Artificial Intelligence-related invention in Japan and its disclosure with a focus on human intervention
p. 225-247
Texte intégral
I. Introduction
1In general, patent law is aimed at protecting and using inventions, and since its framework is continuously closely related to innovation, it has always been flexible to meet the needs of the times. At the same time, it may present us with novel issues concerning legal protection. For instance, advances in information and communication technology have brought about major changes in various technical fields. In particular, rapid advances in artificial intelligence (“AI”) technology are bringing about massive changes in modern society1. In Japan, “Society 5.0,” which is a system that highly integrates cyberspace (virtual space) and physical space (real space), was proposed in the 5th Science and Technology Basic Plan in 20162. In the modern information society, value has been created by human beings accessing and analysing information. AI has been analysing data so far, but the target of information transmission is humans, and ultimately humans make decisions and act. It was humans who changed the real world. Society 5.0 is a concept of a new human-centred future society realised by incorporating innovative technologies, such as AI, the Internet of things (“IoT”), robots, and Big Data, into all industries and societies. The Japan Patent Office (“JPO”) has also frequently published case examples since 2016 regarding the determination of AI inventions in order to improve the predictability of rights.
2As AI becomes increasingly important in the invention process, it is important to scrutinise current acts and case law and consider methods to deal with AI’s increasing importance in the future3. Under current law, inventions cannot be legally protected under the Patent Act unless they are created by a natural person in the first place. For example, the invention created by AI/DABUS, which was filed in recent years, has received a great deal of attention. International organisations, such as the World Intellectual Property Organisation (WIPO) and The International Association for the Protection of Intellectual Property (AIPPI), are cautiously debating whether to identify AI as an inventor4. AI learn from input data, potentially may “output” the level technical solutions like human-created inventions. On the other hand, the internal operation of AI is often characterised by non-disclosure, or at least is difficult to understand. How should the disclosure of an invention be considered to ensure that a person skilled in the art can carry out the claimed invention based on the disclosure of the patent specification? We will focus on the ingenuity of human beings engaged in AI inventions.
3In this chapter, in order to examine the handling of AI inventions, we will share the interpretation of the case examples for AI inventions in the JPO. First, we will explain the patent eligibility of AI inventions in Japan through the case examples in the JPO. Second, we will provide the theories of inventorship so far and a recent topic of AI inventions. Third, we will discuss the Japanese way of thinking about AI inventions in terms of the inventive step and a person skilled in the art. Ultimately, we also consider appropriate written description requirement through the JPO case examples and European Patent Office (EPO) decisions as well. We will consider some future issues.
II. AI world and Inventions
4AI has become a familiar term, but people may have different understandings of it. The definition is not always clear, but it can be said to be software capable of intelligent information processing like humans. AI can be classified in various ways and can be classified into specialised AI and general purpose AI according to the scope of task processing. This chapter mainly deals with issues related to specialised AI that is currently in practical use. The progress of AI technology is also profoundly related to the progress of IoT-related technology. From the perspective of the IoT, the JPO has created guidelines by defining the basic elements as follows:
- Acquire various data (Acquisition)
- Collect and manage data via the network (Management)
- Analyse and learn Big Data by using AI, etc. (Analysis・Learning)
- Utilise data to find new values and services (Utilisation)
5In (3), the AI learning system plays a major role. AI aims to mimic the structure of the neural circuits in the human brain. These neural networks are among the machine learning mechanisms that aim to realise the same learning ability functions that humans naturally perform. AI needs a program to analyse Big Data; therefore, data and programs are essential elements for AI development. In particular, since this digital information can be copied without deterioration, the incentive for AI development is impaired if others use it without permission for AI development work on which money and effort has been spent. Therefore, appropriate legal protection is required. Through protection under patent rights, it is possible to obtain a patent for an AI program. Furthermore, data structures for computer processing may also be evaluated as programmatic, and there is a possibility that the invention of a business method using AI can also be patented. The subject matter of AI inventions in this chapter shall be that if part or all of the invention is made by AI, or if the invention is composed of (new or improved) AI, the invention will include AI (hereafter, “AI inventions”). Next, we will consider the possibility of patent protection for AI inventions in Japan based on case examples that were released by the JPO.
III. Protection of AI inventions in Japan as “inventions”
6How should AI be evaluated under the Patent Act? In other words, how should AI be regarded as a patent subject5. The Japanese Patent Act, Article 2, paragraph (1) defines inventions, and Article 29(1) of the Patent Act provides that a patent shall be granted for an invention as defined in the Patent Act; an “invention” is clearly defined to mean a “highly advanced creation of technical ideas utilising the laws of nature.” “The laws of nature” refer to fundamental rules and principles that have physical, chemical or biological rules such as mental activities, mathematical laws, and man-made agreements. Accordingly, the common eligibility test applicable to inventions is whether the invention “utilises the laws of nature” as a whole and embodies “technical ideas.” The following is an example of the invention utilising the laws of nature:
- (i) Concretely performing control of an apparatus, or processing with respect to the control (e.g., engine control).
- (ii) Concretely performing information processing based on the technical properties of an object (e.g., image processing).
7Furthermore, AI inventions are also considered to be a type of computer-implemented invention, and thus the provisions in “Computer Software-related Inventions” in the Examination Handbook for the Patent and Utility Model (hereafter “Examination Handbook”) apply. The current JPO examination determines the eligibility of a software-related invention according to “whether information processing by the software is specifically implemented by using hardware resources.” More specifically, by examining “based on the statement of the claims, whether or not specific calculation or processing of information depending on the intended use is implemented by specific means or procedures on which software and hardware resources cooperate6.” This procedure is the same not only for software inventions, but also for AI development regarding “data having a structure” and “data structure,” which are similar to a program. Since AI inventions are expected to connect to many fields, the JPO publishes examination case examples. We will consider AI inventions with reference to the following examples.
A. AI inventions including programs and data structures
8The JPO conceptualises AI inventions about data structure by citing the following cases: “Data Structure of Dialogue Scenarios in a Voice Interactive System7” “[Claim 1] A data structure of dialogue scenarios utilised in a voice interactive system composed of a client’s device and a server, comprising: unit IDs that identify dialogue units constituting dialogue scenarios; messages including contents of utterances and information presented to users; a plurality of candidate answers in response to answers from users; information on communication mode; and a plurality of branch information mapped to each of the candidate answers and information on communication mode, wherein the branch information indicates the following dialogue unit which contains messages corresponding to the said candidate answers and whose data size corresponds to the said information on communication mode; wherein, the said data structure of dialogue scenarios is utilised for the following processing performed by the said client’s device:
- (1) outputting a message included in the current dialogue unit;
- (2) acquiring an answer from the user in response to the message;
- (3) specifying the said candidate answer based on the answer from the user;
- (4) selecting the branch information based on the candidate answer and information on communication mode; and
- (5) receiving from the server a following dialogue unit indicated by the selected branch information.”
Figure 1: Case2-13 Data Structure of Dialogue Scenarios in a Voice Interactive System

9According to the JPO casebook, the evaluation of the claim of the above invention is as follows: “It can be said that the data structure of Claim 1 enables information processing, that is for voice dialogues based on branch information included in dialogue units.“ (…).” “Voice dialogues in accordance with branch information included in dialogue units, are realised by concrete means or procedures, that is, information processing by a voice interactive system composed of the server and clients’ devices by means of collaboration between the computer software and hardware resources. The data structure is determined to establish an operating method for the specific information processing device in accordance with the purpose of use by means of the collaboration between the computer software and hardware resources. Therefore, as information processing prescribed by the data structure equivalent to the computer program is concretely realised utilising hardware resources, the data structure of Claim 1 is a creation of the technical idea utilising a law of nature and thus falls under [invention].”
10It was explained that information processing of the server and client device according to the purpose of voice dialogue meets the Japanese software invention requirements for collaboration between software and hardware devices. On the other hand, in terms of inventions concerning data, training data plays an important role in constructing machine learning optimisation. However, training data is an algorithm that is in the stage before some learning is performed. The training data in machine learning technology is the material of the learning process, and the reasoning by machine learning is derived from the trained model. Therefore, it is considered not to be subject to protection by the Patent Act.
11The concept of “data” is so diverse that it seems difficult to capture its precise meaning. It will be necessary to carefully consider whether it is appropriate to protect it with an exclusive patent right. Next, regarding the data structure, although it depends on the algorithm and is not a direct command to the computer, however, the optimised output can be equated with a programmatic mechanism. Accordingly, a unique technical configuration is constructed according to the purpose of utilisation. This case example can be interpreted as an “invention” of the Patent Act as equivalent to an existing software-related invention, as one can observe from the JPO case.
B. AI inventions including learning models
12The structure of the neural network is also subject to patents due to it being similar to programs that perform computer processing. Moreover, a learning model is a combination of a program and trained parameters, and is patentable if the idea constitutes the creation of a technical idea using the laws of nature. The JPO casebook conceptualises the “Trained Model for Analysing Reputations of Accommodations” as an example of trained models8. “[Claim 1] A trained model for causing a computer to function to output quantified values of reputations of accommodations based on text data on reputations of accommodations, wherein the model is comprises a first neural network and a second neural network connected in a such way that the second neural network receives output from the first neural network. The first neural network is comprises an input layer to intermediate layers of a feature extraction neural network in which the number of neurons of at least one intermediate layer is smaller than the number of neurons of the input layer, the number of neurons of the input layer and the number of the output layers are the same, and weights are trained in such a way that each value input to the input layer and each corresponding value output from output layer become equal. The weights of the second neural network are trained without changing the weights of the first neural network; and the model causes the computer function to perform a calculation based on the trained weights in the first and second neural networks in response to the appearance frequency of specific words obtained from the text data on reputations of accommodations input to the input layer of the first neural network and to output the quantified values of reputations of accommodations from the output layer of the second neural network.”
Figure 2 : Case 2-14 Trained Model for Analysing Reputations of Accommodations

13As for the learned model for analysing reputations of accommodations of Claim1. It shown as follows: “a computer to ‘function to perform a calculation based on the said trained weights in the said first and second neural networks in response to the appearance frequency of specific words obtained from the text data on reputations of accommodations input to the input layer of the said first neural network and to output the quantified values of reputations of accommodations from the output layer of the said second neural network.’ For this reason, in the trained model of Claim 1, a specific information processing system depending on intended use is constructed through cooperation of software and hardware resources. Therefore, since the information processing by software is concretely realised by using hardware resources, the trained model of Claim 1 is a creation of technical idea utilising a law of nature and thus falls under ‘invention.’ ” As demonstrated in this case example, in order to obtain a patent, a description of specific parameter information is not necessary; what is necessary is the existence of correlations between the data.
C. Invention of business method using IoT
14Business-related inventions in Japan are not patentable unless they utilise the laws of nature and are not technical ideas. However, in the case of a business-related invention having the aforementioned collaboration requirements for software-related inventions, there is an option to obtain a patent. In the case of the JPO, the “system and method of allocating an unmanned autonomous vehicle” has been published in two forms as a way to utilise data of IoT-related technologies9. Claim 1 of Type 1 is as follows:
15“Claim 1 is a system comprising a vehicle allocation server, a portable terminal which a person who desires a vehicle allocation has, and unmanned autonomous vehicles: wherein the portable terminal comprises a transmitting unit for transmitting the user ID and a vehicle allocation position to the vehicle allocation server. The vehicle allocation server comprises: a storing unit for storing information on a face image of a user corresponding to a user ID; an acquiring unit for acquiring information on the face image made to correspond to the user ID received from said portable terminal from the storing unit; a specifying unit for specifying an unmanned autonomous vehicle which can be allocated based on position information and the utilisation state of the unmanned autonomous vehicle; and a transmitting unit for transmitting information on the vehicle allocation position and information on the face image to the specified unmanned autonomous vehicle. The unmanned autonomous vehicle comprises: an autonomous driving unit for performing autonomous driving up to the vehicle allocation position; a face authentication unit for performing face authentication processing for surrounding people; and a judging unit for judging whether a person’s face matches the received face image as the person who desires vehicle allocation, thereby permitting utilisation of the unmanned autonomous vehicle.”
Figure 3: Case 2-9 System and method of allocating unmanned autonomous vehicle 1

16Type 1 of Figure 3 explains that it is patent-eligible as follows: “specific calculation or processing of information depending on the intended use, which is the allocation of unmanned autonomous vehicles, is implemented by specific means in which software and hardware resources collaborate.”
17Furthermore, Claim 1 of Type 2 of Figure 4 is “A system comprising a vehicle allocation server, a portable terminal which a person who desires vehicle allocation has, and unmanned autonomous vehicles. When the vehicle allocation server receives a vehicle allocation request for the unmanned autonomous vehicle for which a vehicle allocation position is specified from the person who desires the vehicle allocation, the vehicle allocation server allocates an unmanned autonomous vehicle to the person who desires the vehicle allocation10.”
Figure 4: Case 2-10 System and method of allocating unmanned autonomous vehicle 2

18In Type 2, “no information processing is specified. Therefore, it is not possible to determine that specific means or procedures for specific calculation or processing of information depending on the intended use, which is the an allocation of unmanned autonomous vehicles, have been specified. For this reason, in the invention of Claim 1, a specific information processing system or an operation method thereof depending on intended use is not constructed through cooperation between software and hardware resources.” It is explained that Type 2 does not fall under the category of “invention.”
19These two cases illustrate the handling of business method inventions utilising the IoT. In the case of an invention utilising software, it is necessary that specific information processing is performed by collaboration between software and hardware, as in the case described above. Type 1 consists of a vehicle allocation server equipped with a storage unit, a system consisting of an unmanned vehicle equipped with a face recognition unit, and a mobile terminal. Type 1 is demonstrated in an example realised by concrete means in which software and hardware collaborate. This is an example of obtaining a business-related patent for the data acquisition method. The explanation of Type 2 only refers to the business method of allocating unmanned vehicles, so the reason why it is not subject matter to patent is justified. The JPO case focuses on whether it consistently builds a unique technical configuration according to the purpose of utilisation. However, in these two cases, it can be pointed out that patent eligibility can be obtained by adding a sentence that identifies the hardware. In the case of a new business method, if there is no prior art, there is no novelty and underlying citation invention, thus, the inventive step may be affirmed. The determination of patent eligibility in Japan should be based on the definition of the invention (Patent Act, Art. 2, para.1). There should be no intention to easily clear the hurdle of patent eligibility for abstract ideas by using claim writing skills11.
IV. AI inventions in Japan and “inventorship”
20The recognition that “inventions” are made by natural persons seems to be common in many countries. Therefore, the case of AI/DABUS has been a topic of frequent discussion in recent years. Of particular note is how inventorship should be applied to the various roles that humans play in the creation, training and use of AI systems. It will provide an opportunity to rethink the rationale regarding inventorship. In this paragraph, we will discuss the concept of inventorship in the Japanese legal system and judicial precedents, and analyse some issues that may arise at this stage and in the future12.
A. Structure and criteria of the patent system in Japan
21The Japanese Patent Act does not clearly stipulate what kind of person the “inventor” is. An invention is defined as an “advanced creation of a technical idea utilising the laws of nature” (Patent Act, Article 2, Paragraph 1). Furthermore, Article 29, Paragraph 1, states that a person who has made an invention that can be used industrially can obtain a patent for the invention (inventor principle). For instance, the situations in which inventorship recognition is disputed are often as follows: usurped application, joint application violation, claim for compensation for an employee invention, and an honoured right for an inventor. However, the judicial precedent does not specify the “inventor” recognition method for each situation.
22If there is an application by a person who is not related to the invention (a person who does not have the right to obtain a patent) or a person who has not succeeded to the right to obtain a patent for the invention, this represents a usurped application and is a cause for rejection (Art. 49(vii)). When a patent is granted under these circumstances, there are grounds for invalidation (Art. 123(1)(vi)). A violation of this rule is a reason for rejection (Article 49, (2)) as a joint application violation (Article 38), and if a patent is granted, such a violation is a reason for invalidation. The inventor may be a single person, or there may be multiple persons involved. When an invention is jointly created, the right to obtain a patent is shared, and the application must be filed by all the inventors. Moreover, if the right to obtain a patent, or the patent right, is shared, the application, transfer of equity, and licence cannot be completed without the consent of all the co-owners. Furthermore, while the application is pending at JPO, amendments such as the addition or deletion of an inventor are possible, but after the patent right is registered, inventors cannot be added or deleted, regardless of inadvertent errors13.
23The Supreme Court of Japan states as follows: “It must be structured as concrete and objective to the extent that a person ordinarily skilled in the art in the relevant technical field can repeatedly carry out and achieve the desired technical effect14.” That is, the inventor must be involved in the technical idea of the invention. Furthermore, the subsequent Supreme Court decision was decided as follows: “It will be completed through the steps of setting a certain technical problem (purpose), adopting technical means to solve the problem, and confirming the effect that the desired purpose can be achieved by the technical means15.” Those involved in such inventions must be identified as inventors. According to previous judicial precedents regarding the criteria of inventorship, those who were creatively involved in the completion of the characteristic parts of the invention are to be considered inventors16. In the Porous Glass Case, the Intellectual Property (IP) High Court found that, “An inventor is a person who was involved in the highly advanced creation of technical ideas utilising the laws of nature, in other words, a person who was involved in the creative activity of the technical ideas concretely and objectively enough to enable a person ordinarily skilled in the art to implement the technical ideas17.” As an indication of the criteria of co-inventorship, in the Pfizer Case, the court also determined that the following cannot be deemed to have been actually involved in the creation of technical ideas and should therefore be excluded from the scope of joint inventors: (i) a person who has merely provided general management for the inventor (mere manager), such as by only having given the inventor general research themes without a specific concept; general guidance in the course of the creation of the invention; or abstract advice regarding the solution to the problem; (ii) a person who has merely followed the inventor’s instructions or assisted the inventor (mere assistant), for example, only having compiled data, prepared documents, or conducted experiments; (iii) a person who has merely supported the inventor in completing the invention (mere supporter), for example, by having provided the inventor with funds or allowed the inventor to use their equipment18. Therefore, in Japan, the real value of an invention lies in the technical idea, that is, a new solution to a technical problem, and the true inventor must actually be involved in the solution to the technical problem described in the specification.
B. Theories of inventor recognition in Japan
24There are two main ways of conceptualising the recognition of inventors in Japan based on the accumulation of judicial precedent thus far. The first approach has been adopted in a few court decisions. It is a way to make a determination by dividing it into two steps: providing a conception of an invention and reduction to practice (two-step test)19. 1) There is a decision that the person who provided the concept is the inventor when the concept provided is novel. For instance, in the Grain Processing Method Case, the true inventor was evaluated based on the time the invention was completed20. 2) In a judicial precedent in which a person who embodies a new idea when the materialisation is not obvious to a person skilled in the art is recognised as a co-inventor, the creation of a technical idea is not just an idea. Rather, the idea must be embodied, and the person who embodies the idea should been determined to be a co-inventor21.
25Secondly, the primary theory is to identify a person who was actually involved in the creation of the “characteristic part of the invention” as an inventor22. With this method, first, one identifies a characteristic part of the invention (a part that is not found in the prior art, in other words, a part that is fundamental to problem-solving and is specific to the invention). Then, one must identify 1) the person who presented the problem to be solved, 2) the person who devised the means of solving the problem, and 3) the person who confirmed that the problem was solved via those means. Finally, from among persons 1) to 3), a person who has made a contribution that is not obvious to a person skilled in the art in the process of creating the technical idea is deemed an “inventor23.”
26In this regard, there are two approaches: the first is to identify the conception and embodiment in two steps, and the second is to first determine the characteristic part of the invention. However, there is no difference in both approaches to determine whether a given person is actually involved in the creation of the technical idea of the invention, namely, the solution to the problem of the invention. In addition, these two approaches are not clearly distinguished in the court decisions mentioned above. These are merely a way to sort out in court decisions. Each does not exclude the other. Thus, the critical point is how to determine the characteristic parts of technical idea of the invention.
27Furthermore, there are also judicial precedents that consider the technical field in question, particularly in the chemical field, in which it is often not clear, without an experiment, whether a specific invention produces the desired effect. The Porous Glass Case was decided as follows: “In the chemical field, when a particular phenomenon is observed, this itself alone does not mean that the technical idea is concrete and objective enough to be put into practice by the person skilled in the art.” The court denied inventorship due to the absence of a practical contribution on the part of the plaintiff. In the mechanical field, it is often the case that the results embodied at the conception stage can be predicted. Regarding the previously mentioned antibacterial agent bag, the person who provided the idea for the invention was recognised as a co-inventor. However, technology is diverse, so the credit for each invention should be decided in light of its technical idea24.
C. AI inventions and human intervention
28An AI project team led by Professor Abbott of the University of Surrey filed two applications to raise questions about the modern patent system in 201925. DABUS is an AI developed by Dr. Stephen Thaler that can create ideas without human intervention26. These applications, which designated DABUS as an inventor in the UKIPO, The United States Patent and Trademark Office (USPTO) and EPO were rejected. The primary reason for these outcomes concerned the fact that DABUS was a machine. However, the Federal Court of Australia ruled that “AI would be recognised as an inventor” on July 202127. The problem here is the concern that the inventors of these AI inventions will be none. The discussion of AI autonomy invites people to a thought experiment, and usually human creativity has immeasurable consequences. AI can memorise enormous amounts of information; however, it does not work based on meaning. Actual AI development involves processes such as system design, data collection/selection, and demonstration, and is also a product of wisdom that many people are involved in. As mentioned above, an inventor needs to intervene directly/indirectly in an invention created by a technical idea. Thus, it is important to have a theoretical discussion focusing on human intervention28. In the future, with Society 5.0 realised, AI will analyse Big Data and provide feedback to humans in the best possible manner. The patent system should protect the ingenuity that humans have demonstrated. Without human intervention behind AI technology, AI inventions would not be created. Discussions on AI inventions and human intervention (direct and indirect) will continue to be necessary. As long as humans who use AI play a major role in inventing or unless the Act is amended, it is unlikely that AI will be identified as an inventor. However, it is necessary to adjust the interests of the rights entity and transaction activities in preparation for expanding the use of AI in the future.
V. AI inventions, the inventive step, and the person skilled in the art
29Regarding the inventive step in Japan, Article 29, Paragraph 2 of the Patent Act stipulates that a patent shall not be obtained if a person skilled in the art can easily invent at the time of filing a patent application. In determining the inventive step, the examiner examines whether or not it is possible to construct (logicalise) the logic that a person skilled in the art could easily conceive the claimed invention based on prior art29. A person skilled in the art is “a person of ordinary skill in the art of the invention in the patent application” (Article 29 (2) of the Patent Act), a person who assumes the following conditions: (1) having common general knowledge in the art at the time of filing, (2) the ability to use ordinary technical means for research and development, (3) the ability to exercise an ordinary level of creative ability, and (4) the ability to absorb as own knowledge all of the state-of-the-art knowledge in the art at the time of filing and technical knowledge in technical fields related to problems to be solved by the invention. A person skilled in the art may be considered to be a “team of experts” from multiple technical disciplines rather than an individual in some cases30. The aforementioned definition is not changed by the increasing use of AI in the inventive process31.
30There is no dispute that the inventive step requirement plays a role as a gatekeeper that distinguishes patented inventions. In Japan, there are several theories depending on the difference in the approach that is utilised as shown below32. For example, one is the idea of selecting inventions which were technically difficult to create33, and the other is to select inventions that are suitable for giving an incentive to protect patents, that is, to evaluate technical contributions34. According to Case 36 in the JPO regarding the determination of the inventive step of the dementia stage estimation apparatus35, claim 1, which identifies the questioner and the respondent, associates the question type with the answer, and trains machine learning as teacher data. It was determined that a person skilled in the art cannot easily conceive identifies the questioner and the respondent and letting them learn as teacher data, compared to the cited invention in which voice is machine-learned without specifying the questioner and the respondent. Furthermore, it was explained that it does not seem to be a mere design modification or matter of design choice for improving an estimation accuracy in the cited invention.
Figure 5: Case 36 Dementia stage estimation apparatus

31Based on the examples discussed above, the evaluation of JPO is based on the ease of generating the relevant technical information based on the state of technological development. That is, the inventive step plays a role in excluding inventions that are easily generated by using the conventional art from the subject of patented inventions. Legal protection of inventions that can be easily made by those skilled in the art does not contribute to the development of industry, which is the purpose of patent law, and there is no need to protect such inventions. The requirements of the patent subject matter are thought the function of denying patent protection for those that are expected to be developed, without giving incentives and those that hinder the development of the industry if giving patent protection36. However, unlike conventional rule-type software, in data-driven AI inventions, the generation of learning models in the process of the invention may be the key to the inventive step. Again, as for the inventive step, it is necessary to objectively evaluate the difference between the invention itself and the state of the art. As long as the constituent elements of the claimed invention are new and could not have easily been conceived when compared to the state of the art, the inventive step will be affirmed. When using published AI systems and data to generate trained model processes, using the publicly available AI system and data may be considered as the exercise of an ordinary level of creative ability and as a result, it is more likely that the invention would be found to be obvious (easily conceived) from the state of the art. Rather, it seems likely that the inventive step will be evaluated by the technical effects produced by data tuning. However, even in this case, how to disclose the claimed invention will be a problem. Therefore, in the following section, the necessity of disclosure of the invention will be examined.
VI. AI inventions and description requirements
32Sufficiency of disclosure shall disclose the claimed invention in a sufficiently clear and complete manner for the person skilled in the art to be able to carry it out. Since the inner workings of an AI are often undisclosed (black box), such enablement might be hampered by AI inventions. Thus, whether the sufficiency of disclosure requirements can be dealt with by the current law is an important issue37.
33In relation to the standard of sufficiency of disclosure, the Japan Patent Act provides for the enablement requirement (Article 36(4)(i)), definiteness requirement (Article 36(6) (ii) and the support requirement (Article 36(6)(i)). According to Article 36(4)(i), the statement of description needs to be clear and sufficient to enable a person ordinarily skilled in the art associated with the invention to work (produce and use) the invention. More specifically, the description needs to be so clear and sufficient that the person skilled in the art could work the invention without the need for a trial-and-error process or a highly complex experiment (Part II, Chapter 1, Section 2 (2) of the Patent Examination Guidelines). Software-implemented inventions have specific guidelines that are given for these fields (Part VII, Chapter 1). Moreover, the support requirement in the Patent Act provides that the claimed invention shall not go beyond the scope of the detailed description of the invention. In a court case on a Method for Producing Polarising Film38, the decision states: “when assessing the support requirement, the claims should be compared to the detailed description to determine whether the claimed invention is the invention in the detailed description. It is also necessary to determine whether the description is detailed to such an extent that a person skilled in the art could have solved the technical problems based on what is stated or suggested in the detailed description or a person skilled in the art could have solved the problems based on common technical knowledge at the time of filing, without such statements or suggestions.”
34When AI constitutes a claimed invention, it is necessary to support the specification of a certain correlation among each data in a training dataset. JPO case 46 about the description requirement states the following: “Claim: A sugar content estimation system comprising: a storage means for storing facial images of people and sugar contents of vegetables produced by the people; a model generation means for generating a determination model through machine learning, to which a facial image of a person is input and from which the sugar content of a vegetable produced by the person is output, using training data containing the facial images of the people stored in the storage means and the sugar contents of the vegetables, a reception means for receiving an input of an face image; and a processing means for outputting, using the generated determination model that has been generated by the model generation means, the sugar content of a vegetable produced by a person that is estimated based on the face image of the person inputted to the reception means39.”
Figure 6: Case 46 A sugar content estimation system

35In the claim, a facial image of data and sugar contain data is described as teacher data. However, the description only discloses that there is a certain correlation between a facial image of a person and the sugar content of a vegetable produced by the person and does not disclose any correlation between them. Case 46 is an example of a violation of enablement requirements, explaining that the details of claim was not disclosed in the description in a manner that would allow a person with ordinary skill in the art to create and use the system. In Europe, guidelines for AI inventions were provided in October 2018, and the EPO Board of Appeals has already rejected an application on the lack of disclosure sufficiency (T161/18)40. In this way, if AI itself constitutes the claimed invention, the future trends in AI technology may make it difficult to ensure its reproducibility by merely using an algorithm or flowchart, as in the case of a conventional software-implemented invention. When it becomes necessary to disclose the generation process of a trained model and/or the details of a learning model to meet the enablement requirement, the standard of sufficiency of disclosure should be reviewed to prevent a situation in which an excessive level of disclosure is required to compensate for a lack of transparency of AI technology41. There is also a proposed position to propose to resolve the uncertainty of AI by depositing data42. If such an alternative means is used to meet the sufficiency of disclosure requirement, it will become necessary to understand the details of the software at issue when assessing a patent in dispute. Furthermore, the process for analysing the claims may be extremely complicated.
VII. Conclusion
36Rather than focusing solely on what AI has generated, as we leverage AI in human society, we should accurately consider the intervention of human designers, programmers, and data trainers behind AI. As a premise, it is necessary that a sufficient description that can recognise the technical ideas of these inventions is provided in the specification. The advent of AI technology has brought about not only convenient technology, but also an opportunity to reconsider the potential challenges of modern patent law. Until now, there has been no need for deep debate concerning human intervention in the invention itself. However, when considering inventions that unfold in virtual space and real space, it is necessary to continue to consider the theory of how humans intervene in the emergent development of inventions. The businesses related to AI technology has unknown potential. Therefore, better creative work in the digital era is necessary to maximise this potential. In conclusion, the patent system should be an accelerator for innovation in a human-centred society rather than an obstacle to development. It is necessary to carefully continue theoretical discussions that respect human ingenuity.
Notes de bas de page
1 Ryan Abbott, Everything is Obvious, 66 UCLA L. Rev. 1-52 (2019).
2 Cabinet Office, <https://www8.cao.go.jp/cstp/english/society5_0/index.html>, accessed Aug. 30, 2021.
3 Ryan Abbott, I Think, Therefore I Invent: Creative Computers and the Future of Patent Law, 57 B.C.L. Rev. 1079 (2016); Peter Blok, The inventor’s new tool: Artificial Intelligence – how does it fit the European patent system?, European Intellectual Property Review, 39(2) 69-73 (2017); Dhenne Matthieu, Artificial Intelligence: Back to the Future of Patent Law (2021). <https://ssrn.com/abstract=3787526or http://dx.doi.org/10.2139/ssrn.3787526>, accessed Aug. 30, 2021; Ichiro Nakayama, Patentability and PHOSITA in the AI Era- A Japanese Perspective, in Jyh-An Lee, Reto Hilty & Kung-Chung Liu ed., Artificial Intelligence and Intellectual Property, Oxford Scholarship Online, 99-119 (2021); Dan L. Burk, AI Patents and the Self-Assembling Machine, 105 MINN. L. REV. Headnotes 313 (2021).
4 WIPO, Conversation on Intellectual Property (IP) And Artificial Intelligence (AI) (2020),
<https://www.wipo.int/edocs/mdocs/mdocs/en/wipo_ip_ai_3_ge_20/wipo_ip_ai_3_ge_20_inf_5.pdf>; AIPPI, Study Question Inventorship of Inventions made using Artificial Intelligence, Study Guidelines, Q272-SGL-P-2020, <https://aippi.info/wp-content/uploads/2020/02/Q272-SGL-P-2020-Inventorship-of-inventions-made-using-AI.pdf.>, accessed Aug. 30, 2021.
5 Ryuta Hirashima, New Issues in patent law brought about by inventions related to machine learning and deep learning, PATENT (separate vol. No. 23), 73, 8, 163-188 (2020) (in Japanese).
6 “Examination Handbook for the Patent and Utility Model in Japan” (JPO, 2019), Annex B Chap. 1, 2.1.1.2.
7 “Case Examples pertinent to AI-related technology” (JPO, 2019), Case 2-13, p. 63. <https://www.jpo.go.jp/e/system/laws/rule/guideline/patent/document/ai_jirei_e/jirei_e.pdf> accessed August 31, 2021.
8 “Case Examples pertinent to AI-related technology” (JPO, 2019), Case 2-14, p. 68.
9 “Examination Guidelines pertinent to IoT-related technology” (JPO, 2016), Case 2-9 (Annex B). <https://www.jpo.go.jp/e/system/laws/rule/guideline/patent/document/iot_shinsa/01.pdf> accessed August 31 2021.
10 “Examination Guidelines pertinent to IoT related technology”(JPO, 2016), Case 2-10 (Annex B).
11 Yoshiyuki Tamura, Significance of the original functional theory of things in defining patent-eligible objects, PATENT (separate vol. No. 26), 74, 4 (2021) (in Japanese).
12 AIPPI Study Committees “Inventorship of inventions made using Artificial Intelligence” Japan National Group Report <https://aippi.soutron.net/Portal/DownloadImageFile.ashx?objectId=8418>, accessed August 31, 2021.
13 In contrast, in the United States, patent co-owners are free to license co-patents to third parties without the consent of the co-owner. Therefore, as long as the inventor’s indication error is not of deceptive intent, it can be corrected even if it is pending in the USPTO or if the patent is issued., Etsuko Yoshida, “The Theory and Practice of Determining Co-inventorship: A Case Study from University Research Activities”, Patents & Licensing, 50, 6, 6-16 (2021).
14 Supreme Court of Japan of October 13, 1977, Minshu Vol. 31, No. 6, p. 805.
15 Supreme Court of Japan of October 3, 1986, Minshu Vol. 40, No. 6, p. 1068.
16 Tokyo District Court of August 27, 2002, Hanrei jiho No. 1810, p. 102.
17 Intellectual Property High Court of May 29, 2008, Hanrei jiho, No. 2018, p. 146.
18 Tokyo District Court of September 13, 2005, Hanrei jiho No. 1916, p. 133.
19 Kosaku Yoshifuji, Kenichi Kumagai, “Tokkyo Hou Gaisetsu [revised & enlarged 13th ed. ]” (Overview of Patent Law) p. 188 (Yuhikaku 1998) (in Japanese).
20 Tokyo District Court of April 16, 1979, Hanrei Times, No. 395. p. 155.
21 Tokyo High Court of April 27, 1976, Torikeshishu 1976, p. 449.
22 Nobuhiro Nakayama, Naoki Koizumi, “Shin-tyuukai Tokkyo Hou (2nd ed)”, p. 386 (Seirinshoin 2017) (in Japanese).
23 Intellectual Property High court of July 30, 2007, Heisei 18 (Gyo-Ke)10048; As the same determination method, Intellectual Property High court of September 30, 2008, Heisei 19 (Gyo-Ke)10278 [Wafer Case], Intellectual Property High Court of March 13, 2013, Hanrei Jiho No. 2201, p. 116 [Double eyelid forming tape Case] and so on.
24 Takakuni Yamane, Shin Tokii, “Certification of the true inventor-Fine-grained Core Case,” Intellectual Property law and Policy Journal, 20, 287, 293-296 (2008) (in Japanese).
25 The Artificial Inventor Project, https://artificialinventor.com/ Accessed 31 Aug. 2020.
26 Two patents for a food container and a flashing light for use in emergencies were filed (EP18275163, EP18275174).
27 The South African patent office, the Companies and Intellectual Property Commission (CIPC) has registered a patent on AI/DABUS inventions and published it in the South African Patent Journal in July 2021. – AI/DABUS as the inventor and the machine’s owner as the patent owner (unexamined system). Moreover, the Japanese Patent Office issued a notice “regarding the indication of the inventor” on July 30, when the Federal Court of Australia ruled. It is as follows (Japanese only): “We understand that the indication of inventor is limited to natural persons. In the column of the inventor described in the application, etc., it is not permitted to describe a machine that is not recognised as a natural person, for example, a machine containing artificial intelligence (AI), etc. as an inventor.” The case concerning transfer of PCT applications to Japan is ongoing.
28 Daria Kim, ‘AI-Generated Inventions’: Time to Get the Record Straight, GRUR Int., 69, 5, 443-456 (2020); Eva Stanková, Human Inventorship in European Patent Law, Cambridge Law Journal, 80, 2, 342 (2021).
29 “Examination Guidelines for Patent and Utility Model in Japan” (JPO, 2019), Part III Chapter 2 Section 2.
30 “Patent Examination Guidelines” (JPO, 2019), Part III, Chapter 2, Section 2.2. (According to general practices in Japan, it is quite rare that the person skilled in the art is identified in individual cases where the inventive step is assessed, as in the UK and the US.), The following paper examines the concept of those skilled in the art in order to determine the inventive step of AI inventions from the viewpoint of international harmony: Ramalho Ana, Patentability of AI-Generated Inventions: Is a Reform of the Patent System Needed?, Institute of Intellectual Property report, 1-32 (2018).< https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3168703>, accessed August 31, 2021.
31 Although it is not about AI inventions, the determination of the referral case G1 / 19 (the invention of computer-implemented simulations) of the Enlarged Board of Appeal of the European Patent Office in 2021 will be helpful. The Enlarged Board of Appeal has determined that computer-implemented simulations are patentable following the previous criteria (COMVIK approach).
32 Shin Tokii, Statistical Analysis of Judicial Decisions on Inventive Step, and Its Theoretical Implications: A Comparative Study of Japan, China, Germany, EPO and US (1), Intellectual Property Law and Policy Journal Vol.54 at 61 (2019); Takeshi Maeda, The Meaning of inventive step requirements and method of determination, Annual of Industrial Property law Vol. 44, (Yuhikaku, 2020) at 92 (in Japanese).
33 Takeshi Maeda, An Empirical Study of the Inventive Step in Japanese IP High Court Cases and Reconstruction of its Test from a Functional View, IIP Bulletin 2014, Vol. 23, pp. 150-168.
34 Kazuaki Yamashita, Reason for rescinding the trial decision (ruling), Practice and doctrine of a patent trial decision revocation suit, edited by Minoru Takeda et al. (Japan Institute of Invention and Innovation, 2003) at 160 (in Japanese).
35 “Case example pertinent to AI-related technology” (JPO, 2019), Case 36.
36 Yoshiyuki Tamura, Chitekizaisan Hougaku no Kadai: Tabi no Totyuu(Intellectual Property Law Challenges: During the Journey), Intellectual Property law and Policy Journal, 51, 1-46 (2018): Ryuta Hirashima, Scope of invention protection from the viewpoint of the role and function of patent law-Suggestions for Japanese law obtained from trends in American law-, Annual of Industrial Property Law Vol. 40, (Yuhikaku, 2017) at 121-122 (in Japanese).
37 AIPPI Study Question, Inventiveness and Sufficiency of Disclosure in AI Inventions, Q276-SGL-P-2021(2021), <https://aippi.soutron.net/Portal/DownloadImageFile.ashx?objectId=8542>, accessed August 31, 2021; Tabrez Y. Ebrahim, Artificial Intelligence Inventions & Patent Disclosure, 125 Penn State Law Review 1, 147-221 (2020).
38 Intellectual Property High Court Grand Panel of 11 November 2005, Hanrei Jiho, No. 1911 p. 48.
39 “Case Example pertinent to AI-related technology”(JPO, 2019), Case 46.
40 T 0161/18 (Äquivalenter Aortendruck/ARC SEIBERSDORF) of 12.5.2020.
41 See supra note 5, where it is pointed out that the machine learning-related inventions have a low affinity with the description requirements under the current Japanese patent law.
42 Dan L. Burk, op. cit., 313; Tabrez Y. Ebrahim, op. cit., 214.
Auteur
Assistant Professor
Intellectual Property and Legal practice Centre, Osaka University
Le texte seul est utilisable sous licence Licence OpenEdition Books. Les autres éléments (illustrations, fichiers annexes importés) sont « Tous droits réservés », sauf mention contraire.
Qu’en est-il du code du commerce 200 ans après ?
États des lieux et projections
Corinne Saint-Alary-Houin (dir.)
2008
Qu'en est-il de la simplification du droit ?
Frédérique Rueda et Jacqueline Pousson-Petit (dir.)
2010
La réorientation européenne de la TVA à la suite du renoncement au régime définitif
Francis Querol (dir.)
2014
Regards critiques sur quelques (r)évolutions récentes du droit
Tome 1 : Bilans et Tome 2 : Réformes-Révolutions
Maryvonne Hecquard-Théron et Jacques Krynen (dir.)
2005