2024.08.16
Addition of Case Examples on AI-related Technologies to "Examination Handbook" of Japan Patent Office
PATENT
Addition of Case Examples on AI-related Technologies to "Examination Handbook" of Japan Patent Office
1. Introduction
In association with development of AI-related technologies in various technical fields, the Japan Patent Office has published case examples on AI-related technologies with the aim of providing clear guidance on determination on inventive step, description requirement, etc.
The first five case examples were published in March 2017, and ten case examples were added in January 2019. Now, ten case examples were further added in March 2024. These case examples are shown in the Examination Handbook referred to by examiners at the Japan Patent Office during examination.
Among the recently added ten case examples, this article explains four case examples related to determination on the inventive step (Article 29(2) of the Patent Act), and also explains important points to be careful of when filing an application of AI-related technologies in Japan in light of those case examples.
2. Overview of Recently Added Case Examples Related to Determination on Inventive Step
(1) Case 37
① Title of Invention: Automatic Response Generator for Customer Service Centers
② Determination: inventive step is not satisfied
③ What is claimed:
(2) Case 38
① Title of Invention: Method of Generating Texts for Prompt for Input into Large Language Model
② Determination: claim 1 lacks inventive step, and claim 2 has an inventive step
③ What is claimed:
④ Overview of Case:
⑤ Key Points on Determination of Inventive Step:
① Title of Invention: Method of Learning Trained Model for Use in Radiographic Image Brightness Adjustment
② Determination: inventive step is satisfied
③ What is claimed:
④ Overview of Case:
(4) Case 40
① Title of Invention: Laser Processing Device
② Determination: claim 1 lacks inventive step, and claim 2 has an inventive step
③ What is claimed:
④ Overview of Case:
⑤ Key Points on Determination of Inventive Step:
3. Important Points on Filing Application of AI-related Technologies in Japan
In view of recently added Cases 37-40 related to inventive step, it is likely that an invention in which a process usually performed by humans is replaced to be performed by an AI like in “Case 37” is not regarded to have an inventive step.
On the other hand, in order for an AI-related invention to be recognized to have an inventive step as in “Case 38,” “Case 39,” and “Case 40,” it is required that characteristic features of the AI such as processing steps, learning method, data configuration used, etc. were not publicly known or common general knowledge as of the filing, and that an advantageous effect over a cited reference is attained.
Further, also in the case examples previously added in 2019 (Cases 33-36), an invention which merely uses an AI for estimating output data from input data, an invention which merely systematizes human tasks by using an AI, and an invention in which a change of training data used in learning is a combination of known data and no significant effect is recognized are judged to lack inventive step.
Thus, in order to acquire patent right for AI-related technologies in Japan, it should be noted that an invention is required to have a novel feature, such as AI processing steps, learning method, data configuration used, etc., and to attain an advantageous effect over publicly-known techniques by the feature, not to merely use an AI to perform a publicly-known process or merely use publicly-known data in an AI.
In association with development of AI-related technologies in various technical fields, the Japan Patent Office has published case examples on AI-related technologies with the aim of providing clear guidance on determination on inventive step, description requirement, etc.
The first five case examples were published in March 2017, and ten case examples were added in January 2019. Now, ten case examples were further added in March 2024. These case examples are shown in the Examination Handbook referred to by examiners at the Japan Patent Office during examination.
Among the recently added ten case examples, this article explains four case examples related to determination on the inventive step (Article 29(2) of the Patent Act), and also explains important points to be careful of when filing an application of AI-related technologies in Japan in light of those case examples.
2. Overview of Recently Added Case Examples Related to Determination on Inventive Step
(1) Case 37
① Title of Invention: Automatic Response Generator for Customer Service Centers
② Determination: inventive step is not satisfied
③ What is claimed:
[Claim 1]
An automatic response generator for a customer service center for receiving a question text of an inquiry about a financial product from an inquirer and automatically generating a response text to the question text, wherein the response text is automatically generated by inputting the question text into a large language model.
④ Overview of Case:An automatic response generator for a customer service center for receiving a question text of an inquiry about a financial product from an inquirer and automatically generating a response text to the question text, wherein the response text is automatically generated by inputting the question text into a large language model.
·According to the invention of claim 1 of the subject application, when responding to inquiries about financial products, generation of a response text to a question text of an inquiry about a financial product from an inquirer is performed by an automatic response generator for a customer service center which uses a large language model, in replacement of a person in charge of the customer service center.
·Cited invention 1 relates to a method of preparing a response text in which an operator performs preparation of a response text to a question text of an inquiry in the same field.
·In many business fields including customer services, it is an obvious task ordinarily considered by one skilled in the art to improve efficiency by systematizing and automating human tasks with a computer. Further, in the technical field of information processing, “replacing human decisions by machine decisions using a trained machine learning model,” which is the solution to the task, is a common technique. Thus, it would have been easily conceived by one skilled in the art to apply the common technique to Cited Invention 1 to make “an automatic response generator for a customer service center for automatically generating a response text, wherein the response text is automatically generated by inputting a question text of an inquiry into a large language model.”
·Cited invention 1 relates to a method of preparing a response text in which an operator performs preparation of a response text to a question text of an inquiry in the same field.
·In many business fields including customer services, it is an obvious task ordinarily considered by one skilled in the art to improve efficiency by systematizing and automating human tasks with a computer. Further, in the technical field of information processing, “replacing human decisions by machine decisions using a trained machine learning model,” which is the solution to the task, is a common technique. Thus, it would have been easily conceived by one skilled in the art to apply the common technique to Cited Invention 1 to make “an automatic response generator for a customer service center for automatically generating a response text, wherein the response text is automatically generated by inputting a question text of an inquiry into a large language model.”
(2) Case 38
① Title of Invention: Method of Generating Texts for Prompt for Input into Large Language Model
② Determination: claim 1 lacks inventive step, and claim 2 has an inventive step
③ What is claimed:
[Claim 1]
A method of generating a text for a prompt that is generated by a computer for input into a large language model by adding reference information to an inputted question text,
wherein the large language model has a character limit that is a maximum number of characters in a prompt that can be inputted, and when a prompt containing a question text is inputted, the large language model outputs a response text related to the question text, and
wherein the computer executes:
an additional text generation step in which an additional text related to the question text is generated based on the inputted question text so that a total number of characters including the number of characters of the question text and that of the additional text is equal to or less than the character limit; and
a prompt generation step in which the prompt is generated by adding the additional text generated in the additional text generation step to the inputted question text as reference information.
[Claim 2]
The method of claim 1, wherein in the additional text generation step a plurality of relevant texts related to the question text are obtained based on the inputted question text, a plurality of keywords suitable as the reference information are extracted from the obtained plurality of relevant texts, and the additional text is generated using the plurality of keywords so that the total number of characters does not exceed the character limit.
A method of generating a text for a prompt that is generated by a computer for input into a large language model by adding reference information to an inputted question text,
wherein the large language model has a character limit that is a maximum number of characters in a prompt that can be inputted, and when a prompt containing a question text is inputted, the large language model outputs a response text related to the question text, and
wherein the computer executes:
an additional text generation step in which an additional text related to the question text is generated based on the inputted question text so that a total number of characters including the number of characters of the question text and that of the additional text is equal to or less than the character limit; and
a prompt generation step in which the prompt is generated by adding the additional text generated in the additional text generation step to the inputted question text as reference information.
[Claim 2]
The method of claim 1, wherein in the additional text generation step a plurality of relevant texts related to the question text are obtained based on the inputted question text, a plurality of keywords suitable as the reference information are extracted from the obtained plurality of relevant texts, and the additional text is generated using the plurality of keywords so that the total number of characters does not exceed the character limit.
④ Overview of Case:
·The invention of claim 1 of the subject application relates to a method of generating a text for a prompt that is generated by a computer for input into a large language model, in which an additional text related to a question text is generated as reference information and added to the question text.
·Here, the large language model has the character limit, and the additional text is generated so that the total number of characters including the number of characters of the question text and that of the additional text is equal to or less than the character limit.
·Regarding the invention of claim 2 of the subject application, a concrete method of the additional text generation step is specified, and one skilled in the art would be able to understand effects of the invention from the Detailed Description of the Disclosure.
·According to Cited Invention 1, a prompt for input into a large language model is generated by generating an additional text related to a question text as reference information and adding it to the question text. However, Cited Invention 1 does not have a configuration in which the large language model has a character limit or the additional text is generated so that the number of characters of the prompt is equal to or less than the character limit.
·Here, in the technical field of language processing, it is an obvious task ordinarily considered by one skilled in the art to avoid an excessive information processing amount. Further, as a solution to the task, it is a well-known technique as of the filing to set a character limit, which is a maximum number of inputtable characters in a text, and if the text exceeds the character limit, discard the excess part to generate a text to be actually inputted so that the number of characters thereof is equal to or less than the character limit.
·The invention of claim 1 of the subject application lacks inventive step in view of Cited Invention1 and the well-known technique, whereas the invention of claim 2 of the subject application has an inventive step.
·Here, the large language model has the character limit, and the additional text is generated so that the total number of characters including the number of characters of the question text and that of the additional text is equal to or less than the character limit.
·Regarding the invention of claim 2 of the subject application, a concrete method of the additional text generation step is specified, and one skilled in the art would be able to understand effects of the invention from the Detailed Description of the Disclosure.
·According to Cited Invention 1, a prompt for input into a large language model is generated by generating an additional text related to a question text as reference information and adding it to the question text. However, Cited Invention 1 does not have a configuration in which the large language model has a character limit or the additional text is generated so that the number of characters of the prompt is equal to or less than the character limit.
·Here, in the technical field of language processing, it is an obvious task ordinarily considered by one skilled in the art to avoid an excessive information processing amount. Further, as a solution to the task, it is a well-known technique as of the filing to set a character limit, which is a maximum number of inputtable characters in a text, and if the text exceeds the character limit, discard the excess part to generate a text to be actually inputted so that the number of characters thereof is equal to or less than the character limit.
·The invention of claim 1 of the subject application lacks inventive step in view of Cited Invention1 and the well-known technique, whereas the invention of claim 2 of the subject application has an inventive step.
⑤ Key Points on Determination of Inventive Step:
[Difference from Cited Invention 1]
While in the additional text generation step of the invention of claim 2 the plurality of relevant texts related to the question text are obtained based on the inputted question text, the plurality of keywords suitable as the reference information are extracted from the obtained plurality of relevant texts, and the additional text is generated using the plurality of keywords so that the total number of characters does not exceed the character limit, the additional text generation step in Cited Invention 1 is not specified as such.
[Reason]
Although the invention of claim 2 has a configuration related to the above difference, no prior art disclosing such a configuration has been found, and further, the configuration was not common general knowledge as of the filing. In addition, according to the invention of claim 2, a prompt including the additional text highly relevant to the question text and suitable as the reference information can be generated within the predetermined character limit so that a more reliable and appropriate response text is obtained, which is an advantageous effect over Cited Invention 1, due to such a configuration. The above configuration cannot be regarded as design variation which may be made in applying the well-known technique to Cited Invention 1.
Therefore, the invention of claim 2 of the subject application has an inventive step.
(3) Case 39While in the additional text generation step of the invention of claim 2 the plurality of relevant texts related to the question text are obtained based on the inputted question text, the plurality of keywords suitable as the reference information are extracted from the obtained plurality of relevant texts, and the additional text is generated using the plurality of keywords so that the total number of characters does not exceed the character limit, the additional text generation step in Cited Invention 1 is not specified as such.
[Reason]
Although the invention of claim 2 has a configuration related to the above difference, no prior art disclosing such a configuration has been found, and further, the configuration was not common general knowledge as of the filing. In addition, according to the invention of claim 2, a prompt including the additional text highly relevant to the question text and suitable as the reference information can be generated within the predetermined character limit so that a more reliable and appropriate response text is obtained, which is an advantageous effect over Cited Invention 1, due to such a configuration. The above configuration cannot be regarded as design variation which may be made in applying the well-known technique to Cited Invention 1.
Therefore, the invention of claim 2 of the subject application has an inventive step.
① Title of Invention: Method of Learning Trained Model for Use in Radiographic Image Brightness Adjustment
② Determination: inventive step is satisfied
③ What is claimed:
[Claim 1]
A method of learning a trained model by a machine learning process, in which a radiographic image of a human body is inputted to output a brightness adjustment parameter of the radiographic image, the method comprising:
a step of obtaining training data including a learning radiographic image and a training image associated therewith in which the learning radiographic image is adjusted in brightness;
a step of inputting the learning radiographic image included in the training data to output the brightness adjustment parameter of the learning radiographic image by the learning model during learning;
a step of obtaining a value for a loss function using the brightness adjustment parameter outputted by the learning model; and
a step of optimizing the learning model so that the value for the loss function is reduced,
wherein the steps are executed by a computer,
wherein the loss function is a function based on an error between pixel values of the training image and pixel values of a brightness-adjusted image in which the learning radiographic image is adjusted in brightness based on the brightness adjustment parameter outputted by the learning model; and
wherein the loss function is configured to bias the learning to suppress occurrence of pixel value saturation by integrating a predetermined weight for the error in pixels where the pixel value saturation of the brightness-adjusted image occurs so that the value for the loss function is estimated to be relatively large.
A method of learning a trained model by a machine learning process, in which a radiographic image of a human body is inputted to output a brightness adjustment parameter of the radiographic image, the method comprising:
a step of obtaining training data including a learning radiographic image and a training image associated therewith in which the learning radiographic image is adjusted in brightness;
a step of inputting the learning radiographic image included in the training data to output the brightness adjustment parameter of the learning radiographic image by the learning model during learning;
a step of obtaining a value for a loss function using the brightness adjustment parameter outputted by the learning model; and
a step of optimizing the learning model so that the value for the loss function is reduced,
wherein the steps are executed by a computer,
wherein the loss function is a function based on an error between pixel values of the training image and pixel values of a brightness-adjusted image in which the learning radiographic image is adjusted in brightness based on the brightness adjustment parameter outputted by the learning model; and
wherein the loss function is configured to bias the learning to suppress occurrence of pixel value saturation by integrating a predetermined weight for the error in pixels where the pixel value saturation of the brightness-adjusted image occurs so that the value for the loss function is estimated to be relatively large.
④ Overview of Case:
·The invention of claim 1 of the subject application relates to a method of learning a trained model by a machine learning process, in which a radiographic image of a human body is inputted to output a brightness adjustment parameter of the radiographic image for obtaining an image with brightness suitable for diagnosis.
·The learning radiographic image is inputted into the learning model during learning, the brightness-adjusted image is generated in which the learning radiographic image is adjusted in brightness based on the outputted brightness adjustment parameter, the loss function based on the error between the pixel values of the training image included in the training data and the pixel values of the brightness-adjusted image is obtained, and the learning model is optimized so that the value for the loss function is reduced, thereby learning the learning model by the machine learning process.
·Here, in the invention of claim 1 of the subject application, the loss function is configured to bias the learning to suppress occurrence of pixel value saturation by integrating the predetermined weight for the error in pixels where the pixel value saturation of the brightness-adjusted image occurs so that the value for the loss function is estimated to be relatively large (Difference from Cited Invention 1), but no prior art disclosing such a configuration has been found, and further, the configuration was not common general knowledge as of the filing.
·In addition, “an effect of being able to learn a trained model used for adjusting the brightness of a radiographic image which can prevent pixel value saturation and improve visibility by biasing the learning to suppress the occurrence of pixel saturation” provided by the invention of claim 1 of the subject application is an advantageous effect over Cited Invention 1, and is difficult to predict from Cited Invention 1, as Cited Invention 1 focuses merely on a general issue of improving the accuracy of the machine learning process. Therefore, the subject invention has an inventive step.
⑤ Key Points on Determination of Inventive Step:·The learning radiographic image is inputted into the learning model during learning, the brightness-adjusted image is generated in which the learning radiographic image is adjusted in brightness based on the outputted brightness adjustment parameter, the loss function based on the error between the pixel values of the training image included in the training data and the pixel values of the brightness-adjusted image is obtained, and the learning model is optimized so that the value for the loss function is reduced, thereby learning the learning model by the machine learning process.
·Here, in the invention of claim 1 of the subject application, the loss function is configured to bias the learning to suppress occurrence of pixel value saturation by integrating the predetermined weight for the error in pixels where the pixel value saturation of the brightness-adjusted image occurs so that the value for the loss function is estimated to be relatively large (Difference from Cited Invention 1), but no prior art disclosing such a configuration has been found, and further, the configuration was not common general knowledge as of the filing.
·In addition, “an effect of being able to learn a trained model used for adjusting the brightness of a radiographic image which can prevent pixel value saturation and improve visibility by biasing the learning to suppress the occurrence of pixel saturation” provided by the invention of claim 1 of the subject application is an advantageous effect over Cited Invention 1, and is difficult to predict from Cited Invention 1, as Cited Invention 1 focuses merely on a general issue of improving the accuracy of the machine learning process. Therefore, the subject invention has an inventive step.
[Difference from Cited Invention 1]
While in the invention of claim 1 the loss function is configured to bias the learning to suppress the occurrence of pixel value saturation by integrating the predetermined weight for the error between the pixel values of the training image and the pixel values of the brightness-adjusted image in pixels where the pixel value saturation of the brightness-adjusted image occurs so that the value for the loss function is estimated to be relatively large, Cited Invention 1 does not have such a configuration.
[Reason]
In performing a machine learning process using a loss function, it is common practice and just adoption of design variation or design choice for one skilled in the art to change the configuration of the loss function to improve the estimation accuracy of a trained model generated by the machine learning process.
However, regarding a method of learning a trained model used for image processing of radiographic images, no prior art disclosing a specific configuration related to the difference has been found, and further, such a loss function was not common general knowledge as of the filing.
In addition, according to the configuration related to the above difference, “the effect of being able to learn a trained model used for adjusting the brightness of a radiographic image which can prevent pixel value saturation and improve visibility by biasing the learning to suppress the occurrence of pixel saturation” is provided, which is an advantageous effect over Cited Invention 1 and is difficult to predict from Cited Invention 1, as Cited Invention 1 focuses merely on a general issue of improving the accuracy of the machine learning process.
Synthetically taking the above circumstances into consideration, it cannot be concluded that the configuration related to the difference would have been easily conceived by one skilled in the art from Cited Invention 1.
Therefore, the invention of claim 1 of the subject application has an inventive step.
While in the invention of claim 1 the loss function is configured to bias the learning to suppress the occurrence of pixel value saturation by integrating the predetermined weight for the error between the pixel values of the training image and the pixel values of the brightness-adjusted image in pixels where the pixel value saturation of the brightness-adjusted image occurs so that the value for the loss function is estimated to be relatively large, Cited Invention 1 does not have such a configuration.
[Reason]
In performing a machine learning process using a loss function, it is common practice and just adoption of design variation or design choice for one skilled in the art to change the configuration of the loss function to improve the estimation accuracy of a trained model generated by the machine learning process.
However, regarding a method of learning a trained model used for image processing of radiographic images, no prior art disclosing a specific configuration related to the difference has been found, and further, such a loss function was not common general knowledge as of the filing.
In addition, according to the configuration related to the above difference, “the effect of being able to learn a trained model used for adjusting the brightness of a radiographic image which can prevent pixel value saturation and improve visibility by biasing the learning to suppress the occurrence of pixel saturation” is provided, which is an advantageous effect over Cited Invention 1 and is difficult to predict from Cited Invention 1, as Cited Invention 1 focuses merely on a general issue of improving the accuracy of the machine learning process.
Synthetically taking the above circumstances into consideration, it cannot be concluded that the configuration related to the difference would have been easily conceived by one skilled in the art from Cited Invention 1.
Therefore, the invention of claim 1 of the subject application has an inventive step.
(4) Case 40
① Title of Invention: Laser Processing Device
② Determination: claim 1 lacks inventive step, and claim 2 has an inventive step
③ What is claimed:
[Claim 1]
A laser processing device for welding that irradiates a laser beam onto a workpiece, comprising:
a control unit that controls the laser processing device based on a plurality of processing parameters related to laser processing;
a light intensity detection unit that detects a light intensity in a predetermined wavelength band of reflected light generated from the workpiece by the irradiation of the laser beam as a light intensity signal;
an average value extraction unit that extracts an average value obtained from a time-series signal of the light intensity signal;
a machine learning unit that performs a machine learning process of a learning model in which the average value is used as input data, an adjustment amount of the plurality of processing parameters is used as output data, and past actual values of the input data and the output data are used as training data; and
a processing parameter adjustment unit that inputs the input data to the trained model obtained by the machine learning process in the machine learning unit, outputs the adjustment amount of the plurality of processing parameters as the output data, and inputs the adjustment amount of the plurality of processing parameters to the control unit.
[Claim 2]
The laser processing device of claim 1 further comprising an accumulated usage time storage unit that stores accumulated usage time of a laser oscillator, wherein the input data further includes the accumulated usage time of the laser oscillator.
A laser processing device for welding that irradiates a laser beam onto a workpiece, comprising:
a control unit that controls the laser processing device based on a plurality of processing parameters related to laser processing;
a light intensity detection unit that detects a light intensity in a predetermined wavelength band of reflected light generated from the workpiece by the irradiation of the laser beam as a light intensity signal;
an average value extraction unit that extracts an average value obtained from a time-series signal of the light intensity signal;
a machine learning unit that performs a machine learning process of a learning model in which the average value is used as input data, an adjustment amount of the plurality of processing parameters is used as output data, and past actual values of the input data and the output data are used as training data; and
a processing parameter adjustment unit that inputs the input data to the trained model obtained by the machine learning process in the machine learning unit, outputs the adjustment amount of the plurality of processing parameters as the output data, and inputs the adjustment amount of the plurality of processing parameters to the control unit.
[Claim 2]
The laser processing device of claim 1 further comprising an accumulated usage time storage unit that stores accumulated usage time of a laser oscillator, wherein the input data further includes the accumulated usage time of the laser oscillator.
④ Overview of Case:
·According to claim 1 of the subject application, a process for adjusting a plurality of processing parameters of a laser processing device which was performed by an operator in the field of machining is replaced by arithmetic processing using a trained model.
·According to Cited Invention 1, a plurality of processing parameters of a laser processing device are adjusted by an operator.
·In many technical fields including general machining, it is an obvious task ordinarily considered by one skilled in the art to improve efficiency by systematizing and automating human tasks with a computer. Further, in the technical field of information processing, it is common to replace human decisions by machine decisions using a trained machine learning model in order to improve the efficiency of human tasks. Thus, it is judged that it would have been easily conceived by one skilled in the art to apply the common technique to Cited Invention 1 to “perform adjustment of a plurality of processing parameters by arithmetic processing using a trained model.”
·On the other hand, regarding claim 2 of the subject application, it is recognized that a new feature (training data used for the learning) added to the AI-based systematization of operator’s tasks provides an advantageous effect over Cited Invention 1. Therefore, claim 2 has an inventive step.
·According to Cited Invention 1, a plurality of processing parameters of a laser processing device are adjusted by an operator.
·In many technical fields including general machining, it is an obvious task ordinarily considered by one skilled in the art to improve efficiency by systematizing and automating human tasks with a computer. Further, in the technical field of information processing, it is common to replace human decisions by machine decisions using a trained machine learning model in order to improve the efficiency of human tasks. Thus, it is judged that it would have been easily conceived by one skilled in the art to apply the common technique to Cited Invention 1 to “perform adjustment of a plurality of processing parameters by arithmetic processing using a trained model.”
·On the other hand, regarding claim 2 of the subject application, it is recognized that a new feature (training data used for the learning) added to the AI-based systematization of operator’s tasks provides an advantageous effect over Cited Invention 1. Therefore, claim 2 has an inventive step.
⑤ Key Points on Determination of Inventive Step:
[Difference from Cited Invention 1]
In the invention of the laser processing device according to claim 2, the laser processing device includes the accumulated usage time storage unit that stores accumulated usage time of the laser oscillator, and the input data includes the accumulated usage time of the laser oscillator. On the other hand, Cited Invention 1 relates to a laser processing method, and a laser processing device does not include the accumulated usage time storage unit that stores accumulated usage time of a laser oscillator, and determination on an adjustment amount of a plurality of processing parameters is not based on the accumulated usage time of the laser oscillator.
[Reason]
According to the invention of claim 2, the laser processing device includes the accumulated usage time storage unit that stores accumulated usage time of the laser oscillator, and the input data includes the accumulated usage time of the laser oscillator, but no prior art disclosing such a configuration has been found, and further, such a configuration was not common general knowledge as of the filing.
In addition, in the technical field of laser processing devices, no prior art has been found that discloses a laser processing device including an accumulated usage time storage unit which stores accumulated usage time of a laser oscillator, in which the accumulated usage time of the laser oscillator is taken into consideration when an operator adjusts processing parameters. Further, such a configuration was not common general knowledge as of the filing.
Moreover, according to the invention of claim 2, the configuration related to the difference provides “an effect of being able to dramatically improve the estimation accuracy of the adjustment amount of a plurality of processing parameters,” which is an advantageous effect over Cited Invention 1 and cannot be regarded as design variation which may be adopted in applying the common technique to Cited Invention 1.
Therefore, the invention of claim 2 of the subject application has an inventive step.
In the invention of the laser processing device according to claim 2, the laser processing device includes the accumulated usage time storage unit that stores accumulated usage time of the laser oscillator, and the input data includes the accumulated usage time of the laser oscillator. On the other hand, Cited Invention 1 relates to a laser processing method, and a laser processing device does not include the accumulated usage time storage unit that stores accumulated usage time of a laser oscillator, and determination on an adjustment amount of a plurality of processing parameters is not based on the accumulated usage time of the laser oscillator.
[Reason]
According to the invention of claim 2, the laser processing device includes the accumulated usage time storage unit that stores accumulated usage time of the laser oscillator, and the input data includes the accumulated usage time of the laser oscillator, but no prior art disclosing such a configuration has been found, and further, such a configuration was not common general knowledge as of the filing.
In addition, in the technical field of laser processing devices, no prior art has been found that discloses a laser processing device including an accumulated usage time storage unit which stores accumulated usage time of a laser oscillator, in which the accumulated usage time of the laser oscillator is taken into consideration when an operator adjusts processing parameters. Further, such a configuration was not common general knowledge as of the filing.
Moreover, according to the invention of claim 2, the configuration related to the difference provides “an effect of being able to dramatically improve the estimation accuracy of the adjustment amount of a plurality of processing parameters,” which is an advantageous effect over Cited Invention 1 and cannot be regarded as design variation which may be adopted in applying the common technique to Cited Invention 1.
Therefore, the invention of claim 2 of the subject application has an inventive step.
3. Important Points on Filing Application of AI-related Technologies in Japan
In view of recently added Cases 37-40 related to inventive step, it is likely that an invention in which a process usually performed by humans is replaced to be performed by an AI like in “Case 37” is not regarded to have an inventive step.
On the other hand, in order for an AI-related invention to be recognized to have an inventive step as in “Case 38,” “Case 39,” and “Case 40,” it is required that characteristic features of the AI such as processing steps, learning method, data configuration used, etc. were not publicly known or common general knowledge as of the filing, and that an advantageous effect over a cited reference is attained.
Further, also in the case examples previously added in 2019 (Cases 33-36), an invention which merely uses an AI for estimating output data from input data, an invention which merely systematizes human tasks by using an AI, and an invention in which a change of training data used in learning is a combination of known data and no significant effect is recognized are judged to lack inventive step.
Thus, in order to acquire patent right for AI-related technologies in Japan, it should be noted that an invention is required to have a novel feature, such as AI processing steps, learning method, data configuration used, etc., and to attain an advantageous effect over publicly-known techniques by the feature, not to merely use an AI to perform a publicly-known process or merely use publicly-known data in an AI.
END OF ARTICLE
