlation, exists between multiple types of data based on the description of the detailed description of the invention, or that the existence of a correlation, etc. can be inferred in light of common general technical knowledge.
For example, in Case 46, "Sugar content estimation system," it is considered a violation of the enablement requirement because it cannot be inferred as common technical knowledge that there is a correlation between a person's face image and the sugar content of vegetables grown by that person. On the other hand, in Case 47, "Business planning support device," it is determined that the feasibility requirement is met because it can be assumed as common technical knowledge that there is a correlation between advertising activity data on the web and sales numbers.
When inventing a product that is estimated to have a certain function using AI, it is important to consider whether the results of AI estimation can replace the evaluation of the actually manufactured product. For example, in Case 51, "Anaerobic adhesive composition," the predictive accuracy of the trained model's predicted values has not been verified, and there is no evaluation data for the actually manufactured product, so it is considered a violation of the enablement requirement.
If a claim ends with a term such as "trained model" and it is unclear whether it means "program" or not, the category may be unclear and a violation of the clarity requirement may occur. Case 55 shows the difference between the description format that is acceptable as a program and the description format that is not acceptable for "a trained model for outputting the work details to be performed in response to an abnormality."
When determining the inventive step of AI-related technology, the following viewpoints are important.
The issues and effects of applying AI to conventional technology are important factors in determining whether the mere application of AI constitutes an inventive step. For example, in Case 33, "Cancer Level Calculation Device," the inventive step of using AI to systemize cancer level calculations performed by doctors is denied, as it is merely the exercise of the ordinary creative ability of a person skilled in the art.
Similarly, in claim 1 of Case 40 "Laser processing equipment," the inventive step is denied because replacing the work performed by humans with a machine-learned model is conventional technology.
The effect of changing teaching data is also an important element in determining inventive step. In Case 34, "Hydroelectric power generation estimation system," an invention that simply uses a neural network (Claim 1) is denied an inventive step, while an invention that produces a remarkable effect by adding the temperature of the upstream area as input data (Claim 2) is recognized as an inventive step.
The preprocessing of training data also affects the judgment of inventive step. Case 36, ``Dementia Level Estimation Device,'' has been recognized as being effective in achieving highly accurate dementia level estimation by associating the type of question asked by the questioner with the content of the answers given by the respondent as training data, and is recognized as an inventive step.
The case added in March 2020 also shows an inventive step determination regarding the application of generative AI. In Case 37, "Answer automatic generation device for customer center," the inventive step is denied because inputting a question into a large-scale language model and automatically generating an answer is a conventional technology.
On the other hand, in claim 2 of Case 38, "Method for generating prompt sentences for input to a large-scale language model," an inventive step is recognized in the invention of extracting multiple keywords suitable as reference information from multiple related sentences and generating a prompt within the limited number of characters.
The following points are important when determining whether AI-related technology is eligible for invention.
Mere data or parameter sets are judged to be "mere presentation of information" and do not fall under the category of "invention." For example, in Case 5, “Teacher data and image generation method for teacher data,” the teacher data itself does not fall under the category of “invention,” but the method for generating the teacher data is determined to fall under the category of “invention.”
An important criterion is whether information processing by software is concretely realized using hardware resources. For example, in Case 2-14, "A trained model for analyzing the reputation of accommodation facilities," it falls under an "invention" if it is described as a function that causes a computer to perform a specific calculation, but it is determined that it does not fall under an "invention" if it is simply configured as a parameter set.
When applying for AI-related patents, it is important to keep the following points in mind.
For AI-related inventions, it is necessary to specifically state in the detailed description of the invention that there is a correlation between input data and output data, or to show that it can be inferred from common technical knowledge. If the correlation is not clear, it would be desirable to enrich it with experimental data and theoretical explanations.
In order to show that AI estimation results can replace actual evaluation, it is important to include verification results of the prediction accuracy of the trained model and specific examples of actual manufacturing and evaluation. In particular, in the fields of chemistry and materials, it is often necessary to include not only AI predictions but also actual verification results.
When requesting a trained model, it is important to clearly state its nature as a "program." Specifically, it should be written in a format such as ``a trained model that makes the computer function ~'' to avoid misunderstanding that it is just a parameter set.
Mere systemization of AI application is difficult to recognize as an inventive step, so the following ideas should be considered.
In order to ensure patentability, it should be clearly shown that software and hardware resources work together to realize specific information processing. It is effective to aim to obtain rights not just for data structures or parameter sets, but for specific information processing methods and devices that use them.
When applying for an AI-related patent, there are many points to consider that are different from ordinary patent applications. In particular, attention must be paid to the correlation between AI input data and output data, effectiveness verification of trained models, claim description format, etc. Furthermore, since it is difficult to recognize inventive step in the mere application of AI, it is important to clearly explain the unique ingenuity used to solve a specific technical problem.
Our firm has a wealth of experience and expertise in filing patent applications for AI-related technologies, and provides support to obtain the rights for your company's AI technology in the best possible manner. If you are having trouble determining patentability or are considering building a stronger patent portfolio, please feel free to contact us.
🤖 For those who created a draft of a patent specification with ChatGPT
You are welcome to bring in your AI drafts. A patent attorney who specializes in IT and AI patents will rewrite the generated AI draft into a "strong title deed that protects your business."
Leave the final check of the statement quality to a professional with over 15 years of experience in intellectual property practice.
AUTHOR
Takefumi SUGIURA （杉浦 健文）
EVORIX Intellectual Property Law Firm Managing Patent Attorney
Supports clients across IT, manufacturing, startups, fashion, and medical industries, covering patent, trademark, design, and copyright filings through trials and infringement litigation. Specialized in IP strategy for AI, IoT, Web3, and FinTech. Member of the Japan Patent Attorneys Association (JPAA), Asian Patent Attorneys Association (APAA), and Japan Trademark Association (JTA).