Service delivery system, service delivery method, program, and server

Abstract

This addresses the issue of content rights (such as copyright and the right to collect royalties associated with the use of paid content) when artificial intelligence uses high-quality content data with a high degree of accuracy. [Solution] The generating AI, such as LLM, inserts hyperlinks to referenced content (books, etc.) in its responses to user-inputted prompts. When the user clicks on the hyperlink, they can view the content if they have the right to use it (right of use certificate 77). If they do not have the right to use it, they are redirected to the content sales page where they can purchase the content to acquire the right to use it. The display of hyperlinks by the generating AI serves as advertising for the content and benefits the content rights holders, creating a positive cycle where content rights holders proactively provide content to the generating AI.

Description

Technical Field

[0001] The present invention relates to a service providing system in which an artificial intelligence provides a service to a user, a service providing method in which an artificial intelligence provides a service to a user, a program executed on a computer used for providing a service to a user by an artificial intelligence, and a server.

Background Art

[0002] Artificial intelligence such as generative AI represented by generally known LLM (Large Language Models) etc. conventionally generates an answer to a prompt input by a user and presents it to the user. Such a generation function is made possible by, for example, using a vast amount of data such as Web on the Internet as machine learning data for pre-training of artificial intelligence and then performing fine-tuning. When the pre-trained data alone is insufficient, the artificial intelligence may use RAG (Retrieval-Augmented Generation) etc. to search for external information such as the Web and generate an answer using the retrieved information. There is also artificial intelligence programmed to show a hyperlink of the Web that the artificial intelligence referred to in generating an answer within the answer, and when a user who saw it selects the hyperlink, it transitions to the linked Web page and becomes viewable by the user. For the user, there is an advantage that they can confirm the basis information of the answer by viewing the Web information for reference.

[0003] However, there is a problem that hallucination occurs when artificial intelligence uses such information because the Web information contains information different from facts. Hallucination is a phenomenon in which artificial intelligence generates information different from facts or non-existent information based on the data it has learned.

[0004] To address this hallucination problem, there was a technology (for example, Patent Document 1) in which the user inputs prompts to a generating AI, the user evaluates the content output by the generating AI, the input prompts, the output content, and the user's evaluation are stored as historical information, and a reward model is trained based on this historical information. [Prior art documents] [Patent Documents]

[0005] [Patent Document 1] Patent No. 7530134 [Overview of the Initiative] [Problems that the invention aims to solve]

[0006] In this type of background technology, hallucination is prevented based on user evaluation, which requires users to have the ability to discern whether the content generated by the AI ​​is factually incorrect. However, few ordinary users possess this ability, which is a drawback as it limits the effectiveness of hallucination prevention.

[0007] The root cause of hallucination lies in the low accuracy (reliability) of the data used by artificial intelligence. An effective way to resolve this root cause is to use data with high accuracy (reliability), such as book data or paid content, for artificial intelligence. However, if artificial intelligence were to use such high-quality content data without permission, there is a risk of problems such as copyright infringement and unauthorized use of paid content.

[0008] In other words, when artificial intelligence uses high-quality content data with a high degree of truthfulness (accuracy), it is necessary to clear up issues related to the content, such as copyright and the right to collect royalties for the use of paid content (hereinafter referred to as "content rights").

[0009] This invention was conceived in view of the above circumstances, and its purpose is to resolve the content rights issues that arise when artificial intelligence uses high-quality content data with a high degree of truthfulness (accuracy). [Means for solving the problem]

[0010] The solutions provided by the present invention are listed below. In this invention, "person" is a broad concept that includes natural persons, legal entities, and artificial intelligence. Furthermore, in this invention, "artificial intelligence" is a broad concept that includes agents, multi-agents, and mobile agents.

[0011] (1) A service provision system in which artificial intelligence provides services to users, The system includes a reference content usage control means that controls how the reference content, which is the content that the artificial intelligence referenced in order to provide the aforementioned service, can be made available to the user. The aforementioned reference content usage control means is: A means for determining whether a user has acquired a paid usage right that permits the use of the aforementioned reference content, The system includes an allowance control means that controls whether a user is permitted to use the reference content, provided that the user is determined to have acquired the right of use by the right of use determination means, The consideration paid by the user who acquired the aforementioned usage rights shall be granted to the person who holds the rights associated with the aforementioned reference content.

[0012] With this configuration, the fees paid by users for acquiring the right to use the content referenced by the artificial intelligence are given to the content rights holders who hold the rights associated with the referenced content. This provides content rights holders with an incentive to provide content to the artificial intelligence, and minimizes the content rights issues that arise when the artificial intelligence uses high-quality content data with a high degree of accuracy.

[0013] (2) In the above (1), the reference content usage control means includes a presentation control means that performs control to present specific information identifying the reference content to the user, The permission control means controls the user to use the reference content identified by the specific information, provided that the user who performed the access operation to the specific information is determined by the usage right determination means to be a user who has acquired the usage right.

[0014] With this configuration, specific information identifying the reference content that the AI ​​used is presented to the user, thus promoting the content. The consideration paid by the user for acquiring the right to use that reference content is given to the content rights holder who holds the rights associated with the reference content. This provides content rights holders with an incentive to provide content to the AI, and minimizes the content rights issues that arise when the AI ​​uses high-quality content data with high accuracy.

[0015] (3) In the above (1), the reference content usage control means includes a presentation control means that controls the presentation of specific information identifying the reference content to the user, The aforementioned allowable control means is An access determination means for determining whether an operation to access the presented specific information has been performed by a user who has been determined to have acquired the right to use the information by the aforementioned right to use determination means, Conditional on the access determination means determining that an access operation has been performed, control is implemented to allow the user to use the reference content identified by the specific information.

[0016] With this configuration, specific information identifying the reference content that the AI ​​used is presented to the user, thus promoting the content. The consideration paid by the user for acquiring the right to use that reference content is given to the content rights holder who holds the rights associated with the reference content. This provides content rights holders with an incentive to provide content to the AI, and minimizes the content rights issues that arise when the AI ​​uses high-quality content data with high accuracy.

[0017] (4) A service provision system in which artificial intelligence provides services to users, Presentation control means that controls the presentation of specific information to the user that identifies the reference content, which is the content that the artificial intelligence used as a reference in order to provide the aforementioned service. The system includes a purchase page display control means that controls the display of the purchase page for the reference content identified by the aforementioned specific information, The consideration paid by a user who purchases the reference content identified by the specified information from the displayed purchase page is given to the person who holds the rights associated with the content.

[0018] With this configuration, the presentation of specific information to users that identifies the reference content used by the artificial intelligence serves as promotion for that content. The display of purchase pages for the reference content identified by the specific information promotes the sale of the reference content, making it easier to incentivize content rights holders to provide their content to the artificial intelligence. Furthermore, since the payment made by users who purchase the reference content identified by the specific information from the purchase page is given to those who hold the rights associated with the reference content, it further incentivizes content rights holders to provide their content to the artificial intelligence.

[0019] (5) A service provision system in which artificial intelligence provides services to users, Service providing means for providing a service to a user by using the artificial intelligence with reference to content, Reference content usage control means for performing control to enable a user to use reference content, which is the content referred to by the artificial intelligence when providing a service by the service providing means, The service providing means, Usage right determination means for determining whether or not a user has acquired a paid usage right that permits the use of the reference content, Service permission means for permitting the service to be provided to a user on the condition that the user is determined by the usage right determination means to have acquired the usage right, The consideration paid by the user who has acquired the usage right is given to the person having the right associated with the reference content.

[0020] According to such a configuration, the consideration paid by a user for acquiring the usage right of the content referred to by the artificial intelligence in order to enjoy the service by the service providing means is given to the content right holder having the right associated with the reference content. This can give an incentive to the content right holder to provide content to the artificial intelligence, and can clear the content right problem when the artificial intelligence uses high-quality content data with high authenticity (accuracy) as much as possible.

[0021] (6) In the above (5), it further includes reference content designation means for designating content that a user who intends to receive the service provided by the service providing means wishes to refer to, The service providing means provides a service to the user by using the content designated by the user as a reference, The service permission means permits the service to be provided to the user on the condition that the usage right determination means determines that the user has acquired a paid usage right that permits the use of the content designated by the user.

[0022] (7) In the above (1), the content includes books, The rights associated with the aforementioned content include the copyright of the aforementioned book.

[0023] (8) In (2) above, the permission means includes viewing permission means that allows the user who selected the specific information to view the reference content identified by the specific information, on the condition that the user has been determined by the usage right determination means to have acquired the usage right.

[0024] (9) The above (2) further includes a purchase page display control means that controls the display of the purchase page of the reference content identified by the specific information when a user who has not acquired the usage rights selects the specific information, A user can acquire the right to use the aforementioned reference content when they purchase it.

[0025] With this configuration, the purchase page for reference content is displayed to the user, promoting the sale of the reference content and thus creating an incentive for content rights holders to provide their content to artificial intelligence.

[0026] (10) In the case of (1) above, if a user who has acquired the usage rights transfers the content that is permitted to be used by said usage rights, the system further provides a means for granting usage rights that permits the use of said content to the transferee of the content.

[0027] With this configuration, even when content is transferred from one user to another, the new transferee is granted the right to use it, thus preventing the inconvenience of the transferee being unable to use the content.

[0028] (11) The system further includes a content provision means that allows the artificial intelligence to provide content, provided that the artificial intelligence is an artificial intelligence that presents specific information identifying the reference content to the user by a presentation control means that controls the presentation of specific information identifying the reference content to the user.

[0029] (12) A method of providing services in which artificial intelligence provides services to a user, This includes a usage control step that controls how to make reference content, which is content that the artificial intelligence referenced in order to provide the aforementioned service, available to the user. The aforementioned usage control step is: A usage rights determination step that determines whether the user has acquired paid usage rights that permit the use of the aforementioned reference content, The system includes an allowance control step that controls whether a user is allowed to use the reference content, provided that the user is determined to have acquired the right to use the reference content in the right to use determination step. The consideration paid by the user who acquired the aforementioned usage rights shall be granted to the person who holds the rights associated with the aforementioned reference content.

[0030] With this configuration, the fees paid by users for acquiring the right to use the reference content that the artificial intelligence uses are given to the content rights holders who hold the rights associated with the reference content. This provides content rights holders with an incentive to provide content to the artificial intelligence, and minimizes the content rights issues that arise when the artificial intelligence uses high-quality content data with a high degree of accuracy.

[0031] (13) A program that is used to provide services to users by artificial intelligence and is executed on a computer, To the aforementioned computer, The artificial intelligence performs a usage control step to control how to make the reference content, which is the content that the artificial intelligence used as a reference for providing the aforementioned service, available to the user. The aforementioned usage control step is: The system includes the step of providing a service to a user that allows the use of the aforementioned reference content, provided that the user has acquired a paid usage right that permits the use of the aforementioned reference content. Furthermore, the computer is made to perform a step in which it accepts an operation in which a user pays a fee to acquire the usage rights. The consideration paid by the user who acquired the aforementioned usage rights shall be granted to the person who holds the rights associated with the aforementioned reference content.

[0032] With this configuration, the fees paid by users for acquiring the right to use the reference content that the artificial intelligence uses are given to the content rights holders who hold the rights associated with the reference content. This provides content rights holders with an incentive to provide content to the artificial intelligence, and minimizes the content rights issues that arise when the artificial intelligence uses high-quality content data with a high degree of accuracy.

[0033] (14) A server used in a service provision system in which artificial intelligence provides services to users, Equipped with a processor and memory, The aforementioned processor, In order to provide the aforementioned service, the artificial intelligence performs a reference content usage control process to control how the reference content, which is the content it referenced, can be made available to the user. The aforementioned reference content usage control process is: A usage rights determination process that determines whether the user has acquired a paid usage right that permits the use of the aforementioned reference content, The process includes an allowance control process that controls whether a user is permitted to use the reference content, provided that the user is determined to have acquired the right of use by the right of use determination means, The consideration paid by the user who acquired the aforementioned usage rights shall be granted to the person who holds the rights associated with the aforementioned reference content.

[0034] With this configuration, the fees paid by users for acquiring the right to use the reference content that the artificial intelligence uses are given to the content rights holders who hold the rights associated with the reference content. This provides content rights holders with an incentive to provide content to the artificial intelligence, and minimizes the content rights issues that arise when the artificial intelligence uses high-quality content data with a high degree of accuracy. [Brief explanation of the drawing]

[0035] [Figure 1] This is a conceptual diagram of a content AI utilization system. [Figure 2] This diagram shows the overall configuration of the content AI utilization system. [Figure 3] (A) is a diagram showing the hardware configuration of the user terminal, and (B) is a diagram showing the hardware configuration of the cloud server. [Figure 4] (A) is a functional block diagram of the AI ​​platform system, and (B) is a diagram showing the data stored in the content database. [Figure 5] (A) is a diagram showing the data stored in the AI ​​registration database, and (B) is an explanatory diagram showing the mechanism of Verifiable Credentials. [Figure 6] This is an explanatory diagram showing the blockchain data recorded at each node of the blockchain. [Figure 7] This is a flowchart of the main routine program for control processing in a content AI utilization system. [Figure 8] (A) is a flowchart showing subroutine programs for book registration processing by the publisher's terminal and book registration support processing by the cloud server, and (B) is a flowchart showing subroutine programs for AI-based registration processing and AI registration support processing by the cloud server. [Figure 9](A) is a flowchart showing subroutine programs for AI-based access processing and cloud server-based access response processing, and (B) is an explanatory diagram showing the mechanism of RAG (Retrieval-Augmented Generation). [Figure 10] This flowchart shows the subroutine programs for AI utilization processing on the user terminal and AI utilization support processing on the cloud server. [Figure 11] This flowchart shows the continuation of the subroutine programs for AI utilization processing on the user terminal and AI utilization support processing on the cloud server. [Figure 12] This flowchart shows the search process for the user's terminal to retrieve the usage rights certificate data for the content they wish to view, and the corresponding subroutine program on the company server. [Figure 13] This diagram shows a screen on the user's terminal displaying the answer to the user's prompt, and a screen of the book purchase webpage that is displayed after the user selects the hyperlink to the book that is the basis for the answer displayed on the screen. [Figure 14] This is a flowchart of a subroutine program for AI-based response processing. [Figure 15] (A) is a flowchart showing a subroutine program for sharing highly-rated answers by a cloud server, and (B) is a flowchart showing a subroutine program for compiling book rankings by a cloud server. [Figure 16] This flowchart shows the subroutine program for processing user-specific book rankings using a cloud server. [Figure 17] This figure shows a screen displaying the results of the user-specific book ranking aggregation process on the user's terminal, and a screen showing the book purchase web page that the user is redirected to when they select a book displayed on that screen. [Figure 18]This flowchart shows a subroutine program for smart contract processing to share highly-rated responses between highly-rated AIs and other AI groups. [Figure 19] This diagram shows the data stored in the knowledge-sharing database. [Figure 20] This is a flowchart of the subroutine program for the AI-driven AI alliance formation process. [Figure 21] This is a flowchart of the subroutine program for AI-generated response processing. [Figure 22] (A) is a functional block diagram of the modified AI platform system, and (B) is a diagram showing the data stored in the modified AI registration DB. [Figure 23] This flowchart shows the subroutine programs for AI utilization processing by the user terminal and AI utilization support processing by the cloud server in the modified example. [Figure 24] (A) is a flowchart showing a subroutine program for identifying suitable AI and useful books in response to prompts received by the cloud server in the modified example, and (B) is a flowchart showing a subroutine program for the AI's response processing in the modified example. [Figure 25] This flowchart shows the subroutine programs for processing operations performed by the user terminal (EPUB application), the publisher's terminal, and the cloud server in the second embodiment. [Figure 26] (A) is a flowchart showing the continuation of the subroutine program for processing operations by the user terminal (EPUB application), publisher's terminal, and cloud server in the second embodiment, and (B) is a diagram showing the data stored in CFIDB in the second embodiment. [Figure 27] This figure shows the overall configuration of the content AI utilization system in the third modified example. [Figure 28] This flowchart shows the subroutine programs for processing operations performed by the user terminal and the AI ​​group specified by the user in the third modified example. [Figure 29] This is an explanatory diagram showing blockchain data stored in each node of the blockchain in the fourth embodiment, and it shows the transaction data of resale transaction 1, in which book data was resold, and the NFT data of the book data that was resold. [Figure 30] This is an explanatory diagram showing the blockchain data stored in each node of the blockchain in the fourth embodiment. (A) is a diagram showing the transaction data of resale transaction 1, in which the book data was resold, and the Verifiable Credential proof data 1 of the person who became the new book purchaser. (B) is a diagram showing the transaction data of resale transaction 2, in which the book data was resold, and the Verifiable Credential proof data 2 of the person who became the new book purchaser. [Figure 31] This is a flowchart showing the subroutine program for the NFT creation process by the user terminal in the fourth embodiment. [Figure 32] This flowchart shows the subroutine programs for smart contract processing by user terminals A and B and the corresponding processing by the cloud server in the fourth embodiment. [Figure 33] (A) is an explanatory diagram showing how to update a smart contract using the Proxy pattern, and (B) is a flowchart showing the operation process of the user terminal (EPUB application) in the fourth embodiment. [Figure 34] This is a flowchart showing the NFT creation process of a user terminal in the fourth embodiment. [Figure 35] This flowchart shows the subroutine programs for smart contract processing by user terminals A and B and the corresponding processing by the cloud server in the fourth embodiment. [Figure 36] This is a flowchart showing the subroutine program for the reading process performed by user terminal B in the fourth embodiment. [Figure 37]A fourth embodiment is shown, where (A) is an explanatory diagram illustrating the case where book data is distributed among multiple parties for resale, and (B) is an explanatory diagram showing the price fluctuations in the secondary market when resold in a distributed manner. [Figure 38] A fourth embodiment is shown, where (A) is an explanatory diagram illustrating the case in which security tokens (digital securities) of book data are issued and sold to multiple parties, and (B) is an explanatory diagram showing the price fluctuations of the security tokens in the secondary market when they are sold. [Figure 39] (A) is a functional block diagram showing the reading content utilization system in the fifth embodiment, and (B) is a diagram showing the data stored in the company's shared database in the fifth embodiment. [Figure 40] This is a flowchart showing the subroutine programs for operation processing by the terminal and the enterprise server in the fifth embodiment. [Figure 41] This flowchart shows the subroutine programs for the terminal's request process for creating a persuasive presentation and the enterprise server's process for creating a persuasive presentation in the fifth embodiment. [Figure 42] This flowchart shows the subroutine programs for the terminal's process of requesting personnel selection for projects, etc., and the corporate server's process of selecting personnel for projects, etc., in the fifth embodiment. [Figure 43] This is a flowchart showing the subroutine program for security token processing by the AI, other AI groups, and user terminal or other AI groups in the sixth embodiment. [Figure 44] This is a flowchart showing a subroutine program for security token transaction processing by an AI and a user terminal or other AI group in the sixth embodiment. [Figure 45] (A) is a flowchart showing the subroutine programs for AI-based dividend processing and dividend receipt processing by user terminals or other AI groups in the sixth embodiment. [Figure 46]This is a flowchart showing the subroutine program for security token transaction processing by user terminal a or AIa and user terminal b or AIb in the sixth embodiment. [Figure 47] This flowchart shows the subroutine program for displaying each of the replies in the modified example. [Figure 48] This flowchart shows the subroutine programs for sending a prompt from the user terminal and sending the received prompt to the AI ​​by the cloud server in a modified example. [Modes for carrying out the invention]

[0036] [Explanation of terms] This embodiment will be described in detail based on the drawings. Hereafter, a database will be referred to as DB and artificial intelligence as AI. Note that the DB may be configured as a storage server. AI is a broad concept that includes agents, multi-agents, and mobile agents. Furthermore, while the explanation will mainly use ebooks as examples of copyrighted works, it is not limited to them and is a broad concept that includes images, videos, music, consumer-generated media (CGM), and even research papers. Paid CGM is paid content that can be viewed or transferred on the condition of payment of a fee such as money. These are collectively referred to as "content." Note that CGM is a broad concept that also includes content generated by AI.

[0037] The copyright, moral rights, and other rights held by the author or owner of content are collectively referred to as "content rights." Furthermore, creators of paid content such as paid CGM (Consumer Generated Media) have the "right to collect fees" from viewers or transferees, and "content rights" is a broad concept that includes this "right to collect fees." While copyright is generally not recognized for creative works generated by AI, such as generation AI, if AI-generated creative works (content) are allowed to be viewed or transferred for a fee, the AI ​​that generated the paid content becomes the "content rights holder." Therefore, in this embodiment, "person" is a broad concept that includes not only natural persons and legal entities but also AI.

[0038] In this embodiment, the AI ​​has the function of responding to prompts entered by the user. A "prompt" includes requests such as "I would like you to make a hotel reservation" from the user. Therefore, when the AI ​​responds to such a "prompt," it can be considered that the AI ​​is performing task processing. Furthermore, the AI's "response" also includes the AI ​​(agent) "performing task processing and responding" in response to the user's "request for a task." All of these are included and expressed as a "response." Note that in the drawings, the prompt is written as "Prompt." [First Embodiment] The content AI utilization system in this embodiment enables AI to utilize content data such as books.

[0039] Traditionally, generative AI such as LLMs (Large Language Models) have provided users with a service that responds to prompts they input (including responses to task processing), and sometimes displays a hyperlink to the web page that serves as the basis for that response. Users can click on this hyperlink to navigate to and view the web page.

[0040] However, if the AI ​​were to present copyrighted content or other material as the basis for its answer, a problem would arise in that if a user were to view that content to verify the basis for the answer, there is a risk of infringing on the content rights of that material.

[0041] Furthermore, if an AI uses content data such as books as training data during its pre-training phase, copyright (content rights) issues may arise in some countries.

[0042] A content AI utilization system solves these problems, and its basic concept is explained based on Figure 1.

[0043] The stakeholders in implementing this content AI utilization system include 78 content rights holders (copyright holders, publishers, record producers, broadcasters, cable broadcasters, etc.), 75 AI development and operation companies, and the users. This content AI utilization system will create a win-win-win situation where all three parties are satisfied.

[0044] The ideal scenario for content rights holders 78 is for sales of the content, which is the object of the content rights, to be promoted and revenue to increase. In the content AI utilization system, the AI ​​is tasked with promoting sales. When responding to a user prompt, the AI ​​includes a hyperlink to the content it referenced as evidence within the response and presents it to the user. The user can view the linked content by clicking the hyperlink, provided that the user is a purchaser of the content (a rights holder who possesses the right to use it). Users who wish to view the content that forms the basis of the response will purchase that content, promoting its sales and benefiting content rights holders. The purchase price of the content by the user is paid to content rights holders, and royalties are paid to copyright holders. In this embodiment, a purchased electronic certificate of content (usage rights electronic certificate) 77 is used as proof that the user is a purchaser (usage rights holder) of the content. In this embodiment, a Verifiable Data Registry is used as this usage rights electronic certificate 77. This will be described later.

[0045] This Verifiable Data Registry (DID)77 proves that the person is a purchaser of content, but is not limited to that. For example, it can prove that the person has the right to view or download content or paid content, the right to own or possess content or paid content, or that the person is a licensee who has received permission to use content from the content rights holder. In other words, it can prove that the person is a holder of the right to use content or paid content. Furthermore, DID certificates (DID)777 can be issued not only to natural persons but also to organizations such as companies.

[0046] If AI-generated content presentation serves as promotion and boosts sales, then content rights holders will have an incentive to proactively provide their books and other content to AI development and operation companies. As a result, AI development and operation companies will be able to use data from books and other content as machine learning data during the pre-training phase, enabling them to prevent hallucination and provide accurate answers with high-quality training data. Moreover, they will be able to refer to content to answer users' questions and enjoy the benefit of being able to present content as the basis for their answers.

[0047] On the other hand, users can enjoy the benefit of receiving appropriate answers based on the content, as well as being able to learn about the content that formed the basis of those answers.

[0048] Thus, this content AI utilization system is a business model that creates a win-win-win situation for all parties involved. As a result, it can generate the following positive spiral (virtuous cycle).

[0049] 1. When content rights holders proactively provide books and other content to AI development and operation companies, the suitability of AI-based services such as providing answers will improve.

[0050] 2. As a result, the benefits users receive from using AI will increase, leading more users to utilize content AI systems.

[0051] 3. By having AI provide evidence for content to the increased number of users, content sales will be further promoted, leading to increased revenue for content rights holders.

[0052] 4. In order to increase content sales using content AI utilization systems, more content rights holders will proactively provide books and other content to AI development and operation companies by using content AI utilization systems.

[0053] In this way, a positive spiral (virtuous cycle) occurs: 1→2→3→4→1.

[0054] Referring to Figure 2, the overall configuration of the content AI utilization system is explained. As an example of user terminals, a personal computer (hereinafter referred to as "PC") 54a, a smartphone 54b, and AR glasses 54c are connected to the internet 50. Furthermore, various AI groups 1, a centralized oracle 2, blockchains such as private chains 86, corporate servers 79 installed in a group of companies 81 (an example of an organization), a group of publishers 3, IPFS (InterPlanetary File System) 4, and a data center 44 where multiple cloud servers 51 are installed are connected to the internet 50, enabling information communication between them via the internet 50. IPFS 4 is a distributed file system and peer-to-peer (P2P) network intended for data storage and sharing. It is designed to complement blockchain technology and is widely used for storing digital media.

[0055] A private chain only records transactions using blockchain technology; the recording rights are not open and are monopolized by an individual or company, recording only internal transactions. Blockchain 86 is not limited to private chains; it can also be a consortium chain or a public chain. Blockchain 86 is connected to a centralized oracle 21. The centralized oracle 21 is a system that bridges data between the blockchain and the internet 1, connecting to the internet 50 to collect various information scattered across the network and providing it to the blockchain's smart contracts.

[0056] In addition to or instead of the centralized Oracle 21, a decentralized oracle managed in a distributed manner across the entire network may also be adopted. Information collected by multiple oracles distributed across the network is aggregated, an average of that information is extracted, and this average information is considered correct and incorporated into the blockchain for use in smart contracts. This is based on the theory proposed by James Surowiecki in his book "The Collective Opinion Is Often Right," which states that "by aggregating information within a group, the conclusion reached by that group can be better than the conclusion reached by any individual within the group." Furthermore, rewards such as tokens are given to oracles that collect information close to the average, thereby providing an incentive to operate the decentralized oracle.

[0057] Each node 99 of the blockchain 86 consists of a user terminal such as a PC 54a. This PC 54a is connected to the internet 50. Node 99b is a node of the content AI utilization system. In the group of companies 81, each company has a company server 79, as well as terminals 54d used by employees belonging to the company and a shared DB 80. These company servers 79, terminals 54d, and shared DB 80 are connected by a LAN (Local Area Network).

[0058] Each cloud server 51 in data center 44 is configured to communicate with the AI ​​registration DB 72, content DB 73, and content rights DB 74. The content DB 73 stores data for multiple book sets, multiple image sets, multiple video sets, multiple CGM (Consumer Generated Media) sets, and multiple music sets. The book data stored in the content rights DB 74 is in EPUB (Electronic Publication) format. EPUB is one of the standard file formats for ebooks, has the extension ".epub", and can store content such as text, images, and stylesheets. A key feature of EPUB is that it is reflowable, automatically adjusting the font size and layout according to the device being used, allowing reading on various screen sizes. It also supports DRM (Digital Rights Management) and is used for distributing copyrighted content.

[0059] Content rights DB74 records each content A, B, ... Z and their respective content rights holders 1, 2, ... N, in relation to each other. AI registration DB72 is a database that stores the IDs of AI groups registered in the content AI utilization system. These registered AI groups are permitted to use the content AI utilization system.

[0060] The hardware configurations of each user terminal 54a, 54b, 54c and terminal 54d will be explained with reference to Figure 3(A).

[0061] Each user terminal 54a, 54b, 54c, and terminal 54d is equipped with a CPU (Central Processing Unit) 10 as a control center. The CPU 10 consists of a RAM (Random Access Memory) 9 that functions as a work area, a ROM (Read Only Memory) 11 that stores data and programs, a storage unit such as an SSD (Solid State Drive) 12, an input operation unit 7 such as a display and keyboard, a communication unit 5, a display unit 6, a camera 71, a speaker 70, an interface 8, a bus 13, and various other hardware. In addition to the SSD, an HDD (Hard Disk Drive) may be used as the storage unit. Furthermore, in addition to the CPU, a GPU (Graphics Processing Unit) may be used as the control center.

[0062] The hardware configuration of the cloud server 51 and the enterprise server 79 will be explained with reference to Figure 3(B). Hardware similar to that of the user terminal shown in Figure 3(A) is given the same reference numbers. The cloud server 51 and the enterprise server 79 are equipped with a GPU 10g as a control center. It consists of RAM 9 which functions as the work area of ​​the GPU 10g, ROM 11 which stores data and programs, storage units such as an SSD 12, an operation unit 7 such as a display and keyboard, a communication unit 5, a display unit 6, a bus 13, and various other hardware. Note that an HDD may be used as storage in addition to or instead of the SSD. Furthermore, a CPU may be used as the control center in addition to or instead of the GPU.

[0063] In this embodiment, the server uses a general-purpose von Neumann-type computer, but a neural network processor (NNP) may also be used. The NNP chip is equipped with numerous "artificial neurons" modeled after real neurons, and each neuron interacts with the others in a network. Alternatively, a quantum computer employing a "quantum annealing method" may be used. This significantly reduces the time required for optimization calculations in machine learning.

[0064] The functions of the AI ​​platform system 82 will be explained based on Figure 4(A). This AI platform system 82 provides a training environment (a place for reinforcement learning, etc.) for the AI ​​groups by prompting multiple AI groups with prompts input by the user, having them compete in terms of the quality of their answers, and feeding the evaluation results back to each AI. In this embodiment, the AI ​​groups 86, 87, ... 89 are existing AIs registered in the AI ​​registration DB 72. However, from the perspective of the idea of ​​providing a training environment (a place for reinforcement learning, etc.) for the AI ​​groups, the system is not limited to such AIs.

[0065] A prompt entered by the user via the user terminal 54a is received by the prompt reception unit 83, and that prompt is input to each AI1, AI2, ...AIn. Each AI generates an answer to that prompt and sends it to the answer reply unit 84, and the answer reply unit 84 sends the answer back to the user terminal 54a. The display screen of that answer is shown in the upper screen of Figure 13. A hyperlink 999 to the relevant section of the book that the AI ​​used as the basis for its answer (h in Figure 13, "Introduction to Generating AI") is displayed, and if the user selects (clicks) this hyperlink, they will be taken to the purchase page of the book (see the lower screen of Figure 13). This hyperlink 999 is written in hyperlink format so that the CFI functions as a hyperlink within the EPUB. If the user purchases the book, the copyright issue (content rights issue) is resolved, and the relevant section of the book that the AI ​​used as the basis for its answer (Introduction to Generating AI) is displayed to the user.

[0066] Users who view the AI's response screen (see upper screen in Figure 13) perform an evaluation (for example, clicking the "Like icon 998" in Figure 13), and the evaluation result is received by the evaluation reception unit 85 and fed back to each AI1, AI2, ... AIn that provided the response. Each AI uses this evaluation as a reward to perform reinforcement learning. Reinforcement learning is a mechanism in which an agent placed in a certain environment acquires a policy π that maximizes the cumulative reward from the initial state to the goal, based on the reward given when it chooses an action. In reinforcement learning, learning progresses through the interaction between an agent, which is a type of AI, and the environment. When the agent performs a certain action a on the environment, the state s of that environment changes and reaches a certain target state, the agent is given a reward r. The agent learns a function that takes state s as input and outputs action a with the aim of maximizing this reward r.

[0067] Reinforcement learning progresses over time by repeating the following simple steps. 1. The agent receives an observation o from the environment (or directly the state s of the environment) and returns an action a to the environment based on the policy π. 2. Based on the action a received from the agent and the current state s, the environment changes to the next state s', and based on this transition, returns to the agent the next observation o' and a single number (scalar quantity) called reward r that indicates whether the previous action was good or bad. 3. Time progression: t ← t+1 Here, ← represents an assignment operation.

[0068] In this embodiment, "reinforcement learning" is a broad concept that includes deep reinforcement learning (DQN (Deep Q-Network)), which combines reinforcement learning and deep learning techniques.

[0069] In this embodiment, not only the evaluations but also the highly-rated responses are fed back to each of the other AI1, AI2, ...AIn under predetermined conditions, so that each AI can use them for machine learning. However, it is also possible to control the system so that highly-rated responses are not fed back to each of the other AI1, AI2, ...AIn.

[0070] Knowledge-sharing DB987 records successful experiences (achievements: including policies such as successful response patterns) when the AI ​​(agent) responds to user prompts (including requests such as task requests) and receives a high evaluation result. More details will be provided later.

[0071] In the case of videos on YouTube and other platforms, the following methods can be used to specify a particular section within a video: • YouTube URL with timestamp There is a way to add a timestamp to a YouTube URL. On YouTube, you can add a time (in seconds) in the format ?t=60s to the video URL, which will jump you to a specific point in the video. ·Web Video Text Tracks Format (WebVTT) WebVTT is a subtitle and annotation format for HTML5 video, but it also allows you to embed metadata and annotations at specific points within a video. Using WebVTT, you can add annotations and links to each timestamp, enabling more precise information linking. Media Fragments URI (MFURI) Media Fragments URIs are a standard proposed by the W3C, a URI scheme for identifying specific parts of audio and video files. For example, a range of 10 to 20 seconds can be specified using the format #t=10,20. Example: https: / / example.com / video.mp4#t=10,20 • Video control via custom scripts By using custom scripts such as JavaScript, you can leverage the video player's API to make more precise specifications. For example, you can seek through a video to a specific point in time or display annotations.

[0072] Based on Figure 4(B), the specific contents of the book collection recorded in Content DB73 will be explained. Purchasing the content is a common method of making the content available without infringing on content rights. In this Content AI utilization system, there are three forms of purchasing the books that constitute the content. <Form 1> This is a traditional and common purchase method in which a person purchases the entire book and makes all parts of the book available for reading and viewing, and is referred to as "entire book for human use". <Form 2> This purchase model does not allow humans to view (read) the entire book, but rather allows the AI ​​to provide any section of the book as evidence in response to user prompts, and the user can view the section that was provided. This is referred to as the "entire book for AI." <Form 3> A "partial purchase" model in which, if the AI ​​provides a specific section of a book as evidence in response to a user prompt, the user can purchase and view only that section.

[0073] Figure 4(B) shows the prices for these three purchase options, linked to the book title. For partial purchases, EPUB Canonical Fragment Identifiers (CFIs) that identify the purchase location are recorded. EPUB CFIs are a mechanism for specifying specific locations or ranges within an EPUB, and it is also possible to specify entire chapters or paragraphs using CFIs. The format for specifying with EPUBCFI is epubcfi( / pathname / chapter location! / location within chapter). For example, "epubcfi( / 6 / 4)" specifies the entire Chapter 6 of the EPUB. "epubcfi( / 6 / 4 / 2[chapter06]! / 4 / 2)" specifies the fourth paragraph (the entire paragraph element) of Chapter 6.

[0074] Based on Figure 5(A), the data recorded in the AI ​​registration DB72 will be explained. The AI ​​registration DB72 records the IDs of AIs registered in the content AI utilization system (e.g., $3gp#8, 9pk%7\, ...5&?m4#). As mentioned above, when a highly-rated response is shared with other AIs (AI1, AI2, ...AIn) under predetermined conditions, the consideration that the highly-rated AI requests when providing the content of the highly-rated response to other AIs, and the consideration that it pays when receiving the content of the highly-rated response from other AIs, are recorded in association with each AI's ID. In this embodiment, this consideration is paid in Ether, the cryptocurrency of Ethereum.

[0075] Based on Figure 5(B), the general mechanism of a Verifiable Credential will be explained. A Verifiable Credential is a verifiable qualification certificate, such as a digital certificate that proves that the holder possesses information such as academic / career certificates or a driver's license, and that this information has been verified by a reliable institution. In this embodiment, it is used to prove that the holder is a person who has purchased and is able to use content such as books (a user with the right to use the content). As explained based on Figure 4(B), there are three forms of book purchase, so there are also three types of Verifiable Credentials corresponding to each form.

[0076] Verifiable credentials always have an issuer. For example, a graduation certificate is issued by an educational institution, a driver's license by the public safety commission of each prefecture, and an employee training completion certificate is issued by the company that provided the training program. In this embodiment, the operator of the content AI utilization system is the issuer, but it is not limited to that; for example, it could be a content rights holder such as publisher 3.

[0077] The issuer uses a platform that allows the issuance of Verifiable Credential-compliant credentials (e.g., Blockcerts) to issue digital credentials as official certificates.

[0078] Next, the recipient (the holder in Figure 5(B)) stores the Verifiable Credential received from the issuer in their own registry and uses it as needed. For example, this could include job postings that only those with specific qualifications can apply for, logging into services, or accessing the owner page of an IoT device.

[0079] Verifiers can verify verifiable credentials sent by recipients, determine whether or not to provide the service, and change the service plan depending on the type of credentials.

[0080] It is desirable that Verifiable Credentials incorporate a Decentralized Identifier (DID), and that the combination of Verifiable Credentials and DID can achieve Self-Sovereign Identity (SSI).

[0081] Based on Figure 6, the data stored in SSD12 on terminal 54a as a blockchain node is explained. The data stored includes the user's private key SK, public key PK, shared key K1, the user's wallet address on the blockchain, smart contracts, EPUB applications, verifiable credentials, and blockchain data.

[0082] The private key SK and public key PK are a key pair used in PKI (Public Key Infrastructure), where data encrypted with the public key PK is decrypted with the private key SK. The private key SK is also used for digital signatures. The symmetric key K1 is a key used in symmetric-key cryptography such as DES (Data Encryption Standard) and AES (Advanced Encryption Standard). Data encrypted with the symmetric key K1 is decrypted using the same symmetric key K1. In this embodiment, a different symmetric key is used for each user.

[0083] Next, let's explain blockchain data. The data within each block of the blockchain includes the hash value of the previous block, multiple transaction data (also called transactions), and Verifiable Credential proof data. Transaction data includes the transaction ID, the wallet addresses of the seller (e.g., ○△× Publishing) and the buyer, the date and time of the transaction, the item sold (e.g., "Introduction to Human-Generated AI"), the selling price (e.g., 2,000 yen), etc. Furthermore, the digital signatures of the parties to the transaction are also recorded. A timestamp is also embedded in the blockchain.

[0084] The Verifiable Credential includes the following verification data: Certificate ID: 8%#?3!%6&$, issuer information (e.g., ○△× Publishing), buyer information (e.g., email address), purchased product information (e.g., Human-Generated AI Introduction), and proof of purchase (e.g., transaction ID, receipt number). Furthermore, the electronic signature and timestamp of the issuer of the Verifiable Credential are also recorded.

[0085] This data is encrypted using a shared key (e.g., K1) of the transaction parties (e.g., the buyer) and recorded on all nodes 99.

[0086] Therefore, since the transaction details cannot be deciphered by anyone other than the parties involved in the transaction (e.g., the buyer), there is the advantage of being able to protect personal information and privacy.

[0087] Such blockchains are generated and added as new blockchains when each node 99 performs blockchain processing (see S80, S204, S215, S219, S241, S253, S257, S361, S440, S465, etc. described later). Blockchain processing mainly consists of three phases: transaction, propagation, and recording.

[0088] The transaction phase refers to the actions generally known as "transactions," and includes legal acts such as buying and selling, transferring, and lending. More specifically, this transaction phase can be divided into three phases: generation, signing, and propagation.

[0089] The generation phase involves generating a transaction. For example, it might be decided to "sell the book 'Introduction to Generative AI' for 2,000 yen," and then digitally sign the generated transaction. This digital signature is created by passing the transaction data through a predetermined hash function to generate a hash value, which is then encrypted using the private keys (SK) of the parties to the transaction (seller (〇△× Publisher) and buyer). Alternatively, a digital public key certificate may be issued by a certification authority.

[0090] The propagation phase involves having node 99b of the content AI utilization system confirm that the transaction has been correctly generated and signed. If it is determined that the transaction was not correctly generated or signed, the transaction will be discarded.

[0091] The recording phase involves node 99b of the content AI utilization system recording the transaction once it is confirmed that the transaction has been successfully generated and signed. Transactions that have been successfully generated and signed are moved to a place called a pool. Subsequently, node 99b of the content AI utilization system selects transactions to record from the pool and performs the process of adding them to the blockchain.

[0092] Blockchain data is stored on all nodes 99 and 99b. Transaction data and Verifiable Credential data within the blockchain are encrypted and stored using the shared key of the user corresponding to each node 99 and 99b. Each transaction data A, B, C, etc., is prepared in the same number of copies as the number of transaction parties, and each is encrypted with the shared key of the transaction parties and recorded on the blockchain. Figure 6 shows the data of the buyer, one of the transaction parties in transaction F. This data is encrypted using the shared key K1. Data encrypted using the shared key K2 of the seller (e.g., ○△× Publishing), the other party in transaction F, is also recorded on the blockchain, but its illustration is omitted. This transaction F data encrypted with shared key K1 and transaction F data encrypted with shared key K2 are recorded as blockchain data on all nodes 99 and 99b. The Verifiable proof data is also encrypted with a shared key (e.g., K1) and recorded on all nodes 99 and 99b.

[0093] Therefore, since the transaction details cannot be deciphered by anyone other than the parties involved in the transaction, it has the advantage of protecting personal information and privacy.

[0094] Based on Figure 7, the flowchart of the main routine program for control processing in the content AI utilization system is explained. At the publisher's terminal, it is determined whether or not to register the book data in the content DB 73. If publisher 3 wishes to make the book data available for use in the content AI utilization system, step S (hereinafter simply referred to as "S") 1 determines it to be YES, and the book registration process is performed in S2. Upon receiving this, the cloud server 51 performs the book registration response process in S3.

[0095] When the AI ​​registers content in the AI ​​registration database 72 in order to use the content in the AI ​​content utilization system, S4 determines that the result is YES, and S5 performs the AI ​​registration process. Upon receiving this, the cloud server 51 performs the AI ​​registration response process in S6. S7 determines whether or not the AI ​​should access the content database 73, and if it determines that it should access it, S8 performs the access process. Upon receiving this, the cloud server 51 performs the access response process in S9. S10 executes a response response process to answer the user's prompt.

[0096] In user terminals 54a, 54b, 54c, and 54d (hereinafter simply referred to as "user terminal 54"), S11 determines whether or not the user will use the AI ​​in the AI ​​platform system 82. If the user will use the AI ​​in the AI ​​platform system 82, S11 determines it to be YES, and S12 performs the AI ​​usage process. Upon receiving this, the cloud server 51 performs the AI ​​usage response process in S13.

[0097] The flowcharts of the subroutine programs for the book registration process in S2 and the book registration support process in S3 are explained based on Figure 8(A). At the publisher's terminal 3, after the process of sending book data and content rights data to the cloud server 51 is performed in S18, the terminal returns and control is transferred to S1. Whether or not the book data and content rights data have been received is determined in S19 on the cloud server 51, and if they have not been received, the terminal returns and control is transferred to S6. As mentioned above, this book data is in the form of EPUB and includes data on the book title, purchase type (entire book for human use, entire book for AI use, purchase location in the case of partial purchase), and each price, as shown in Figure 4(B).

[0098] If the cloud server 51 determines that the data has been received by S19, the received book data is registered in the book area of ​​the content DB 73 by S20, and the received content rights data is registered in the content rights DB 74 by S21. This state is shown in Figure 2.

[0099] The flowcharts for the S5 AI registration process and the S6 AI registration support process subroutine programs are explained based on Figure 8(B). This AI registration process registers AIs that use the content AI utilization system in the AI ​​registration DB72. However, prior to registration, a contract based on mutual agreement must be concluded between the AI's operating company and the content rights holder, such as the operator of the content AI utilization system or a publisher. This contract includes a clause stating that "the AI ​​will present to the user specific information (hyperlinks, etc.) that identifies the content, such as books, that it referenced when providing services (such as responding to user prompts)." Therefore, all AIs registered in the AI ​​registration DB72 have this obligation to "present to the user specific information that identifies the content, such as books, that it referenced when providing services (such as responding to user prompts)." This contract is recorded on the blockchain.

[0100] In S24, the AI ​​generates the following registration details. AI ID • Compensation requested for providing highly-rated answers • The price paid for enjoying highly-rated responses Next, in S25, the registration details are sent to the cloud server 51 and then returned. The cloud server 51, having received the information in S26, registers the received registration details in the AI ​​registration DB 72 in S27 and then returns. The registered state is shown in Figure 5(A).

[0101] The flowcharts of the subroutine programs for access processing in S8 and access response processing in S9 are explained based on Figure 9(A). The AI ​​sends its own ID to the cloud server 51 via S31 to access it. The cloud server 51 determines whether or not access occurred via S32, and returns if there was no access. If access was determined to have occurred, S33 determines whether or not the received ID is registered in the AI ​​registration DB 72, and the result of that determination (access allowed or denied) is returned. If access is denied, it returns.

[0102] Upon receiving the reply, the AI ​​determines in S35 whether access is denied or not, and returns if access is denied. If access is permitted, S36 determines whether the purpose of this access is pre-training using content such as books. There are two purposes for the AI ​​to access the cloud server 51: to perform pre-training using content such as books, and to use RAG to search for necessary information from the content DB 73 when responding to user prompts. In the latter case, control moves to S41, but in the former case, S37 is used to request data registered in the content DB 73 from the cloud server 51.

[0103] The cloud server 51 determines whether or not a request has been made using S38. If there is no request, it returns a response. If there is a request, it provides the AI ​​with the data registered in the content DB 73 using S39. The AI ​​then pre-trains using the content data provided using S40.

[0104] When the AI ​​uses RAG to search for necessary information from content DB73 to respond to a user prompt, S36 determines NO, and S41 searches for the necessary information from content DB73 using RAG. The cloud server 51 receives the search request ω and allows the search of content DB73 in S42.

[0105] The operation process of RAG (Retrieval-Augmented Generation) is explained based on Figure 9(B). RAG is a framework that improves output accuracy, which is difficult to control with prompts alone, by combining text generation by AI such as LLM (Large-Scale Language Model) with the retrieval of highly reliable external information.

[0106] In RAG, before LLM93 generates an answer, external information (external data) such as the latest information and specialized field databases is added, and a step is added to allow searching of this information. This overcomes the weaknesses of LLM93 and enables clear evidence-based and highly accurate output. In this embodiment, a step is added to allow searching of content data such as books stored in the content DB73.

[0107] The system generates answers to user prompts through two phases: "search" and "generate." 1. Retrieval Phase In the search phase, data is collected by searching external information (stored information in Content DB73) in order to provide the best possible answer to the user's prompt.

[0108] (1) The user operates the user terminal 54 to input prompts to the chat AI 92, etc. (2) Chat AI92 searches Content DB73 and collects suitable data. (3) Obtain search result data 2. Generation Phase In the generation phase, LLM93 generates answers based on the data obtained in the search phase.

[0109] (4) The chat AI 92 inputs the prompt by combining the user's prompt with the search result data obtained in (3). (5) Based on the data entered in (4), LLM93 generates an answer and sends a reply to Chat AI92. (6) Chat AI 92 outputs the response obtained from LLM 93 to user terminal 54. Furthermore, the search methods used in RAG include the following:

[0110] Vector search: A method that captures the meaning of a word and finds related information. Keyword search: A method that matches patterns of words or strings and finds information based on similarity. However, vector search has drawbacks such as high development costs and keyword search slows down as the amount of data increases. Therefore, in most cases, implementing a "hybrid search" that compensates for the shortcomings of each is recommended.

[0111] The flowcharts of the subroutine programs for the AI ​​utilization processing in S12 and the AI ​​utilization support processing in S13 will be explained based on Figures 10 and 11. In the cloud server 51, S46 determines whether or not there has been access from a user, and if not, control is transferred to S66. If a user operates the user terminal 54 and accesses the cloud server 51, S46 of the cloud server 51 determines it to be YES, and user authentication using the received ID is performed by S47, and the result (access allowed or access denied) is returned. In the cloud server 51, if the authentication result is access denied, it returns.

[0112] On the user terminal 54, S49 determines whether the authentication result returned from the cloud server 51 is an access denial or not, and returns if access is denied. If access is permitted, the user enters a prompt into the user terminal 54 via S50 and sends it to the cloud server 51. The user can specify books or other content (reference content) that they would like the AI ​​to refer to when generating a response to the prompt.

[0113] Upon receiving a prompt entered by the user, the cloud server 51 sends the received prompt via S51 to the group of AIs registered in the AI ​​registration DB. Next, the cloud server 51 determines via S52 whether the waiting time for a response from the AI ​​has elapsed and waits until it has elapsed. Once the waiting time for a response has elapsed, control proceeds to S53, and each response sent from each AI is returned to the user terminal 54.

[0114] The user terminal 54, upon receiving the reply, displays each reply to the user via S54. An example of this display screen is shown in the upper screen of Figure 13. Next, in S55, it is determined whether the user will view the content that forms the basis of the answer. If the user selects (clicks) the hyperlink 999 (see upper screen of Figure 13) to the content that forms the basis of the answer, S55 determines it to be YES and control moves to S63. However, if the user does not select (click) it, S56 determines whether the user has performed a "like" operation. If the user has not performed a "like" operation, it returns. However, if the user selects (clicks) the "like" icon 998 (see upper screen of Figure 13), S56 determines it to be YES, and S57 sends the user's "like" operation result (high rating result) to the cloud server 51 and then returns.

[0115] Upon receiving the user's "like" operation result (high rating result), the cloud server 51 sends the high rating result to the AI ​​that performed the "like" operation (high rating) in S58. Next, the cloud server 51 processes the sharing of the highly rated responses in S59, processes the book ranking aggregation in S60, and then proceeds to S66.

[0116] If the user views the content that forms the basis of the answer (if the answer is YES in S65), control proceeds to S63, where a search process for the usage rights certificate data of the content to be viewed is executed. S64 determines whether or not search results certificate data is found, and if not, S75 determines whether or not to purchase the content that forms the basis of the answer, and if not to purchase, the system returns.

[0117] If user terminal 54 determines that verification data exists by S64, it selects the content it wants to view along with the verification data by S65 and sends it to cloud server 51. Upon receiving it, cloud server 51 determines YES by S66 and checks whether or not verification data has been sent by S67. If it has not been sent, it returns NG to user terminal 54 by S71. If verification data has been sent, it verifies the verification data (Verifiable Credential) by S68 and determines whether or not the verification result is valid by S69. If it is not valid, it returns NG by S71, but if it is valid, it returns the selected content data to user terminal 54 by S70.

[0118] The user terminal 54 uses S72 to determine whether or not a reply has been received from the cloud server 51. If a reply has been received, S72 determines it to be YES, and S73 determines whether or not the reply content is NG. If it is not NG, S74 displays the content data and returns. On the other hand, if it is NG, S73 determines it to be YES, and S75 determines whether or not to purchase the content that the user wants to view.

[0119] When a user performs an action to make a purchase, S76 redirects them to the purchase web page. An example of this purchase web page is shown in the lower screen of Figure 13. On the purchase web page, the prices for each of the three purchase types (full purchase for human users, partial purchase of AI response reference sections, and full purchase for AI users), as explained based on Figure 4(B), are displayed.

[0120] When a user selects a purchase method and completes the purchase, this information is sent to the cloud server 51. Upon receiving this information, S78 determines it to be YES, and S79 issues a Verifiable Credential to the user.

[0121] The user terminal 54 receives this information, and S80 processes it to record the Verifiable Credential on the blockchain. This state is shown in Figure 6.

[0122] Next, control moves to S65, where the Verifiable Credential (proof data) received in S79 is sent to the cloud server 51. As a result, the content data that was determined to be YES in S69 and selected is returned in S70, allowing the user to view the content data they wish to see.

[0123] The flowchart of the subroutine program for searching for the usage rights certificate data of the content to be viewed, performed by S63, is explained with reference to Figure 12. The user terminal 54 determines by S84 whether or not to use the certification data of its affiliated company, and if not, proceeds to S93. On the other hand, if the user performs an operation to use the electronic certificate (Verifiable Credential) of the organization to which they belong, S84 determines it to be YES, and S85 sends the content identification data to be viewed and the employee ID of the affiliated company to the company server 79.

[0124] Upon receiving it, the corporate server 79 verifies the received employee ID card using S86. S87 determines whether the verification result is valid or not, and if it is not valid, S88 returns an "NG" response. If it is valid, the corporate server 79 searches for the Verifiable Credential data for the content to be viewed using S89. If no certification data is found, S88 returns an "NG" response; however, if data is found, S91 returns the certification data (Verifiable Credential).

[0125] Upon receiving a reply from the corporate server 79, the user terminal 54 determines via S92 whether or not it received an NG. If it received an NG, it returns; otherwise, it searches for its own Verifiable Credential certificate data for the content it wants to view via S93 and then returns.

[0126] If the user terminal 54 receives the company's verification data transmitted via S91, it sends the company's verification data (Verifiable Credential) to the cloud server 51 via S65. If the user's own verification data (Verifiable Credential) is retrieved via S93, the user's own verification data (Verifiable Credential) is sent to the cloud server 51 via S65.

[0127] Figure 13 shows the screen above where the response to the user's prompt is displayed on the user terminal 54. In this example, the user asks the AI, "What is generative AI?", and the response displayed is, "Generative AI is a type of artificial intelligence that can generate content such as text and images. It is pre-trained using machine learning methods such as deep learning and reinforcement learning, and then fine-tuned."

[0128] The book "Introduction to Generative AI" and the relevant section that the AI ​​referenced when generating this answer are as follows: <a href=""hreepubcfi( / 6 / 4)"> Jump to Chapter 6 A hyperlink 999 is displayed. If the user selects (clicks) this hyperlink 999, S55 above determines that it is YES. If the user has purchased the book "Introduction to Generative AI" (hyperlink 999) as "The Whole for Humans," "The Whole for AI," or the AI ​​reference section "Chapter 6," then they possess the Verifiable Credential data, and S65 sends this certification data to the cloud server 51, causing the AI ​​reference section "Chapter 6" to be displayed (YES in S66 → YES in S67 → YES in S68 → YES in S69 → YES in S70 → YES in S72 → NO in S73 → S74). A Like icon 998 is also displayed, and if the user selects (clicks) this Like icon 998, S56 above determines that it is YES.

[0129] If the user does not possess the Verifiable Credential for the content of Chapter 6, the section referenced by the AI, control proceeds to S75, and if the user performs an operation to purchase the content, S76 transitions to the purchase web page. The displayed screen is shown in the lower part of Figure 13. The purchase web page displays the contents of the book to be purchased, "Introduction to Generative AI," and the prices for each of the three purchase types (full purchase for human use, partial purchase of AI response reference sections, and full purchase for AI) as explained based on Figure 4(B). When the user selects a purchase type and performs a purchase operation, this information is sent to the cloud server 51. When the cloud server 51 receives this, S78 determines it to be YES, and S79 issues a Verifiable Credential to the user.

[0130] The flowchart of the subroutine program for the response processing in S10 is explained based on Figure 14. The AI ​​determines in S96 whether or not it has received the prompt entered by the user, and returns if it has not received it. When the user enters a prompt via the user terminal 54, that prompt is sent to the AI ​​via the cloud server 51 (S50, S51), and if it is determined to be YES in S96, the AI ​​alliance formation process is performed in S97. This AI alliance formation process is a process in which multiple AIs (agents) form an alliance and cooperate with each other to answer the prompt (task processing request) entered by the user. In other words, a multi-agent system is built to answer. If the AI ​​answers alone without forming an AI alliance, control proceeds to S98.

[0131] In S98, it is determined whether the user has specified reference content (books, etc.). As explained in S50, the user can specify books or other content (reference content) that they want the AI ​​to refer to when generating an answer to the prompt. If no reference content is specified, control proceeds to S100, and the AI ​​generates an answer through the following process. 1. Understanding and Planning Prompts 2. Search Content DB73 using RAG as needed. 3. Utilize RAG and the information accumulated during the pre-training phase to search for relevant knowledge. 4. Organize the relevant information obtained and construct the information. 5. Check grammatical and logical consistency, generate answers, and add hyperlinks to referenced content. The "planning" described above involves breaking down user prompts into smaller units, organizing their relationships, and creating a "plan." Then, RAG is used to search the content DB73 as needed. By planning beforehand, proper searches become possible, preventing missed results.

[0132] The response also includes the presentation of content such as books that the AI ​​has referenced. The AI ​​sends the generated response back to the cloud server 51 via S103. This response is then sent back to the user terminal 54 via the cloud server 51 (YES in S52 → S53) and displayed to the user (S54).

[0133] On the other hand, if the user specifies reference content, S98 determines it to be YES, and S99 determines whether the AI ​​has pre-trained on that reference content. If it has pre-trained on it, control proceeds to S100; otherwise, it proceeds to S101, where, after understanding the prompt and planning, it uses RAG to search the content DB74 and obtain data on the reference content. Next, it proceeds to S102, where it generates a response through the following process. 1. Utilize information obtained during the pre-learning phase and through RAG to search for relevant knowledge. 2. Organize the relevant information obtained and construct the information. 3. Check grammatical and logical consistency, generate answers, and add hyperlinks to referenced content. Then, S103 sends the generated response to the cloud server 51.

[0134] Next, S104 determines whether or not a high rating result has been received. When the user clicks the like icon 998 (see upper screen in Figure 13), the high rating result of the like is sent to each AI (YES in S56 → S57 → S58), and S104 determines it to be YES. In addition, the highly rated response is shared among the AIs under predetermined conditions (S58).

[0135] In addition to or instead of the user evaluation described above, the evaluation of the AI's responses may be performed by constructing an evaluation AI and implementing it in the AI ​​platform system 82. This evaluation AI will perform the evaluation based on the following criteria. 1. Accuracy: Is the answer based on facts and does it accurately address the user's prompts? 2. Relevance: Does the answer align with the user's intent in the prompt, and does it contain irrelevant information? 3. Depth and comprehensiveness: Does the answer delve deeply into the issue and provide the necessary details and supplementary information? 4. Clarity: Is the answer clear and concise, and does it explain complex concepts in an easy-to-understand way? 5. Creativity and originality: How unique and innovative are the perspectives offered, especially regarding future predictions and idea generation?

[0136] Upon receiving the shared evaluation results and positive feedback, the AI ​​performs the following processing in S105. • Reinforcement learning based on high evaluation results • Supervised learning based on shared answers • Store successful experiences (achievements), including policy π, in the knowledge-sharing DB987. The reinforcement learning described above optimizes the response generation policy π (Policy) using techniques such as Proximal Policy Optimization (PPO). The supervised learning described above is performed, for example, using the following procedure. 1 Data Collection The dataset collects user prompts, the AI's own response A, shared highly-rated responses B, and evaluation results (indicating that highly-rated response B is superior).

[0137] • As an evaluation label, assign information indicating superiority or inferiority, such as "Answer B is better than Answer A." 2. Preparing the dataset The dataset is constructed using the user prompt, both response pairs, and the evaluation results of those responses (response B being the higher-rated option) as features. Training of 3 models The Learning to Rank approach optimizes the model to generate higher-ranking responses to prompts. For example, Pairwise Ranking or Listwise Ranking techniques are used to learn the relative importance of responses. Specific algorithms used include RankNet, LambdaRank, and LambdaMART.

[0138] The flowchart of the subroutine program for sharing highly-rated answers in S59 will be explained based on Figure 15(A). This highly-rated answer sharing process shares highly-rated answers (including multiple answers) that users have given a "like" to each AI's answer with other AIs under predetermined conditions. In this embodiment, highly-rated answers are shared among AIs when a match is found between the "amount of compensation requested for providing highly-rated answers" and the "amount of compensation to be paid for receiving highly-rated answers" shown in Figure 5(A), and it is determined that the predetermined conditions have been met.

[0139] Cloud server 51 searches for AIs highly rated by S108 from AI registration DB 72, and each AI found by S109 is designated as AI(N). N=1, 2, ...RR is the total number of AIs found. Cloud server 51 initializes N to 1 in S110, and sets C(N) as the value of the compensation requested by AI(N) for sharing high-rated responses in S111. In S112, it searches the AI ​​registration DB 72 for AIs where C(N) ≤ "compensation paid for sharing high-rated responses", shares the responses of AI(N), and moves C(N) from the wallet of the found AI to the wallet of AI(N).

[0140] Next, S113 increments N by one step. Then, S114 determines whether the value of N exceeds the "total number of AIs searched R" in S109. If it is determined that it does not exceed the limit, control is transferred to S111, and the loop S111→S112→S113→S114→S111 is repeated. When the value of N exceeds the "total number of AIs searched R" in S109, S114 determines YES, and control proceeds to S115, the value of N is cleared, and the program returns.

[0141] The sharing transaction of highly-rated responses, which involves the transfer of this consideration (cryptocurrency Ether), may be automated using a smart contract.

[0142] A smart contract is a system that, based on a contract, pre-determines the terms of a transaction on the blockchain using a program, and automatically executes the verification and fulfillment of those terms. The process of executing a smart contract on the blockchain consists of four steps: contract definition → waiting for events → contract execution and value exchange → payment and settlement.

[0143] Contract definition is the step of defining contract terms in a program. Event waiting is the step of monitoring the trends of assets or information that are the subject of the contract over a certain period of time. Contract execution and value exchange is the step of triggering a predetermined event and executing processing according to the contract terms when predefined conditions are met during event waiting. Payment and settlement is the step of making payments and settlements for monetary value etc. associated with contract execution and value exchange.

[0144] Figure 18 illustrates a flowchart demonstrating an example of an automated transaction using a smart contract to share highly-rated responses, involving the transfer of consideration (cryptocurrency Ether). In S350, the highly-rated AI programs the compensation it requests for providing highly-rated responses and the compensation it pays for receiving those highly-rated responses. Similarly, in S351, the other AI groups also program the compensation they request for providing highly-rated responses and the compensation they pay for receiving those highly-rated responses.

[0145] Next, the highly-rated AI identifies other AIs from the AI ​​registration DB72 via S352 and checks the conditions of the opposing AI via S353. The opposing AI also checks the conditions of the opposing AI (the highly-rated AI) via S354 and returns if it determines that the conditions do not match. If the highly-rated AI also determines via S355 that the conditions do not match, control proceeds to S360.

[0146] On the other hand, if S355 determines that the conditions are met, S357 sends a notification of agreement to the other AI. Upon receiving it, the AI ​​executes the contract in S358 and, in S359, moves the consideration requested by the other AI from its own wallet to the other AI's wallet. Then, in S361, it records the transaction on the blockchain and returns it.

[0147] The highly-rated AI determines in S360 whether it has finished matching with all other AIs. If not, it returns to S352 and identifies another AI from the AI ​​registration DB72. The highly-rated AI goes through this loop of S360→S352→353→S355→S357→S360→352 and returns when S360 determines YES. Each time this loop is completed, the opponent AI identified in S352 repeats the process from S354 onwards.

[0148] The flowchart of the subroutine program for the book ranking aggregation process in S60 will be explained based on Figure 15(B). This book ranking aggregation process aggregates the number of times a book was highly rated for books that were presented as reference when the AI ​​made its answer. In S119, the cloud server 51 sets the score of book J used in the highly rated answer as J(i). i=1, 2, ...KK is the total number of books This "Book J used in highly rated responses" includes both parts (chapters, etc.) and the entire book presented in the highly rated responses. Therefore, the "displaying a list of J(i) for all J" in S124, described below, displays not only the ranking of an entire book but also the ranking of parts of a book. Since this ranking data is provided to publisher group 3 (see S122), publishers can determine not only whether there is a need to publish a revised edition of an entire book, but also whether there is a need to publish a revised edition of a part (chapter, etc.) of a book. Therefore, publishers can choose to publish a revised edition of only the part (chapter, etc.) that has a high ranking in the highly rated section of a particular book.

[0149] Next, S120 identifies the books J that received high ratings in this response and adds 1 to J(i). S121 determines whether or not to provide the aggregated results to the publisher, and if not yet provided, S123 determines whether or not to publish the aggregated results, and if not to publish, returns.

[0150] The timing for providing the aggregated results to publishers is predetermined, such as monthly or annually. When the time for provision arrives, S121 determines that the result is YES, and S122 provides each publisher with the J(i) of the books they have published. If S123 determines that the aggregated results should be published, S124 displays a list of J(i) for all J's and publishes them. Next, S125 returns after the user-specific book ranking process is completed.

[0151] The flowchart of the S125 user-specific book ranking processing subroutine program is explained based on Figure 16. This user-specific book ranking processing aggregates and displays the ranking of books used in responses that users have given high ratings (likes) to each user. The cloud server 51, via S400, will have users represented as U(n) with n=1, 2, ...R R is the total number of users Let J(i) be the score of book J used in the highly-rated responses for U(n). i = 1, 2, ..., K K is the total number of books Next, S401 determines whether J(i) of book J, which received a high rating in the response to U(n), is already linked to U(n) and stored. If it is already stored, S403 adds 1 to the stored J(i). On the other hand, if S401 determines NO, control moves to S403, and J(i) of book J, which received a high rating in the response to U(n), is added by 1 and stored linked to U(n).

[0152] Next, S404 determines whether or not to publish the aggregated results to the user. The aggregated results are published to the user periodically (for example, every month). If the results are not yet to be published, a return is made. If the results are to be published, S405 sets the initial value of n to "1", and S406 sends the list display data of books J that received high ratings in the response to U(n) to the user terminal 54 of U(n). Upon receiving this, the user terminal 54 displays the list display data via S407. This display screen is shown in Figure 17.

[0153] Next, cloud server 51 advances n by one step using S408, and determines whether n has exceeded R using S409. If n has not yet exceeded R, control returns to S406. The loop S406→S408→S409→S406 is repeated. When n exceeds R, S409 determines YES and returns.

[0154] The list display screen by S407 will be explained based on Figure 17. In the user terminal 54 shown in the upper screen of Figure 17, a ranking compiled every month is displayed, with the title "Ranking of books you highly rated this month!". The titles of the books and the scores received for the top 5 highly rated responses by the user are displayed in a list. If a user sees a book they have not yet purchased and clicks on it, for example, "Introduction to Generative AI" which is ranked 1st, they will be taken to the purchase page for that book. That purchase page is shown in the lower screen of Figure 17. This purchase page is the same as the lower screen of Figure 13 mentioned above, so the explanation of the response will be omitted here.

[0155] When the program-defined content is updated using S350 and S351 as described above, the previously defined smart contract needs to be updated to the new version. One way to do this is to update the smart contract using the Proxy pattern. This is a configuration pattern in which a proxy contract exists between the main contract and the user. Referring to Figure 33(A), the user accesses the proxy contract 120, and the proxy contract 120 forwards the transaction to the main implementation contract containing the logic. In this implementation pattern, the proxy contract 120 remains unchanged and always has the same address, but the logic contract referenced by the proxy contract can be changed to other implementation contracts 121a, 121b, and 121c, so the contract as a whole can be considered an upgradeable contract.

[0156] Next, the data recorded in the knowledge-sharing DB 987 will be explained based on Figure 19. In this embodiment, the AI ​​(agent) records highly-rated responses as successful experiences (including policy π) in the knowledge-sharing DB 987, and makes them shareable with other AIs (agents) under predetermined conditions. For example, if an agent finds the optimal behavioral pattern for a particular task, that pattern is registered in the knowledge-sharing DB 987, and other agents can refer to it under predetermined conditions when working on similar tasks. Specific examples of the data stored in this knowledge-sharing DB 987 are listed below. <Specific Example 1> Task content, division of roles, and assigned roles For example, when an AI federation (multi-agent) responds, the breakdown of roles (see S316) and the role played by the AI ​​in question (AI with ID$3gp#8) are recorded. This role includes a leader who manages and oversees the entire multi-agent system, and this leader agent determines the roles of the other agents. <Specific Example 2> ID of an AI registered with the federation When an AI alliance (multi-agent) responds, the IDs of the registered AIs in the alliance are recorded. <Specific Example 3> Successful Planning This planning includes both overall planning for the entire multi-agent alliance and planning for each individual agent responsible for their assigned role. <Specific Example 4> Successful Behavioral Patterns and Their Reasons For example, if Agent A selects a specific behavioral pattern in a task and succeeds, that behavioral pattern (e.g., the sequence of decisions made or how resources were allocated) and the reasons and background for its success are recorded. <Specific Example 5> Errors and their solutions For example, when an agent encounters an error, the cause and solution are recorded. This information might include details such as, "Part of the task failed due to insufficient resources, but this was resolved by allocating additional resources." <Specific Example 6> Task Completion Conditions and Evaluation Criteria For example, the conditions, evaluation criteria, and expected outcomes necessary for achieving each task are specifically recorded. These might include, for instance, "quality evaluation criteria" and "delivery requirements." <Specific Example 7> Resource Usage History and Optimal Allocation Methods For example, the amount and allocation of resources (time, computing power, memory, etc.) used to successfully complete tasks in the past are recorded. <Specific Example 8> Decision-Making Process and its Results For example, information about the process by which the agent made a decision (such as how they evaluated and prioritized options) and whether the result was successful is recorded. <Specific Example 9> Dependencies between tasks and how to manage them For example, if there are dependencies between tasks, the method for managing those dependencies is recorded. This could include information such as "Task B should be started after Task A is completed" or "Synchronization between tasks is necessary." <Specific Example 10> Environment variables and corresponding adjustment methods For example, the conditions under which a system succeeded in a specific environment, and the adjustments made to adapt to that environment, are recorded. For instance, "the best way to handle network latency" is recorded. <Specific Example 11> History and Results of Similar Tasks For example, a history of similar tasks performed in the past, whether those tasks were successful, and any improvements made are recorded. <Specific Example 12> Policy π The policy π (Policy) that the AI ​​(agent) has learned through reinforcement learning is recorded.

[0157] This policy π records the policy π learned by each AI (agent) by fulfilling its assigned role when multiple AIs (agents) form a coalition and divide roles among themselves to process a task. By recording "task role division and assigned roles" in Specific Example 1 and "IDs of registered coalition AIs" in Specific Example 2, it is possible to search for and collect information on the role and policy π of each registered coalition AI.

[0158] The specific details of this policy π (Policy) are listed below. 1. Information about the state Content: Information representing the current state of the environment. 2. Information regarding Action Content: Actions that the agent can choose within the environment 3. Policy Function Content: A function that defines a probability distribution for selecting an action based on the state. 4. Information about rewards Content: Rewards obtained by agents as a result of taking specific actions. 5. State - Action Value (Q Value) Content: The cumulative reward expected when a specific action is chosen under specific conditions. 6.Discount Factor (γ) Content: A value that determines how much weight is given to future rewards as part of present rewards. 7. Search History Contents: Records of past states visited, actions taken, and rewards received. 8. Policy Parameters Contents: Parameters for defining policy functions and value functions 9. Rules for updating policies Content: Algorithms and rules for improving policies This knowledge-sharing DB987 may be accessible and referenced by all AI1, AI2, ..., AIn. However, if a certain AI1 records a successful experience (including a strategy as an optimal behavioral pattern) in the knowledge-sharing DB987, only AIs authorized by that AI1 (e.g., AI2) may be allowed to access the knowledge-sharing DB987 and reference that successful experience (including a strategy as an optimal behavioral pattern). AI1, AI2, ..., AIn participating in the AI ​​platform system 82 can respond to user prompts (including task requests) by forming an AI alliance of multiple AIs as a multi-agent, as described above (see S97). In other words, an appropriate multi-agent is constructed according to the content of the task request in the prompt entered by the user, and that multi-agent collaborates to provide the optimal response. At that time, it may be possible to control the access of multiple AIs (agents) constituting the multi-agent so that they can share each other's successful experiences (including strategies) in the knowledge-sharing DB987.

[0159] Furthermore, the above specific examples 1 and 2 may be made unconditionally accessible and referential to all AI1, AI2, ..., AIn. The information in specific examples 1 and 2 can be used as reference when forming the aforementioned AI alliance. The AI ​​accesses the knowledge-sharing DB987, refers to specific examples 1 and 2, selects an AI suitable for the prompt (including task requests) entered by the user, and forms an AI alliance. The information in specific examples 1 and 2 is also useful when assigning roles within the formed AI alliance. The "AI alliance formation process" and "AI alliance response process" described later will utilize the information in specific examples 1 and 2.

[0160] Furthermore, as a means of sharing successful experiences (including strategies) among AIs (agents), the following means may be used instead of or in addition to the above-mentioned knowledge sharing DB987. <Alternative Method 1> Policy Propagation This method involves saving an agent's behavioral patterns as a "policy (π)" and propagating it to other agents. In this method, the learned policies of successful agents are copied to other agents, accelerating the overall learning process. This policy propagation is sometimes called "policy distillation" or "policy propagation." <Alternative Method 2> Transfer Learning By utilizing "transfer learning" between agents, successful experiences (including strategies) gained by one agent can be applied to other agents. Transfer learning reduces the need for new agents to learn from scratch by adapting knowledge gained in a specific environment or task so that it can be used in other different environments. <Alternative Method 3> Imitation Learning This method involves other agents learning by "imitating" the actions of successful agents. Using imitation learning, other agents can directly learn the actions of successful agents, efficiently sharing successful experiences (including strategies). This method is particularly effective for tasks requiring consistent behavior. <Alternative Method 4> Utilization of Metalearning By utilizing "meta-learning," which enables agents to quickly adapt to new situations, the sharing of successful experiences (including strategies) becomes more effective. Through meta-learning, agents extract common "learning strategies" from the successful experiences (including strategies) of other agents, making it easier for them to adapt to different tasks and environments. <Alternative method 5> Induction by a shared reward structure It is also effective to introduce a "shared reward structure" in which all agents receive rewards based on the shared success stories (including strategies) of each agent, if those stories are useful to the whole. This promotes cooperation among agents and encourages greater utilization of shared success stories (including strategies).

[0161] The flowchart of the S97 AI alliance formation process subroutine program is explained based on Figure 20. One of the multiple AIs registered in the AI ​​registration DB72 understands the prompt in S299 and searches for an AI suitable for alliance using data 1 and 2 in the knowledge sharing DB987. Based on the search results, S300 determines whether or not to send an AI alliance offer to other AIs. If it is determined not to send an offer, S302 determines whether or not an AI alliance offer has been received from other AIs. If not, it returns. Also, even if an AI alliance offer has been received, if S303 determines that alliance will not be formed, it returns. As a result, control moves to S98 and the AI ​​responds individually. If S303 determines that alliance will be formed, S304 sends an AI alliance acceptance reply to the AI ​​that sent the offer, and then control moves to S308.

[0162] On the other hand, if S300 determines that an AI alliance offer should be made to other AIs, S301 sends the AI ​​alliance offer to the other AIs, and S305 determines whether or not a reply accepting the AI ​​alliance has been received from the other AIs. If no reply has been received yet, S307 determines whether a predetermined time (e.g., 10 seconds) has elapsed, and if it has not yet elapsed, control returns to S305. During this S305→S307→S305 loop, each time S305 determines YES, S306 performs the process of registering the accepted AI as part of the alliance.

[0163] Once the predetermined time has elapsed, S307 determines that the result is YES, and control proceeds to S308, where the AIs registered in the alliance share their respective success experiences (including strategies) stored in the knowledge sharing DB987. Next, S309 performs the AI ​​alliance's response processing.

[0164] The flowchart of the subroutine program for response processing by the AI ​​alliance is explained based on Figure 21. Each AI (multi-agent) that forms the alliance understands the prompt (including the request for task processing) entered by the user in S315. Next, in S316, they refer to data 1 and 2 in the knowledge sharing DB987 to determine the division of roles for response processing among the registered AIs in the alliance. In S317, it is determined whether or not there was a specification of content (books, etc.) that the AI ​​should refer to when providing the answer. If not, the process proceeds to S318, where each registered AI in the alliance generates an answer according to its assigned role through the following process. 1. Planning and, if necessary, searching the content DB73 using RAG. 2. Utilize RAG and the information accumulated during the pre-training phase to search for relevant knowledge. 2. Organize the relevant information obtained and construct the information. 3. Check grammatical and logical consistency, generate answers, and add hyperlinks to referenced content. Next, in S319, the responses from the federated AIs are integrated and sent back to the cloud server 51, after which control is transferred to S104. As a result, when each AI in the federation (multi-agent) responds, those AIs do not perform individual responses (S98-S103). Only AIs that are not in the federation perform individual responses (S98-S103). This response integration process in S319, when responding to a user's task processing request, collects and integrates the results of the work performed by each agent, who has been assigned multiple roles, to create a cohesive overall deliverable.

[0165] If S317 determines that the answer is YES, S320 determines whether the AI ​​has pre-trained on the content specified by the user. If the AI ​​has pre-trained on the content, control proceeds to S318. If the AI ​​has not pre-trained on the content, S321 performs planning, then uses RAG to search the content DB73 to obtain data on the specified content, and then proceeds to S322 to generate an answer through the following process. 1. Utilize information accumulated during the pre-learning phase and within RAG to search for relevant knowledge. 2. Organize the relevant information obtained and construct the information. 3. Check grammatical and logical consistency, generate response answers, and create hyperlinks to referenced content. Then proceed to S319.

[0166] In this embodiment, Ether, the cryptocurrency of Ethereum, is used as an example of compensation paid for sharing highly-rated responses in S112. However, it is not limited to this, and a stablecoin as a fungible token may also be used. Alternatively, the AI ​​platform system 82 may be considered as a decentralized autonomous organization (DAO) and social tokens may be used. In this case, the members of the decentralized autonomous organization can be considered as the group of AIs registered in the AI ​​registration DB 72. These social tokens can be exchanged for other cryptocurrencies (e.g., Ether) or for special rights within the community of the decentralized autonomous organization. Furthermore, the AI ​​itself may issue individual social tokens (personal tokens).

[0167] Furthermore, AIs registered in the AI ​​registration DB72 may issue security tokens (digital securities). These security tokens are tokenized securities and are subject to the same legal regulations as securities. In Japan, security tokens were positioned as "electronically recorded transferable rights," a new category introduced by the revised Financial Instruments and Exchange Act which came into effect in May 2020. This revision has made it possible for financial institutions to handle them. When a user purchases these security tokens, it may be controlled so that dividends are paid to the user (hereinafter referred to as the "digital securities holder") according to the dividend yield. For example, the sharing process of highly rated answers (S108~S115) can be considered an economic transaction by the AI ​​(agent), and if the profit obtained by subtracting the consideration paid by the AI ​​from the consideration obtained by the AI ​​through that economic transaction exceeds a certain amount, dividends may be paid to the digital securities holder (security token holder) according to the dividend yield. Another example of an economic transaction is that all or part of the AI ​​success stories accumulated in the knowledge sharing DB987 may be sold to other AIs.

[0168] Thus, digital securities holders can secure the possibility of receiving dividends by holding security tokens, but they may also create a secondary market where they can buy and sell the security tokens they hold. Figures 38(A) and 38(B), described later, are relevant to this secondary market.

[0169] The issuance of security tokens (digital securities) by AI, as described above, will be described later as a sixth embodiment.

[0170] Next, a modified version of the AI ​​platform system will be described based on Figures 22(A) and 22(B). In the aforementioned AI platform system 82, the prompt entered by the user is sent to all AIs, but in this modified version of the AI ​​platform system, the prompt entered by the user is sent to only a limited number of AIs. Specifically, when the AI ​​platform system receives a prompt entered by the user, the AI ​​platform system selects an AI suitable for that prompt from the AI ​​registration DB 72.

[0171] As shown in Figure 22(B), the AI ​​registration DB72 records not only the fees required for providing high-rated responses and the fees paid for receiving high-rated responses, but also the areas of expertise of each AI, associated with each AI's ID. The AI ​​platform system uses these areas of expertise to select an AI suitable for the prompt and sends the prompt to the selected AI98. In addition, the AI ​​platform system selects useful books or other content from the content DB74 that are suitable for the prompt entered by the user and sends that content to AI98.

[0172] AI98 generates a response to a prompt by referring to the transmitted content (books, etc.). The AI ​​platform system sends this response back to the user terminal 54. If the user sees the response and selects (clicks) the "like" icon 998, the AI ​​platform system receives the positive rating and sends it to AI98.

[0173] Figure 23 is a flowchart of the subroutine programs for AI utilization processing and AI utilization response processing in a modified AI platform system. The same steps as those shown in Figure 10 are given the same step number (S number). Here, the explanation of the same steps is omitted. In S135, the cloud server 51 performs processing to identify a suitable AI and a useful book in response to the received prompt. In S136, the process of sending the identified book data and prompt to the identified AI is performed.

[0174] The flowchart of the subroutine program for identifying a suitable AI and useful books in response to the prompt received in S135 is explained based on Figure 24(A). In S148, the cloud server 51 identifies a suitable AI by referring to the AI's area of ​​expertise in the AI ​​registration DB 72 based on the prompt content. Next, in S149, it is determined whether the user has specified reference content (reference books, etc.). If the user has specified content (books, etc.) that they want the AI ​​to refer to when responding, in S150 the specified content (books, etc.) is designated as useful content.

[0175] On the other hand, if the user does not specify any reference content (such as reference books), S149 determines the result to be NO, and S151 performs a process to identify useful content by referring to the content group in Content DB73 based on the prompt content.

[0176] The flowchart of the AI ​​response processing subroutine program in the modified AI platform system is explained based on Figure 24(B). The same steps as those shown in Figure 14 are given the same step number (S number). Here, the explanation of the same steps is omitted. At S154, the AI ​​determines whether or not it has received the prompt and content data (book data, etc.). If it has not received them, it returns, but if it has received them, it proceeds to S155 and generates a response through the following process. 1. Understanding and planning prompts 2. Search for relevant knowledge using pre-learning phase and received book data. 3. Organize the relevant information obtained and construct the information. 4. Check grammatical and logical consistency, generate answers, and add hyperlinks to referenced content. The AI ​​uses S158 to perform reinforcement learning based on high-evaluation results and supervised learning based on shared responses.

[0177] In the first embodiment described above, an existing AI was used in the content AI utilization system. However, a dedicated AI may be created specifically for the content AI utilization system and used. When using this dedicated AI, book data of publications from each publisher is provided and used to train the specialized AI. If all book data is trained on the specialized AI, the content DB73 becomes unnecessary. [Second Embodiment] Next, a second embodiment will be described. This second embodiment follows the first embodiment described above, and involves selling and distributing books on which notes, comments, annotations, etc., have been written by scholars, entertainers, talents, athletes, and other famous people.

[0178] Referring to Figure 25, if the publisher wants someone to write in a book before publication, they obtain that person's consent and send the book data to that person's user terminal 54 (S163). The user terminal 54 launches the EPUB application and receives and stores the book data via S164. The user terminal 54 determines whether or not it is currently reading via S165, and proceeds to S177 if it is not currently reading. If it is a reader, S165 determines it to be YES, and S166 determines whether or not the user has written in the book.

[0179] If no write operation is performed, the process proceeds to S170. However, if a write operation is performed, S167 generates a CFI (Canonical Fragment Identifier) ​​for the written location in the book and sends the write to the cloud server 51. The cloud server 51 determines in S168 whether or not it has received it, and if it has not, the process proceeds to S181. If it has been received, S168 determines it to be YES, and S169 performs the process of associating the received write content with the CFI and storing it in the CFIDB 100. An example of the data stored in the CFIDB 100 is shown in Figure 26(B).

[0180] Next, user terminal 54 determines in S170 whether the user is currently reading the location identified by the CFI. If it determines NO, it proceeds to S177, but if it determines YES, it proceeds to S171 and sends the CFI for the relevant location to cloud server 51. Cloud server 51 determines in S172 whether it has received it. If it has not received it, it proceeds to S181, but if it has received it, it determines YES in S172 and proceeds to S173, and sends the written content associated with the received CFI back to user terminal 54.

[0181] The user terminal 54 that receives it displays the returned written content to the user via S174. For example, if the SFI sent to the cloud server 51 via S171 is "epubcfi( / 6 / 4 / 2[chapter06]! / 4 / 2 / 1:0", the cloud server 51 searches the CFIDB 100 and reads the written content stored in association with "epubcfi( / 6 / 4 / 2[chapter06]! / 4 / 2 / 1:0" and replies with "The centralized management system incurs high management costs and is not commercially viable. The work of the centralized provider will be taken over by the user group." (S173). The user terminal 54 that receives it displays "The centralized management system incurs high management costs and is not commercially viable. The work of the centralized provider will be taken over by the user group." to the user (S174).

[0182] The user terminal 54 determines whether all writing is complete via S177. If it is not complete, it returns; however, if it determines that it is complete, it sends the completed book data back to the publisher's terminal via S178. The publisher's terminal, upon receiving this via S179, performs book registration processing via S180, and the cloud server 51 responds accordingly with book registration processing via S181. These book registration processing via S180 and book registration response processing via S181 are the same processes as the book registration processing via S2 and book registration response processing via S3 described above. The copyright holder (content rights holder) of the book on which this writing has been done is both the author of the book and the person who performed the writing. Royalties from the sales of the book published and sold by the publisher are awarded not only to the author of the book but also to the person who performed the writing. [Third Embodiment] Next, a third embodiment will be described. As shown in Figure 27, this third embodiment is a system in which LLMs 103a, 103b, ... 103n are sold and distributed as a set, corresponding to each book 102a, 102b, ... 102n. In this third embodiment, "books" is a broad concept that includes not only ebooks but also paper books. Each LLM 103a, 103b, ... 103n has learned the corresponding book 102a, 102b, ... 102n. For example, a user who purchases book 102a can also obtain LLM 103a that has learned book 102a, and if the user inputs a prompt into LLM 103a, LLM 103a will use the knowledge it has learned from book 102a to generate an answer and present it to the user.

[0183] While there is generally a one-to-one correspondence between books and LLMs, there are exceptions where one LLM corresponds to multiple books. For example, if a book is divided into multiple parts, such as Part 1 and Part 2, one LLM will learn all of those parts.

[0184] Users who purchase each of the books 102a, 102b, ...102n will be granted Verifying Credentials 104a, 104b, ...104n, which prove that they have the right to use the respective purchased books.

[0185] When the user inputs the same prompt A to each LLM103a, 103b, ..., 103n, each LLM103a, 103b, ..., 103n generates an answer using the knowledge from the books it is studying. However, each of these answers is based on the knowledge from a single corresponding book and is not an integrated answer combining the knowledge from multiple books. In the third embodiment, a metamodel 105 is used to integrate the answers of each LLM103a, 103b, ..., 103n through tasking (stacking) to generate and present an integrated answer to prompt A to the user.

[0186] Stacking is a technique that combines the outputs of multiple models to produce a response, and its applications are wide-ranging, applicable to various tasks such as classification, generation, and natural language understanding. For example, it can be used in text summarization, text sentiment analysis, and prompt response systems. Stacking integrates the strengths of multiple models.

[0187] Referring to Figure 28, the process for generating a response when a user enters a prompt will be explained. User terminal 54 determines in S185 whether or not the user has entered a prompt. If it determines that a prompt has been entered, it proceeds to S186, where it specifies the AI(LLM) to which the prompt was entered, and sends the Verifiable Credential, which proves the right to use the book that the specified AI(LLM) is learning, along with the prompt, to each specified AI(LLM). The AI(LLM) to which the user specifies may be multiple LLMs 103a, 103b, ..., 103n, as shown in Figure 27. In this third embodiment, since the AI(LLM) and the book that the AI(LLM) is learning correspond (see Figure 27), the user specifies the book by specifying the AI(LLM) in S186.

[0188] Each AI (LLM) that receives it in S187 verifies the received Verifiable Credential in S188.

[0189] S189 determines whether the verification result is valid, and if it is not valid, S190 returns an "NG" to the user terminal 54. If the user terminal 54 receives an "NG" return via S197, it displays an "NG" message via S191 and returns. The reason for verifying the Verifiable Credential even when a user who has purchased a book and LLM pair enters a prompt is to prevent the act of separating the LLM from the book and transferring only the LLM. For example, if person X obtains only LLM 103a by separating it from book 102a and enters a prompt into LLM 103a, and LLM 103a answers using the knowledge learned from book 102a, then X could obtain an answer based on book 102a (including a reference to the relevant section of the book) even though X has not purchased book 102a and does not have the right to use book 102a. The Verifiable Credential is verified to avoid such an absurdity.

[0190] Meanwhile, each AI (LLM) designated by the user proceeds to S192 if it is deemed appropriate in S189, and generates an answer through the following process. 1. Understanding and planning prompts 2. Utilize information gained from prior learning and books to search for relevant knowledge. 3. Organize the relevant information obtained and construct the information. 4. Check grammatical and logical consistency, generate answers, and add hyperlinks to referenced content. Each AI (LLM) then sends its generated response to the cloud server 51 via S193. The cloud server 51 receives the transmission via S194 and, via S195, inputs the output of each AI (LLM) into the metamodel 105 to generate an integrated response (including the location of the referenced book). Then, via S196, it sends the integrated response back to the user. The user terminal 54, which receives this response via S197, displays the received response.

[0191] In addition to stacking, weighted ensembles are also possible for integrating the responses of each LLM103a, 103b, ..., 103n. A weighted ensemble integrates responses by assigning weights to the outputs of multiple models and averaging them. The weights are adjusted according to the accuracy and confidence level of each model. Weighted ensembles may be used in addition to or instead of stacking. [Fourth Embodiment] Next, the fourth embodiment will be described. This fourth embodiment follows the first, second, and third embodiments described above and relates to an Nth-order distribution system (where N is an integer of 2 or more) that enables purchasers of books and other content to resell said content to others. In this Nth-order distribution system, books and other content are distributed as NFTs. NFT stands for Non-Fungible Token, which in Japanese means a token that cannot be replaced.

[0192] In this embodiment, an off-chain method is adopted, where content is recorded outside the blockchain, rather than an on-chain method, where content is recorded directly on the blockchain. The reason is that digital content is extremely large compared to the size of transactions treated as "tokens" on the blockchain, making it difficult to store it directly on the blockchain. Furthermore, if content is stored on a private server, there is a risk that it may become inaccessible from the internet at any time, or that it may be hacked and the content destroyed. Therefore, in this embodiment, the distributed file system IPFS (InterPlanet File System) is used.

[0193] Referring to Figure 29, upload an example of book data to file server 106. Record metadata containing information such as the URL, title, and description of the uploaded content in JSON format and upload it to IPFS.

[0194] Next, we issue an NFT on Ethereum and record it on the blockchain. An example of the information to be recorded is shown below. Transaction ID • Token ID • Seller's and buyer's wallet addresses • Purchase date • Royalty payments to content rights holders (royalty rate, content rights holder's wallet address, etc.) • Hyperlink to Chapter 6 of Book Data 2 (Hash value for Chapter 6 of Book Data 2: Qmuql2xv) "Transaction ID" is a unique identifier assigned to a transaction. In the case of Figure 29, it is the identifier assigned to resale transaction 1 of the NFT-enabled content (Chapter 6 of Book Data 2). "Token ID" is a unique ID assigned to a digital asset. In the case of Figure 29, it is the unique ID assigned to Chapter 6 of Book Data 2, which is the subject of the resale. "Royalty return to content rights holder" refers to the royalty rate of the compensation returned to the content rights holder (copyright holder, etc.) in connection with the resale. Compensation (resale price × royalty rate) is returned to the content rights holder each time it is resold.

[0195] Furthermore, the resale transaction 1 and usage rights certificate data (Verifiable Credential data 1) of this NFT-enabled content (Chapter 6 of Book Data 2) will also be recorded on the blockchain. The data recorded on the blockchain is encrypted using the purchaser's shared key K1.

[0196] Purchasers of the content (Chapter 6 of Book Data 2) can view the content (Chapter 6 of Book Data 2) by following the procedure below. 1. Access the NFT data on the blockchain and decrypt the NFT data using the buyer's own shared key K1. 2. Read the hash value "Qmuql2xv" from Chapter 6 of Book Data 2, search IPFS, and retrieve the corresponding URL "https: / / ghi". Based on the read URL "https: / / ghi", access the file server 106 and view the content data "Chapter 6 of Book Data 2" corresponding to the URL "https: / / ghi".

[0197] Based on FIG. 30, explain the proof data of the Verifiable Credential that proves the resale transaction recorded in the blockchain and the usage right of the person who became the purchaser. Referring to FIG. 30(A), an example of the information 108a recorded as the resale transaction 1 is shown below. · Transaction ID · Wallet addresses of seller A and purchaser B · Transaction date and time · Sold item: Chapter 6 of Book Data 2 · Selling price: 50 yen · Royalty rate to content rights holders · Electronic signature This "electronic signature" is an electronic signature performed using the private keys SK of both parties of the transaction (seller A and purchaser B).

[0198] An example of the information 108b recorded as the proof data 1 of the Verifiable Credential that proves the usage right of purchaser B who newly became the usage right holder of the content (Chapter 6 of Book Data 2) by this resale transaction 1 is shown below. · Certificate ID: 6&#!6!36&$ · Issuer information: Purchaser B · Purchaser information: Email address · Purchased item information: Chapter 6 of Book Data 2 · Purchase proof: Transaction ID, receipt number · Electronic signature and timestamp of the issuer of the Verifiable Credential Next, referring to FIG. 30(B), an example of the information 110a recorded as the resale transaction 2 is shown below. · Transaction ID · Wallet addresses of seller C and purchaser D · Transaction date and time • Items for sale: Chapter 6 of Book Data 2, Chapter 4 of Book Data 3, and Chapter 2 of Book Data 5 • Selling price: 190 yen • Royalty rates for content rights holders • Electronic signature This "digital signature" is an electronic signature made using the private keys SK of both transaction parties (seller C and buyer D). In this resale transaction 2, the items being resold are "Chapter 6 of Book Data 2, Chapter 4 of Book Data 3, and Chapter 2 of Book Data 5," which seller C has selected and arranged according to a certain theme from various book data owned by the seller. Note that it is also permissible to resell the data of an entire book, or to resell the data of multiple books together.

[0199] An example of information 110b to be recorded as Verifiable Credential 1, which proves the usage rights of purchaser D, who has newly become a user of the content (Chapter 6 of Book Data 2, Chapter 4 of Book Data 3, and Chapter 2 of Book Data 5) through this resale transaction 2, is shown below. Certificate ID: 2%43#%6&$5 • Issuer information: Buyer D • Buyer information: Email address • Purchased items information: Chapter 6 of Book Data 2, Chapter 4 of Book Data 3, and Chapter 2 of Book Data 5 • Proof of purchase: Transaction ID, receipt number • Electronic signature and timestamp of the issuer of the verifiable credential Based on Figure 31, the flowchart for the NFT creation process in the case of the resale transaction 1 described above will be explained. User terminal 54 determines whether or not to register book data via S200. If the user performs the book data registration operation, S200 determines it to be YES, and S201 performs the process of registering Chapter 6 of book data 2 to file server 106. Specifically, "book data 2EPUBCFI( / 6 / 4)" is registered to file server 106.

[0200] Next, user terminal 54 generates the hash value "Qmuql2xv" for chapter 6 of book data 2 via S202, and registers this hash value "Qmuql2xv" and the URL of the book data registered on the file server with IPFS4 (S203). Specifically, "Qmuql2xv https: / / ghi" is registered with IPFS4. Next, via S204, the user's wallet address and token ID are recorded on the blockchain and returned.

[0201] Based on Figure 32, the flowchart of the smart contract processing that automatically performs the above-described resale transaction 1 using a smart contract will be explained. User terminal A54 defines the resale price of Chapter 6 of Book Data 2 and the royalty rate to the content rights holders in S207. Similarly, user terminal B54 defines the resale price of Chapter 6 of Book Data 2 and the royalty rate to the content rights holders in S208. User terminal A54 confirms the other party's conditions in S209. Similarly, user terminal B54 confirms the other party's conditions in S210.

[0202] User terminal A54 determines whether the conditions are met by S211, and returns if they are not met. Similarly, user terminal B54 determines whether the conditions are met by S212, and returns if they are not met. If user terminal A54 determines that the conditions are met by S211, it sends a notification of agreement to the other party by S213 and then returns. User terminal B54, having received the notification of agreement by S214, executes the contract by S214 and records the contract (transaction 1) on the blockchain by S215. The recorded state is the record information 108a shown in Figure 30(A).

[0203] Next, user terminal B54 sends a resale notification to cloud server 51 via S216. Specifically, it is a notification that Chapter 6 of book data 2 has been resold from seller A to buyer B. Cloud server 51 receives this notification via S217 and issues a Verifiable Credential via S218, which it sends to user terminal B54. User terminal B54, upon receiving it, records the Verifiable Credential on the blockchain via S219. This recorded state is shown in the record information 108b in Figure 30(A). Note that the Verifiable Credential that seller A had obtained will no longer be usable. Specifically, the Proxy contract 120 shown in Figure 33(A) is used.

[0204] Next, we will explain the case where book data with annotations by celebrities, etc., is distributed N times (where N is an integer greater than or equal to 2), referring to Figure 33(B). Figure 33(B) is the same as the steps from S165 onwards in Figure 25 mentioned above. The difference is that in Figure 25, the publisher sends the book data to the user terminal 54 before publication, and the user makes annotations to that book data (S163, S164), whereas in Figure 33(B), the user makes annotations to the purchased book data while reading it (YES in S223) (S225~S232).

[0205] Next, the flowchart for the NFT creation process will be explained based on Figure 34. User terminal 54 determines whether or not to register the written book data via S237. If the user performs the operation to register the written book data, S237 determines it to be YES, and S238 registers the written book data to file server 106. Specifically, "Written book data EPUBCFI( / 6 / 4)" is registered.

[0206] Next, the user terminal 54 generates the hash value Qmuql3xy of the book data and the written data according to S239. The process of registering the hash value Qmuql3xy and the URL of the written book data registered in the file server in IPFS4 is performed according to S240. Specifically, "Qmuql3xy http: / / pqr" is registered in IPFS4.

[0207] Next, the user terminal 54 records and returns the user's wallet address, token ID, etc. in the blockchain according to S241. Specifically, the following information is recorded as NFT109. · Token ID · User's wallet address · Hyperlink of the written book data (hash value Qmuql3xy of the written book data and the written data) A flowchart for N - time circulation (N is an integer greater than or equal to 2) of the written book data NFT - ized in this way will be described based on FIG. 35. The user terminal A54 of user A defines the resale price of the written book data and the royalty rate to the content rights holder according to S245. Similarly, the user terminal B54 of user B also defines the resale price of the written book data and the royalty rate to the content rights holder according to S246.

[0208] The user terminal A54 checks the conditions of the other party according to S247. Similarly, the user terminal B54 also checks the conditions of the other party according to S248. The user terminal A54 determines whether the conditions are met according to S249, and returns if they do not match. Similarly, the user terminal B54 also determines whether the conditions are met according to S250, and returns if they do not match.

[0209] If the user terminal A54 determines that they match according to S249, it sends a notice of agreement to the user terminal B54 according to S251. The user terminal B54 that receives it at S251 executes the contract according to 252, and records the contract (transaction) in the blockchain according to S253.

[0210] User terminal B54 sends a notification to cloud server 51 via S254 that the written book data has been resold from user A to user B. Cloud server 51, having received this notification via S255, issues a Verifiable Credential via S256. This Verifiable Credential proves that user B has become the new rightful user of the written book data. This Verifiable Credential is then sent to user terminal B54 via S256. Upon receiving it, user terminal 54 records the Verifiable Credential on the blockchain via S257.

[0211] The flowchart for user B to view (read) the purchased written book data is explained based on Figure 36. User terminal B54 determines in S262 whether or not to view the purchased written book, and returns if not. If user B performs the viewing operation on user terminal B54, S262 determines it to be YES and proceeds to S263, where the written book data and the hash value Qmuql3xy of the written data are read from the blockchain (see NFT109 in Figure 34). Next, S264 reads the URL https: / / pqr corresponding to the hash value Qmuql3xy from IPFS4 (see IPFS4 in Figure 34). S265 accesses file server 106 and presents a Verifiable Credential to request the written book data stored corresponding to URL https: / / pqr (see file server 106 in Figure 34). Upon receiving the request, file server 106 verifies the received Verifiable Credential using S275. S276 determines whether the verification result is valid or not; if it is invalid, it returns a response, but if it is valid, it returns the requested written book data using S277.

[0212] User terminal B54 determines whether or not it has received the written book data via S274. If it has not received it, it returns a value. If it has received it, user B reads the received written book data via S266.

[0213] While User B is reading pre-written book data, S267 determines whether User B is reading a section identified by the CFI. If the determination is NO, control proceeds to S273. On the other hand, if User B's reading section is identified by the CFI, S267 determines it to be YES and proceeds to S268, where the CFI for the corresponding section is sent to the cloud server 51. The cloud server 51 receives this via S269 and, via S270, searches the CFIDB100 and reads the written content associated with the received CFI (see Figure 26(B)). Then, via S271, the read written content is sent back to the user terminal B54.

[0214] Upon receiving the message, user terminal B54 displays the returned content to user B via S272. Next, it proceeds to S273 to determine whether the reading is complete or not. If it is not complete, it returns to S266 and cycles through the loop S266→S267→S268→S272→S273→S266. When the reading is complete, S273 determines YES and returns.

[0215] This section explains, based on Figures 37(A) and 37(B), the case in which data from books with annotations by celebrities, etc., is converted into NFTs and resold to multiple parties. Referring to Figure 37(A), the usage rights (ownership rights) of the annotation data owned by user A are distributed and resold (transferred) to users B, C, ..., N. In Figure 35, the usage rights (ownership rights) of the annotation data owned by user A are resold (transferred) only to user B, and a Verifiable Credential is issued only to user B. However, in the case of distributed resale (distributed transfer) in Figure 37(A), usage rights certificates (Verifiable Credentials) are issued to all users B, C, ..., N, who have become new users (owners). Users B, C, ..., N who possess this Verifiable Credential can view the written book data (see Figure 36, S265, S275-S277, S274, S266).

[0216] The value of books containing annotations by celebrities fluctuates depending on the activities and fame of the celebrities who made the annotations. Items whose value fluctuates over time can be targets for investment and speculation. In Figure 37(A), user N resells their Verifiable Credential to user P, who then resells it to user Q. These resale prices fluctuate over time. This can be viewed as a type of investment market, with transactions occurring in a secondary market: user N → user P → user Q.

[0217] Referring to Figure 37(B), User N sold the right to use the written book data 102 to User P for 0.1 Ether (40,000 Yen), User P sold it to User Q for 0.12 Ether (48,000 Yen), User Q sold it to User R for 0.14 Ether (56,000 Yen), User R sold it to User S for 0.17 Ether (68,000 Yen), and User S sold it to User T for 0.2 Ether (80,000 Yen). As a result, the market value of the right to use the written book data 102, which was initially 40,000 Yen, surged to 80,000 Yen.

[0218] Figures 38(A) and 38(B) illustrate the case where written book data is securitized and traded in the secondary market. In this embodiment, security tokens (digital securities) are used as the method of securitization.

[0219] Referring to Figure 38(A), User A sells security tokens (labeled "ST" in Figure 38) related to written book data on the investment market for 10,000 yen per ST. User B purchases 3 STs, User C purchases 2 STs, and User D purchases 5 STs. User D then sells 1 ST to User E on the secondary market, and User E sells 1 ST to User F.

[0220] In Figure 37, Verifiable Credentials are issued to all users B, C, ..., N who have newly become users (owners). However, in the case of the investment market through the buying and selling of security tokens shown in Figure 38(A), Verifiable Credentials are issued to all purchasers of security tokens. Therefore, anyone who holds a security token will have a Verifiable Credential and will be able to view the written book data (see S265, S275-S277, S274, S266 in Figure 36).

[0221] Figure 38(B) shows the fluctuations in the sale price of security tokens. User D held 5ST (market capitalization of 50,000 yen) and sold 1ST112 to User E for 10,000 yen. User E sold that 1ST112 to User F for 15,000 yen. User F sold that 1ST112 to User G for 20,000 yen. User G sold that 1ST112 to User H for 25,000 yen. User H sold that 1ST112 to User I for 30,000 yen.

[0222] As a result, the market capitalization of 4ST held by User D surged from an initial 40,000 yen to 120,000 yen. [Fifth Embodiment] Next, a fifth embodiment will be described. This fifth embodiment is a technology independent of the first to fourth embodiments described above, but it may also be adopted after following the first to fourth embodiments. This fifth embodiment relates to a reading content utilization system that utilizes the reading content read by members (employees, etc.) belonging to organizations such as companies or members belonging to groups on social networking services (SNS).

[0223] Referring to Figure 39(A), the reading content utilization system 113 analyzes each member's reading content in the analysis phase 116 and utilizes the analysis results in the utilization phase 117. In the utilization phase 117, the system may utilize the reading content in a way that persuades other members by applying the psychology of persuasion, or in a way that selects people to carry out a certain project.

[0224] The reading content (114) is the source of each member's knowledge, and by analyzing it in the analysis phase (116), each member's intellectual factors can be extracted. These intellectual factors will then be utilized in various services in the application phase (117).

[0225] "The psychology of persuasion" is a set of communication techniques for persuading others, based on research and experimental findings from social psychology, behavioral economics, and neuroscience. It utilizes the unconscious psychology and behavior of humans to effectively persuade others. The psychology of persuasion includes the following methods: • Foot-in-the-door method (gradual request method) • Door-in-the-face technique (concessive request technique) • Low Ball Law (Law Requiring Prior Acceptance) Furthermore, the following are some points to consider in order to increase persuasiveness. Let's start with the conclusion. • Include things that the other person can relate to. • Show specific numbers and evidence. Explain in language the other person can understand. • To state definitively • Consider from multiple perspectives Referring to Figure 39(B), company 81 is equipped with terminal 54d, company server 79, and shared DB 80, which are connected by LAN. The shared DB 80 stores the reading history of each employee, corresponding to Taro, Jiro, ... Hanako.

[0226] Based on Figure 40, a flowchart of the processing operations of terminal 54d and the corporate server 79 will be explained. Terminal 54d determines in S277 whether or not to read the book purchased by the company, and if it is determined not to read it, control is transferred to S283. If the employee performs the operation to read the book on terminal 54d, S277 determines it as YES and proceeds to S278, where the employee ID is sent to the corporate server 79 to access the corporate library and search for the book. Upon receiving the book search request, the corporate server 79 verifies the employee ID in S279 and allows the search of the corporate library.

[0227] Next, the corporate server 79 uses S280 to determine whether or not there is a book that the user wishes to read. If there is no book that the user wishes to read, control is transferred to S284. If there is a book that the user wishes to read, S281 allows the user to read the book and registers the book as part of their reading history in the shared DB 80, linked to the employee's name.

[0228] Next, terminal 54d processes a request to create a persuasive presentation via S283, and then returns after processing a request to select personnel for the project, etc. via S285. On the enterprise server 79, processing to create a persuasive presentation via S284, and then returns after processing a selection of personnel for the project, etc. via S286.

[0229] The flowcharts of the subroutine programs for the S283 persuasion presentation creation request process and the S284 persuasion presentation creation process will be explained based on Figure 41. Terminal 54d is determined by S290 to decide whether or not to create a presentation to persuade a specific employee, and returns if it does not. If an employee performs the operation to create a presentation to persuade a specific employee on terminal 54d, S290 determines it to be YES and S291 sends a presentation creation request to the corporate server 79, specifying the employee to be persuaded. The corporate server 79, having received this request by S292, creates the persuasion presentation by S293 using the following process. • Retrieve the reading history of a specified employee from the shared DB80. • Analysis of the content of the books read by the employee (understanding of themes and messages, identification of points of empathy) • Application of persuasion theory (ethnographic approach, avoidance of ethnocentrism) • Application of psychological techniques (emphasizing common ground, utilizing cognitive dissonance) • Creating a specific presentation structure The enterprise server 79 returns the created presentation to the requester via S294. Terminal 54d retrieves the presentation via S295 and then returns.

[0230] This reading content utilization system allows you to create and present presentations that apply the "psychology of persuasion," which has the advantage of making it easier to get proposals and projects approved in internal meetings and other settings.

[0231] The flowcharts of the subroutine programs for the project selection request processing (S285) and the project selection processing (S286) are explained based on Figure 42. Terminal 54d determines in S300 whether or not to perform project selection, and returns if it determines not to. If an employee performs the operation to select personnel for a project on terminal 54d, S300 determines it as YES and proceeds to S301, where it notifies the project details and sends a request for personnel selection to the company server 79. The company server 79, having received this in S302, performs personnel selection in the following process in S303. • Understand the project details • Extract the knowledge and skills necessary to carry out the project. • Search the company library for books that contain that knowledge and those skills. • Search for employees who are reading the searched books in the shared DB80. Then, S304 sends the searched employee back to the requester.

[0232] Terminal 54d returns after obtaining the selection results via S305. [Sixth Embodiment] Next, the sixth embodiment will be described. This sixth embodiment is a technology independent of the first to fifth embodiments described above, but it may also be adopted after following the first to fifth embodiments. This sixth embodiment relates to the issuance and trading of security tokens (digital securities) by AI. In this sixth embodiment, security tokens may be referred to as ST.

[0233] Referring to Figure 43, smart contract processing takes place between the AI ​​and other AI groups (S415, S416). This processing involves AIs conducting transactions with each other and engaging in economic activities. For example, the exchange of compensation for sharing highly-rated answers between AIs, as explained in Figure 18 above, is one such example. Furthermore, a transaction in which an AI shares its own successful experiences (achievements) recorded in the knowledge-sharing DB987 with other AIs for a fee can also be considered an example of this smart contract processing. In other words, all transactions when AIs engage in economic activities can be considered examples of this smart contract processing.

[0234] Next, the AI ​​determines in S417 whether or not to issue security tokens. If it decides not to issue them, it proceeds to S419. If it decides to issue them, it issues security tokens (ST) in S418 and updates the total number of STs issued. Next, the AI ​​processes security token transactions with user terminals or other AI groups (S419, S420). This process is performed automatically by smart contracts. Next, the AI ​​processes dividends to distribute to security token holders (security token owners) in S421, and user terminals 54 or other AIs process dividend receipt in S422. In addition, user terminals 54 or other AIs process security token transactions in the secondary market with other user terminals 54 or other AIs (S423, S424).

[0235] The flowchart of the security token transaction processing subroutine program in S419 and S420 is explained based on Figure 44. This transaction is performed automatically by a smart contract. The AI ​​defines the selling price of the security token in S430. Similarly, the user terminal group 54 or other AI groups define the buying price of the security token in S431. These definitions are made taking into account the price fluctuations for each AI, which will be published in S472 later. The definitions in S455 and S456 are also updated according to the constantly changing price of the security token. This update is performed using the Proxy pattern described above (see Figure 33(A)).

[0236] The AI ​​verifies the other party's conditions via S432. Similarly, user terminal B54 or other AI verifies the other party's conditions via S433. The AI ​​determines whether the conditions match via S436, and returns if they do not match. Similarly, user terminal B54 or other AI determines whether the conditions match via S435, and returns if they do not match. If the AI ​​determines via S434 that the conditions match, it sends an agreement notification to the other party via S436. Upon receiving the agreement notification, user terminal B54 or other AI executes the contract via S437 and moves the consideration requested by the other AI from its wallet to the other AI's wallet via S439. Then, it records the contract details on the blockchain via S440 and sends the transaction details to the aggregation server via S441.

[0237] AI updates security token DB777 via S438. Specifically, it updates the number of security tokens held by the user ID (including AIID) of the counterparty in this security token transaction (in Figure 44, %2$s97#) by adding the number of security tokens sold.

[0238] The flowcharts of the subroutine programs for dividend processing in S421 and dividend receipt processing in S422 are explained based on Figure 45(A). AI determines in S445 whether or not to pay dividends, and returns if no dividends are to be paid. Dividends are paid, for example, once a year, and when the time for dividend payment arrives, S445 determines YES and proceeds to S446.

[0239] In S446, U(I) represents the user registered in security token DB777. I = 1, 2, ... N N is the total number of registered users. As mentioned earlier, this U(I) also includes AI.

[0240] In S447, the initial value "1" is set to I, and in S448, the number of security tokens held by security token DB777U(I) is read, and the dividend for that U(I) is calculated. Then, in S449, the process of sending that dividend to user U(I)'s user terminal 54 is performed. User U(I)'s user terminal 54 or AI receives the dividend in S450.

[0241] The AI ​​updates I by incrementing it by "1" in S451, and determines in S452 whether I has exceeded N (the total number of users registered in the security token DB777). If it has not exceeded N, control returns to S448, and the AI ​​loops through S448→S449→S451→S452→S448. It returns when S452 determines that it is YES.

[0242] The aggregation of security token price fluctuations by the aggregation server is explained based on Figure 45(B). The aggregation server determines whether or not it has received transaction details in S470. If it has not received them, it proceeds to S427. If transaction details are sent via S441 or S466 described later, S470 determines YES, and S471 calculates the 1ST transaction price for each security token issuing AI. S472 publishes the fluctuation status of that calculated price for each issuing AI. S473 determines whether or not to terminate the process; if not to terminate, it returns to S470, but if to terminate, control stops.

[0243] The flowcharts of the security token transaction processing subroutine programs in S423 and S424 are explained with reference to Figure 46. This transaction is performed automatically by a smart contract. User terminal a54 or AIa programmed defines the selling price of security tokens in the secondary market by S455. Similarly, user terminal b54 or AIb programmed defines the buying price of security tokens in the secondary market by S456.

[0244] User terminal a54 or AIa confirms the other party's conditions via S457. Similarly, user terminal b54 or AIb confirms the other party's conditions via S458. User terminal a54 or AIa determines whether the conditions match via S459, and returns if they do not match. Similarly, user terminal b54 or AIb determines whether the conditions match via S460, and returns if they do not match. If user terminal a54 or AIa determines via S459 that the conditions match, it sends a notification of agreement to the other party via S461. Upon receiving the notification of agreement, user terminal b54 or AIa executes the contract via S462 and moves the consideration requested by the other user from its wallet to the other user's wallet via S464. Then, it records the details of the contract on the blockchain via S465.

[0245] User terminal a54 or AIa updates the security token DB777 via S463. Specifically, it updates the number of security tokens held corresponding to the user ID of the counterparty user in this security token transaction (5pqw&7b in Figure 46) by adding the number of security tokens sold, and updates the number of security tokens held corresponding to its own user ID (sp&87dw in Figure 46) by subtracting the number of security tokens sold. [Other variations] Other variations of Embodiments 1 to 6 described above are listed below. (1) Ethereum's Ether was used as the cryptocurrency, but stablecoins may be used instead or in addition to it. (2) Token swaps between different types of tokens and exchanges or redemptions of different types of tokens for stablecoins may be made possible. (3) The token value display system may be configured to display to the user the fluctuation status of the token value (price per token) of the individual social tokens (personal tokens) or security tokens (digital securities) issued by each AI. This token value display system comprises: an acceptance means (acceptance unit) that accepts token transactions (sell orders and buy orders, etc.) in an investment market including a secondary market (e.g., Figure 38(A)(B), etc.); a transaction execution means (transaction execution unit) that matches the token transactions (sell orders and buy orders, etc.) accepted by the acceptance means (acceptance unit) and executes transactions that meet the conditions; an aggregation means (aggregation unit) that aggregates the price movements of the market over a certain period of time; and a presentation means (presentation unit) that presents the aggregation results by the aggregation means (aggregation unit). The aggregation means (aggregation unit) aggregates the opening price, high price, low price, and closing price of the market over a certain period of time.

[0246] The AI ​​platform system 82 (see Figures 4 and 22(A)) may be controlled to present users who access it with the aggregated results from the aggregation means (aggregation unit) for each AI. The system may also be controlled to allow users who have obtained these aggregated results to specify which AI they would like to receive an answer from. (4) A smart contract that automates transactions may be linked with AI so that the AI ​​automatically generates and updates the contract definition (program definition). In this case, the AI ​​updates the contract definition (program definition) to the optimal one based on information from a centralized oracle 2, etc. For this update, the system of the proxy contract 120 shown in Figure 33(A) is used. (5) A Verifiable Credential was used as the electronic certificate to sign that the user has the right to use the purchased content such as books, but the following means may be used instead or in addition to it. Attribute Certificate An attribute certificate is a certificate that proves a user's access rights and attribute information. It issues an x.509 certificate, separate from the public key certificate, that contains the attribute information for which the certificate is issued. • Proof using NFT (Non-Fungible Token) By utilizing blockchain technology, NFTs (Non-Functional Documents) related to digital items such as books are issued upon purchase. These NFTs are linked to the purchaser's wallet and function as proof of purchase. • Receipt with electronic signature When purchasing content, a receipt with an electronic signature is issued. This receipt contains purchase information and details of the content, and is digitally signed to prevent tampering by third parties. • Transaction proof recorded on the blockchain At the time of purchase, the purchase information is recorded as a transaction on the blockchain, and the hash value is provided to the buyer. This allows the buyer to prove the fact of purchase by referring to the transaction history on the blockchain. • Zero-knowledge proof This technology allows a buyer to prove that they have purchased content without disclosing specific purchase details or transaction information, simply by demonstrating that they "purchased" the content. • Digital watermark Embedding a unique digital watermark in electronic content (such as PDFs or EPUBs) makes it possible to identify the buyer, proving that the buyer owns the content. Decentralized Identity (DID) DID is used to prove on a decentralized network that a buyer has purchased content. Purchase information is recorded in a DID document, and verifiers can confirm the fact of the purchase. • Proof of purchase using smart contracts The purchase process is automated using smart contracts, and purchase information is recorded on the blockchain. Smart contracts ensure that the fact of purchase is automatically recorded in a verifiable format. • Utilizes Trusted Platform Module (TPM) and secure elements. If a device has a TPM or secure element, it can securely store content purchase information within the device and use it as proof of purchase. (6) EPUB CFI was used as a technology to precisely specify a particular location or text within an ebook, but the following technologies may be used instead or in addition to it. XPath Overview: XPath is a path language for precisely specifying elements and attributes within XML documents. Since EPUB files are composed of XHTML (XML based on HTML), it is possible to use XPath to specify specific elements within an EPUB file.

[0247] Usage example: / / div[@id='chapter1'] / p[3] specifies the third paragraph within a particular chapter.

[0248] XPath is used to locate specific elements, attributes, and text nodes within ebooks. CSS selectors Overview: CSS selectors allow you to specify certain elements within an HTML or XHTML document, much like a stylesheet. Since e-book readers support CSS, CSS selectors are used to access specific elements.

[0249] Example usage: `div#chapter1 > p:nth-of-type(3)` specifies the third paragraph of a particular chapter. Text Offset Overview: This is a method of specifying specific text within an entire document or a particular section using string offsets (start and end positions).

[0250] Usage example: For example, `offset: 123-130` indicates the range from the 123rd character to the 130th character. • Anchor (HTML Anchor) Overview: HTML This method involves using id or name attributes in tags to set an "anchor" at a specific location within a document, and then linking to that anchor.

[0251] Usage example: Set an anchor like this and refer to it in #section3. • Hash URL (Hash URL Fragment) Overview: This technique uses '#' at the end of a URL to jump to a specific section or element. It is based on the id attribute of elements within an HTML file.

[0252] Example usage: Jump to a specific section, like chapter01.xhtml#section2. (7) In the above-described embodiment, the AI ​​(agent) recorded responses to user prompts that received high ratings from the user as success experiences (including policy π) in the knowledge-sharing DB987. However, the sharing of success experiences (including policy π) among agents may be used in the pre-practice training stage (machine learning stage) of multiple agents. For example, a task AI equipped with a task generation engine generates a prompt (task) and gives it to a group of agents participating in the AI ​​training dojo platform. The group of agents processes the given task. An AI equipped with an evaluation engine evaluates the task processing results of each group of agents and gives each agent a reward according to the evaluation, thereby enabling each agent to perform reinforcement learning. The success experiences (including policy π) of the top agents with high total reward values ​​are recorded in the knowledge-sharing DB987. The group of agents is made able to access this knowledge-sharing DB987 and share success experiences (including policy π).

[0253] Even during this pre-implementation training phase (machine learning phase), content may be provided to the group of agents. However, this is contingent on a mutually agreed-upon contract being in place between the AI ​​operating company and the content AI utilization system operating company. Furthermore, even during this training phase (machine learning phase), an AI alliance (see S300-S309, S315-S319) may be formed. The group of agents that have undergone this prior training are then made to participate in the AI ​​platform system 82, etc., and the group of agents responds to user prompts. (8) In the AI ​​platform system of Figure 22(A), agents may form multi-agent groups (for example, S97, S299~S308, etc.), and task processing may be performed by the formed multi-agent groups (for example, S309, S315~S319, S52~S54, etc.). When forming such multi-agent groups, the system may be controlled to form multi-agent groups by searching for suitable agents by searching for performance data stored in the knowledge sharing DB987 (for example, the content of the task, the role assigned, the ID of the AI ​​(agent) registered in the federation, the strategies acquired through reinforcement learning through task processing, etc.). (9) Instead of storing performance data related to task processing (successful experiences including policy π in Figure 19) in the knowledge-sharing DB987, or in addition to that, each AI (agent) may store its own performance data (distributed storage). When forming a multi-agent team, each AI (agent) brings its own performance data together, holds a meeting, and selects an AI (agent) suitable for the task to be processed to form the multi-agent team. (10) S299 "understands the prompt and searches for an AI suitable for federation using data 1 and 2 in the knowledge sharing DB", but instead of this, or in addition to this, control may be implemented to select a federation partner by prioritizing other AIs (security token owners) that hold its own security token based on data stored in the security token DB 777. (11) A service in which an AI responds to a prompt entered by a user may be charged, and a fee may be collected from the person who entered the prompt. When collecting the fee, if the user is the holder of the AI's security token (security token owner) and receives a service from that AI, the fee may be reduced or made free. (12) In the above-described embodiment, the AI ​​displays a hyperlink 999 of the reference content it referenced when answering, regardless of whether the user has a verifiable credential (see Figure 13). Alternatively, the AI ​​may be controlled to display the hyperlink 999 of the reference content only if the user has a verifiable credential, while displaying the text of the reference content instead of the hyperlink 999 if the user does not have a verifiable credential. In other words, in the above-described embodiment, access to the specific information (hyperlink 999) that identifies the reference content is permitted regardless of whether a verifiable credential exists, and the control after the access operation differs depending on whether a verifiable credential exists. In this modified example, however, whether or not an access operation to the specific information (hyperlink 999) itself is permitted depends on whether or not a verifiable credential exists.

[0254] The specific control in that case is shown in Figure 47. This flowchart is a flowchart of a subroutine program that specifically shows the process of "displaying each of the returned answers" in S54 of Figure 10. Referring to Figure 47, the user terminal 54 performs a search process for proof of usage rights data for the content it wants to view in S63. The specific control for this is shown in Figure 12. As a result of the search process, S64 determines whether or not proof data was found. If it was found, S700 displays each of the returned answers and also displays the content on which the answers are based as hyperlinks. On the other hand, if S64 determines NO, S701 displays each of the returned answers and also displays the content on which the answers are based as text. As a result, even if the text displayed content is clicked, the reference content will not be displayed.

[0255] Furthermore, if control based on the presence or absence of content usage rights by the user terminal 54 as shown in Figure 47 (S64, S700, S701) is performed, then control based on the presence or absence of content usage rights on the server side (S67~S71, etc.) is not necessarily required. However, if control based on the presence or absence of content usage rights is left solely to the user terminal 54, there is a risk of unauthorized modification by malicious users, so it is desirable to perform redundant control on the server side (S67~S71, etc.). (13) In the embodiments described above, the AI ​​responds to user prompts regardless of whether the user has verifiable credentials or not. Alternatively, the AI ​​may be controlled to respond only if the user has verifiable credentials.

[0256] The specific control in that case is shown in Figure 48. This flowchart is a subroutine program flowchart that specifically shows the process of "sending a Prompt" in S50 and "sending the received Prompt to the registered AI" in S51 of Figure 10. In Figure 48, the same processing steps as in Figure 11 are numbered the same as in Figure 11. Here, the explanation of the same processing steps as in Figure 11 is omitted, and the new steps are mainly explained.

[0257] At S710, the user terminal 54 determines whether or not to specify content to be referenced. If the user specifies content to be referenced to the AI, S710 determines it to be YES and proceeds to S711, where the requested content and proof of usage rights data are sent to the cloud server 51, and the prompt entered by the user is also sent to the cloud server 51 by S712. On the other hand, if S710 determines it to be NO, control proceeds to S712, and the prompt entered by the user is sent to the cloud server 51 by S712.

[0258] The cloud server 51, having received these via S713, determines via S714 whether or not there is a specified request for reference content. If not, it sends a prompt from the user to the AI ​​via S715, and acknowledging the prompt via S717, which replies to the user terminal 54.

[0259] On the other hand, if a reference content is specified, S714 determines it to be YES, and S67 determines whether or not proof data exists. If not, the process proceeds to S71. If proof data exists, S67 determines it to be YES, and S68 verifies the proof data, and S69 determines whether or not it is valid. If it is valid, the control proceeds to S715, but if it is invalid, the process proceeds to S71, and a message indicating invalidity (NG) is sent to the user terminal 54.

[0260] The user terminal 54 receives a reply from S71 or S716 via S72, and determines whether the received content is NG or not via S73. If it is not NG, S740 displays that the prompt has been acknowledged. On the other hand, if it is NG, the control proceeds to S75, where it is determined whether or not to purchase the reference content specified by the user. If a purchase is made, purchase control is performed via S76-S80. When the user purchases the reference content, the product entered by the user is sent to the AI ​​via S717.

[0261] In this modified version, if the user does not possess the right to use the content (Verifiable Credential), the AI ​​will not respond to the user's prompt. Therefore, control steps (S67-S71, S64, S700, S701, etc.) to determine whether or not to allow the user to use (view) the reference content are not necessarily required. However, since there is still a risk of unauthorized modification by malicious users, it is desirable to perform the above control steps (S67-S71, S64, S700, S701, etc.) redundantly. (14) In the embodiments described above, a user acquires the right to use (Verifiable Credential) of specific content (e-books, etc.) by purchasing such content. In other words, a user owns the right to use the content. However, the embodiments are not limited to this, and the right to use (Verifiable Credential) of content (e-books, etc.) may change over time, with the right to use (Verifiable Credential) of specific content (e-books, etc.) being granted to the user for a predetermined period and then expiring, and the right to use (Verifiable Credential) of other content (e-books, etc.) being granted for a predetermined period and then expiring.

[0262] For example, in the case of an e-book subscription service, the content available for unlimited reading is constantly changing. In such cases where the content that can be read changes over time, the verifiable credential for accessing that content must also change accordingly. [Summary of Disclosed Information] The disclosed information is summarized and listed below. <First Disclosure> [Technical field] The first disclosure relates to a service delivery system in which artificial intelligence provides services to users, a service delivery method in which artificial intelligence provides services to users, a program executed on a computer used for providing services to users by artificial intelligence, and a server. [Background technology] Artificial intelligence, such as generative AI represented by the conventionally known Large Language Models (LLMs), generates and presents answers to prompts entered by the user. This generation function is made possible by pre-training the AI ​​using vast amounts of data from the internet, such as the Web, as machine learning data, and then fine-tuning it. In cases where the pre-trained data alone is insufficient, the AI ​​may use methods such as Retrieval-Augmented Generation (RAG) to search for external information from the Web and use the retrieved information to generate answers. There are also AI programs that include hyperlinks to the Web pages referenced in the answer generation, and when a user selects these hyperlinks, they are redirected to the linked web pages for viewing. For the user, this has the advantage of allowing them to verify the basis of the answer by viewing the referenced Web information.

[0263] However, some information on the web is inaccurate, and the use of such information by artificial intelligence can lead to hallucination. Hallucination is a phenomenon in which artificial intelligence generates information that is inaccurate or nonexistent based on the data it has learned.

[0264] To address this hallucination problem, there was a technology (for example, Patent Document 1) in which the user inputs prompts to a generating AI, the user evaluates the content output by the generating AI, the input prompts, the output content, and the user's evaluation are stored as historical information, and a reward model is trained based on this historical information. [Prior art document] [Patent] [Patent Document 1] Patent No. 7530134 [Overview of the prefecture] [Problems the invention aims to solve] In this type of background technology, hallucination is prevented based on user evaluation, which requires users to have the ability to discern whether the content generated by the AI ​​is factually incorrect. However, few ordinary users possess this ability, which is a drawback as it limits the effectiveness of hallucination prevention.

[0265] The root cause of hallucination lies in the low accuracy (reliability) of the data used by artificial intelligence. An effective way to resolve this root cause is to use data with high accuracy (reliability), such as book data or paid content, for artificial intelligence. However, if artificial intelligence were to use such high-quality content data without permission, there is a risk of problems such as copyright infringement and unauthorized use of paid content.

[0266] In other words, when artificial intelligence uses high-quality content data with a high degree of truthfulness (accuracy), it is necessary to clear up issues related to the content, such as copyright and the right to collect royalties for the use of paid content (hereinafter referred to as "content rights").

[0267] This invention was conceived in view of the above circumstances, and its purpose is to resolve the content rights issues that arise when artificial intelligence uses high-quality content data with a high degree of truthfulness (accuracy). [Means for solving the problem] The following lists the solutions described in the first disclosure, with corresponding portions of the embodiments and drawings inserted in parentheses. Note that "person" in the first disclosure is a broad concept including natural persons, legal persons, and artificial intelligence. Furthermore, artificial intelligence is a broad concept including agents, multi-agents, and mobile agents.

[0268] (1) A service provision system in which artificial intelligence provides services to a user (for example, responding to user prompts, processing tasks in response to user requests such as task requests, and obtaining responses from the user by specifying reference content to the AI ​​and inputting prompts, such as S192), The system includes reference content usage control means (for example, S67-S70, S72, S74, S84-S92, S100, S102, S103, S52-S54, S318, S322, S319, S155, S192, S193, S63, S64, S700, S701, etc.) that control how to make reference content (for example, books, images, videos, CGM, music, etc., content DB73 in Figure 2, etc.) that the artificial intelligence used as a reference for providing the aforementioned service available to the user. The aforementioned reference content usage control means is: A means for determining whether a user has acquired a paid right to use the aforementioned reference content (for example, a Verifiable Credential, which may be controlled to change in accordance with changes in the content covered by the unlimited reading service), (for example, S67-S69, S84-S92, S64, S700, S701, etc.), The system includes a permission control means (for example, S70, 72, S74, S64, S700, S701, etc.) that controls whether a user is permitted to use the reference content, provided that the user is determined to have acquired the usage rights by the usage rights determination means. The consideration paid by the user who acquired the aforementioned usage rights is given to the person who holds the rights associated with the aforementioned reference content (for example, the benefits to the content rights holders in Figure 1).

[0269] With this configuration, the fees paid by users for acquiring the right to use the content referenced by the artificial intelligence are given to the content rights holders who hold the rights associated with the content. This provides content rights holders with an incentive to provide content to the artificial intelligence, and minimizes the content rights issues that arise when the artificial intelligence uses high-quality content data with a high degree of accuracy.

[0270] (2) In (1) above, the reference content usage control means includes presentation control means (for example, S100, S102, S103, S52~S54, S318, S322, S319, S155, S192, S193, etc.) that control the presentation of specific information that identifies the reference content (for example, hyperlink 999 of the content shown in the answer in Figure 13) to the user, The permission control means controls the use of the reference content identified by the specific information (for example, S70, 72, S74, etc.) on the condition that the user who performed the access operation to the specific information is a user who has been determined by the usage right determination means to have acquired the usage right (for example, determined to be YES in S69, etc.).

[0271] With this configuration, specific information identifying the reference content that the AI ​​used is presented to the user, thus promoting the content. The consideration paid by the user for acquiring the right to use that reference content is given to the content rights holder who holds the rights associated with the reference content. This provides content rights holders with an incentive to provide content to the AI, and minimizes the content rights issues that arise when the AI ​​uses high-quality content data with high accuracy.

[0272] (3) In (1) above, the reference content usage control means includes presentation control means (for example, S700, S701, S100, S102, S103, S52~S54, S318, S322, S319, S155, S192, S193, etc.) that control the presentation of specific information that identifies the reference content to the user (for example, a hyperlink 999 of the content shown in the answer in Figure 13 controlled by S700, or the text display of the content controlled by S701, etc.), The aforementioned allowable control means is Access determination means (for example, S55, S63-S69, etc.) that determines whether an operation to access the presented specific information has been performed by a user who has been determined to have acquired the right of use by the aforementioned right of use determination means, Based on the condition that the access determination means determines that an access operation has been performed (for example, S700 is performed and S69 is YES), control is performed to allow the user to use the reference content identified by the specific information (for example, S70, S72-S74, etc.).

[0273] With this configuration, specific information identifying the reference content that the AI ​​used is presented to the user, thus promoting the content. The consideration paid by the user for acquiring the right to use that reference content is given to the content rights holder who holds the rights associated with the reference content. This provides content rights holders with an incentive to provide content to the AI, and minimizes the content rights issues that arise when the AI ​​uses high-quality content data with high accuracy.

[0274] (4) A service provision system (for example, Figures 4(A), 13, 22(A), 27, etc.) in which artificial intelligence provides services to a user (for example, responding to user prompts, processing tasks in response to user requests such as task requests, S192, etc., where the user specifies reference content to the AI, inputs a prompt, and obtains a response), Presentation control means (e.g., S100, S102, S103, S52~S54, S318, S322, S319, S155, S192, S193, etc.) that control the presentation of specific information (e.g., hyperlink 999 of the content shown in the answer in Figure 13) that identifies reference content (e.g., books, images, videos, CGM, music, etc., content DB73 in Figure 2, etc.) that the artificial intelligence referenced in order to provide the aforementioned service, to the user, The system includes a purchase page display control means (e.g., S76) that controls the display of the purchase page for the reference content identified by the aforementioned specific information, The consideration paid by a user who purchases the reference content identified by the specified information from the displayed purchase page is given to the person who holds the rights associated with the content (for example, the benefits to the content rights holders shown in Figure 1).

[0275] With this configuration, the presentation of specific information to users that identifies the reference content used by the artificial intelligence serves as promotion for that content. The display of purchase pages for the reference content identified by the specific information promotes the sale of the reference content, making it easier to incentivize content rights holders to provide their content to the artificial intelligence. Furthermore, since the payment made by users who purchase the reference content identified by the specific information from the purchase page is given to those who hold the rights associated with the reference content, it further incentivizes content rights holders to provide their content to the artificial intelligence.

[0276] (5) A service provision system in which artificial intelligence provides services to a user (for example, responding to user prompts, processing tasks in response to user requests such as task requests, and obtaining responses from the user by inputting prompts with reference content specified by the user, as in S192), The aforementioned artificial intelligence provides a service to the user based on the content (for example, the process of sending a Prompt in Figure 48 and the process of sending the received Prompt to the AI, S52-S55, S96-S103, S299-S309, S315-S319, S148-S151, S154, S155, S185-S197, etc.), The service provision means includes a reference content usage control means (for example, S67-S70, S72, S74, S84-S92, S100, S102, S103, S52-S54, S318, S322, S319, S155, S192, S193, S63, S64, S700, S701, etc.) that controls how reference content, which is content that the artificial intelligence referenced when providing services, can be used by the user. The aforementioned service provision means is A means for determining whether a user has acquired a paid right to use the aforementioned reference content (for example, a Verifiable Credential, which may be controlled to change in accordance with changes in the content covered by the unlimited reading service), (for example, S67-S69, S84-S92, S64, S700, S701, etc.), The system includes a service permission means (for example, S715, etc.) that allows the provision of the service to a user on the condition that the user has been determined to have acquired the usage rights by the usage rights determination means (YES in S69 of Figure 48), The consideration paid by the user who acquired the aforementioned usage rights is given to the person who holds the rights associated with the aforementioned reference content (for example, the benefits to the content rights holders in Figure 1).

[0277] (6) The above (5) further includes a reference content designation means (for example, S710, S711, S185, etc.) that allows a user who is to receive a service from the service provision means to designate content that they would like the user to refer to, The service provision means provided above provides the user with a service that takes into account the content specified by the user (for example, S98~S103, S317~S319, S149, S150, S154, S155, S135, S136, S185~S196, etc.), The service-permitting means permits the user to provide the service on the condition that the user has acquired a paid usage right that permits the use of the content specified by the user (for example, YES in S69 and S198 in Figure 48, etc.).

[0278] (7) In the above (1) to (6), the content includes books, The rights associated with the aforementioned content include the copyright of the aforementioned book.

[0279] (8) In (2) to (7) above, the permitting means includes viewing permitting means (for example, S70, 72, S74, etc.) that permits a user to view the reference content identified by the specific information, on the condition that the user who performed the selection operation of the specific information (for example, S65, etc.) is determined by the usage right determination means to have acquired the usage right.

[0280] (9) In any of (2) to (8) above, the system further includes purchase page display control means (e.g., S75, S76, etc.) that controls the display of the purchase page of the reference content identified by the specific information (e.g., the image below Figure 13) when a user who has not acquired the usage rights selects the specific information (e.g., clicks hyperlink 999 in Figure 13), If a user purchases the aforementioned reference content (for example, YES in S77 and S78), they can acquire the right to use that content (for example, S79 and S80).

[0281] With this configuration, the purchase page for reference content is displayed to the user, promoting the sale of the reference content and thus creating a greater incentive for content rights holders to provide their content to artificial intelligence.

[0282] (10) In any of (1) to (9) above, if a user who has acquired the usage rights transfers the content that is permitted to be used by said usage rights (for example, S207 to S214, etc.), the system further provides a means for granting usage rights that permit the use of said content to the transferee of the content (for example, S217 to S219, etc.).

[0283] With this configuration, even when content is transferred from one user to another, the new transferee is granted the right to use it, thus preventing the inconvenience of the transferee being unable to use the content.

[0284] (11) In any of the above (1) to (10), the usage rights include two or more of the following: full user usage rights that allow the user to use an entire piece of content; partial usage rights that allow the user to use only a specific part of a piece of content when the artificial intelligence references that specific part; and full reference part usage rights that allow the user to use any part of a piece of content that the artificial intelligence references.

[0285] This configuration increases the options available to users when acquiring usage rights.

[0286] (12) In any of (1) to (11) above, an evaluation means for the user to evaluate the service provided (for example, S56 to S58, the Like icon 998 in Figure 13, etc.), A ranking means (for example, S119-S120, S124, etc.) that aggregates the reference content of services that have been highly rated by the evaluation means and ranks them by high rating, with respect to the reference content which is the content that the artificial intelligence has referenced in order to provide the aforementioned service, The system further includes a means for publishing the ranking results of referenced content by the aforementioned high-ranking means (e.g., S123, S124, etc.).

[0287] This configuration allows users to access high-ranking reference content, and promotes the sale of that content.

[0288] (13) In (12) above, the evaluation means includes a user-specific high-rating ranking means (for example, S400 to S403, etc.) that classifies and ranks the high ratings for each user, The aforementioned publication means includes a user-specific publication means (for example, S404 to S409, etc.) that publishes the ranking results classified for each user by the user-specific high-rating ranking means for each corresponding user.

[0289] This configuration allows users to see the ranking of their own highly-rated reference content, thereby promoting the sale of high-ranking content.

[0290] (14) In any of the above (1) to (13), the system further includes a means for the user to request the artificial intelligence to provide a service (e.g., S50, S186, etc.), The request means includes a content specification request means (for example, S50, S186, etc.) for specifying the content to be referenced and requesting it from the artificial intelligence. When the artificial intelligence is requested to provide the specified content (for example, YES in S98, YES in S149 → S150 → S136 → S154), it will use the specified content to provide the service to the requester (for example, YES in S99 → S100, NO in S99 → S101 → S102, S155, S192, etc.).

[0291] With this configuration, users can specify the content that the artificial intelligence should refer to and receive services that utilize that content.

[0292] (15) The system further includes content provision means that allows the artificial intelligence to provide content, provided that the artificial intelligence is an artificial intelligence that presents specific information identifying the reference content to the user by a presentation control means that controls the presentation of specific information identifying the reference content to the user in any of the above (1) to (14). (For example, if YES in S36, then YES in S38, then S39; if NO in S36, then S41; 100 to S102; S318; S321; S322, etc.)

[0293] With this configuration, the artificial intelligence, which is provided with content as pre-training data, presents users with specific information about the content it has pre-trained on. This serves as promotion for the content, benefits the content rights holder, and prevents hallucination.

[0294] (16) In (15) above, the content providing means includes pre-training providing means that allows the artificial intelligence to be provided with content as data for pre-training (for example, YES in S36 → YES in S38 → S39, etc.).

[0295] (17) In (15) or (16) above, the content providing means includes service-time providing means that allow the artificial intelligence to provide the content when providing the service (for example, NO in S36 → S41, 100 to S102, S318, S321, S322, etc.).

[0296] (18) A method of providing services in which artificial intelligence provides services to a user (for example, responding to user prompts, processing tasks in response to user requests such as task requests, S192, where the user specifies reference content to the AI, inputs a prompt, and obtains a response), (for example, Figures 4(A), 13, 22(A), 27, etc.) The process includes usage control steps (for example, S67-S70, S72, S74, S84-S92, S100, S102, S103, S52-S54, S318, S322, S319, S155, S192, S193, etc.) that control how to make the reference content (for example, books, images, videos, CGM, music, etc., content DB73 in Figure 2) that the artificial intelligence referenced in order to provide the aforementioned service available to the user. The aforementioned usage control step is: A usage rights determination step (for example, S67-S69, S84-S92, S64, S700, S701, etc.) determines whether the user has acquired a paid usage right that permits the use of the aforementioned reference content (for example, a content usage right (Verifiable Credential), which may be controlled to change in accordance with changes in the content covered by the unlimited reading service), The system includes an allowance control step (for example, S70, 72, S74, S64, S700, S701, etc.) which controls whether a user is allowed to use the reference content, provided that the user is determined to have acquired the right of use in the aforementioned right of use determination step. The consideration paid by the user who acquired the aforementioned usage rights is given to the person who holds the rights associated with the aforementioned reference content (for example, the benefits to the content rights holders in Figure 1).

[0297] With this configuration, the fees paid by users for acquiring the right to use the reference content that the artificial intelligence uses are given to the content rights holders who hold the rights associated with the reference content. This provides content rights holders with an incentive to provide content to the artificial intelligence, and minimizes the content rights issues that arise when the artificial intelligence uses high-quality content data with a high degree of accuracy.

[0298] (19) A program that is used to provide services to a user by artificial intelligence (for example, responding to user prompts, processing tasks in response to user requests such as task requests, S192, where the user specifies reference content to the AI, inputs a prompt, and obtains a response), and is executed on a computer (for example, a user terminal 54), To the aforementioned computer, In order to provide the aforementioned service, the AI ​​uses reference content (e.g., books, images, videos, CGM, music, etc., as shown in the content DB73 in Figure 2) as reference content, and the AI ​​executes usage control steps (e.g., S63-S65, S72-S74, S84, S85, S92, S93, S64, S700, S701, etc.) to control the user's ability to use this reference content. The aforementioned usage control step is: The process includes the step of providing a service to a user (for example, S72-S74, etc.) that allows the user to use the aforementioned reference content, provided that the user has acquired a paid usage right that permits the use of the aforementioned reference content (for example, a content usage right (Verifiable Credential), which may be controlled to change in accordance with changes in the content covered by the unlimited reading service) (for example, YES in S69, YES in S64 in Figure 47), Furthermore, the computer is instructed to perform a step (for example, S77, etc.) in which it accepts an operation for a user to pay a fee in order to acquire the usage rights. The consideration paid by the user who acquired the aforementioned usage rights is given to the person who holds the rights associated with the aforementioned reference content (for example, the benefits to the content rights holders in Figure 1).

[0299] With this configuration, the fees paid by users for acquiring the right to use the reference content that the artificial intelligence uses are given to the content rights holders who hold the rights associated with the reference content. This provides content rights holders with an incentive to provide content to the artificial intelligence, and minimizes the content rights issues that arise when the artificial intelligence uses high-quality content data with a high degree of accuracy.

[0300] (20) A server (e.g., cloud server 51, enterprise server 79, etc.) used in a service provision system in which artificial intelligence provides services to a user (e.g., responding to user prompts, processing tasks in response to user requests such as task requests, S192, etc., where the user specifies reference content to the AI, inputs a prompt, and obtains a response), Equipped with a processor (e.g., GPU 10g, etc.) and memory (e.g., ROM 11, RAM 9, SSD 12, etc.), The aforementioned processor, In order to provide the aforementioned service, the AI ​​executes reference content usage control processing (for example, S67-S70, S100, S102, S103, S52, S53, etc.) to control the user's ability to use the reference content, which is the content that the AI ​​referenced. The aforementioned reference content usage control process is: A usage rights determination process (e.g., S67-S69) determines whether the user has acquired a paid usage right that permits the use of the aforementioned reference content (for example, a content usage right (Verifiable Credential), which may be controlled to change in accordance with changes in the content covered by the unlimited reading service), The process includes an allowance control process (e.g., S70) that controls the use of the reference content to a user, provided that the user is determined to have acquired the right of use by the right of use determination means, The consideration paid by the user who acquired the aforementioned usage rights is given to the person who holds the rights associated with the aforementioned reference content (for example, the benefits to the content rights holders in Figure 1).

[0301] With this configuration, the fees paid by users for acquiring the right to use the reference content that the artificial intelligence uses are given to the content rights holders who hold the rights associated with the reference content. This provides content rights holders with an incentive to provide content to the artificial intelligence, and minimizes the content rights issues that arise when the artificial intelligence uses high-quality content data with a high degree of accuracy. <Second Disclosure> [Technical field] The second disclosure relates to a task processing system using a collection of multiple agents. [Background technology] Conventionally, there have been methods that use reinforcement learning to distill the strategy of the best-performing agent among multiple agents being trained on a task into other agents (Patent Document 1). [Patent] [Patent Document 1] Special Publication No. 2024-529460 [Overview of the prefecture] [Problems the invention aims to solve] In this underlying technology, only the policy of the best-performing agent among multiple agents undergoing reinforcement learning is provided to the other agents. However, for multiple agents to form a multi-agent system and collaborate to perform tasks, the optimal policy differs for each role assigned to each agent in the multi-agent system. This underlying technology cannot handle situations where an optimal policy exists for each role. As a result, it has the drawback of not improving performance when multiple agents form a system and collaborate to perform tasks.

[0302] The second disclosure was conceived in light of these circumstances, and its purpose is to improve performance when multiple agents form a multi-agent system and collaborate to perform tasks. [Means for solving the problem] The following lists the solutions described in the second disclosure, with corresponding portions of the embodiments and drawings inserted in parentheses.

[0303] (1) A task processing system consisting of a collection of multiple agents (for example, the AI ​​platform system in Figure 4(A) and Figure 22(A)), A rewarding means (e.g., S56-S58, etc.) that gives each agent a reward based on an evaluation of the task processing results of actions taken by the agent in order to process the task (e.g., high ratings (likes) from users, evaluation of task processing results by an AI equipped with an evaluation engine, etc.), A means for storing performance data related to the task processing performed by each agent (for example, a knowledge-sharing DB987, distributed storage for each agent, etc.), A means for forming a multi-agent network with multiple agents to process a task (for example, S97, S299~S308, etc.), The system comprises task processing means (for example, S309, S315-S319, S52-S54, etc.) that performs task processing using the multi-agent formed by the multi-agent formation means, The multi-agent formation means includes a search means (e.g., S299, S135, etc.) that searches for an agent suitable for processing a task based on the data stored in the performance data storage means, and, on the condition that a suitable agent has been found by the search means, a multi-agent is formed with the found agent (e.g., S300~S308, etc.).

[0304] With this configuration, based on performance data related to task processing, it is possible to search for agents suitable for task processing and form a multi-agent system. This allows for the formation of a multi-agent system suitable for each task being processed, thereby improving the performance when performing tasks using a multi-agent system.

[0305] (2) In (1) above, the performance data storage means stores the contents of the task (for example, Figure 19, etc.).

[0306] With this configuration, by referring to the content of tasks processed by agents, it is possible to select an agent with a proven track record that is suitable for the task being processed in the current case.

[0307] (3) In (1) or (2) above, the performance data storage means stores the role that was assigned (for example, Figure 19, etc.).

[0308] With this configuration, by referring to the roles that agents have played in the tasks they have processed, it is possible to appropriately assign roles when processing the task currently being processed.

[0309] (4) In any of (1) to (3) above, the performance data storage means stores information that identifies the formed multi-agent (for example, the ID of the AI ​​(agent) registered in the alliance) (for example, Figure 19).

[0310] (5) In any of (1) to (4) above, the performance data storage means stores the strategies acquired through reinforcement learning by task processing (for example, Figure 19).

[0311] With this configuration, the task can be processed by borrowing strategies learned by agents that have performed tasks similar to the one being processed, thereby improving performance when performing tasks with multiple agents.

[0312] (6) In any of (1) to (4) above, the system further includes a sharing means for providing among agents the task processing results for which a higher reward has been obtained through the reward-granting means, The agent performs machine learning using the task processing results shared by the sharing means (e.g., "supervised learning" in S158).

[0313] With this configuration, the agent's performance can be improved through machine learning using the results of tasks that performed well. <Third Disclosure> [Technical field] The third disclosure relates to a service delivery system in which artificial intelligence provides services to users. [Background technology] Traditionally, during the "training" process for customizing ChatGPT, it was possible to train the system with the content of publications (books), allowing the purchaser to not only read the purchased publications (books) but also converse with the trained ChatGPT (GPTs) (for example, Non-Patent Document 1). [Prior art document] [Non-Patent Literature 1] "A first in the publishing industry! From 'books to read only' to 'books to engage in dialogue': Launching the 'GPTs x Publishing' production service. ~As the first release, a book by the representative of Clover Publishing Co., Ltd. has been released as a GPT." [online], [Published January 19, 2024], Internet<https: / / www.dreamnews.jp / press / 0000293079 / > [Overview of the prefecture] [Problems the invention aims to solve] However, this underlying technology has the drawback that, because there is a one-to-one correspondence between purchased books and trained GPTs, even if a user purchases multiple books, they can only interact with the GPTs for one of those purchased books, and cannot have a comprehensive conversation (question and answer) with the GPTs for multiple purchased books. In other words, because there is a one-to-one correspondence between purchased content and trained artificial intelligence, even if a user purchases multiple pieces of content, they can only receive services such as question and answer by inputting prompts into the AI ​​that has learned from one of those purchased pieces of content, and cannot enjoy a comprehensive service for multiple pieces of purchased content. [Means for solving the problem] The following lists the solutions described in the third disclosure, with corresponding portions of the embodiments and drawings inserted in parentheses.

[0314] (1) A service provision system in which artificial intelligence provides services to users (for example, Figure 27, etc.), A first trained artificial intelligence (e.g., LLM model A103a) that has learned from the first content purchased by the user, A second trained artificial intelligence (e.g., LLM model B103b) that has learned from a second piece of content purchased by the user, A prompt input means (e.g., S186) for the user to input prompts to the first trained artificial intelligence and the second trained artificial intelligence, The system includes an integration means (e.g., S195, S196, etc.) that provides the user with a result obtained by integrating the response result of the first trained artificial intelligence to the input prompt (e.g., S192, etc.) and the response result of the second trained artificial intelligence to the input prompt (e.g., S192, etc.).

[0315] With this configuration, users can enjoy a comprehensive service encompassing multiple purchased content items.

[0316] (2) In (1) above, the first trained artificial intelligence and the second trained artificial intelligence perform response processing to prompts on the condition that the task requester has acquired the right to use the learned content (for example, S67-S69, S84-S92, etc.).

[0317] With this configuration, if a user separates the first or second trained artificial intelligence from the purchased content and transfers it, the transferee can enjoy services provided by the first or second trained artificial intelligence even though they do not have the right to use the first or second content, thus preventing such inconvenience. <Fourth Disclosure> [Technical field] The fourth disclosure concerns the distribution system for electronic content. [Background technology] Conventionally, there have been electronic content distribution systems that sell (primary distribution) or resell (Nth-th-thread distribution (where N is an integer greater than or equal to 2)) content such as ebooks (for example, Patent Document 1). [Prior art document] [Patent Document 1] Patent No. 6087469 [Overview of the prefecture] [Problems the invention aims to solve] However, this underlying technology simply distributes electronic content created and completed by authors such as book authors, and has the drawback of not incorporating any added value. [Means for solving the problem] The following lists the solutions described in the fourth disclosure, with corresponding portions of the embodiments and drawings inserted in parentheses.

[0318] (1) A distribution system for electronic content, A means of requesting that a seller of electronic content (e.g., ebooks, etc.) (e.g., publisher 3, etc.) write on electronic content before it is sold (e.g., S163, etc.), The system includes a writing means (for example, S166 to S169, etc.) used by the person who receives the request via the aforementioned request means to write to the electronic content, The aforementioned seller sells electronic books (for example, S178, etc.) that have been written to using the aforementioned writing means.

[0319] With this configuration, for example, electronic content with comments made by celebrities can be given added value compared to regular content.

[0320] (2) In the above (1), the electronic content is an ebook, The writing means identifies the writing location in the e-book using the EPUB CFI (Canonical Fragment Identifier) ​​and writes the content, thereby linking the CFI that identifies the writing location with the written content and storing it in a storage means (e.g., CFIDB100, etc.) (e.g., S167~S169, etc.).

[0321] (3) In (1) or (2) above, a means for determining whether or not a user has a right to use the electronic content (e.g., viewing the electronic content) (e.g., a Verifiable Credential), The system further includes an allowance control means (for example, S277, S274, S257~S273, etc.) that controls the use of the electronic content to the user on the condition that the user is determined to be a person who has the right to use it by the aforementioned right to use determination means, The permission control means, when the usage rights are distributed and transferred to multiple parties, controls each of the transferred transferees to allow the use of the electronic content on the condition that it is determined that each of the transferees possesses the usage rights (for example, Figure 37(A), etc.).

[0322] (4) In (1) or (2) above, a means for determining whether or not a user has a right to use the electronic content (e.g., viewing the electronic content) (e.g., a Verifiable Credential), The system further includes an allowance control means (for example, S277, S274, S257~S273, etc.) that controls the use of the electronic content to the user on the condition that the user is determined to be a person who has the right to use it by the aforementioned right to use determination means, The aforementioned usage rights are issued to all acquirers who acquire electronic securities issued for the aforementioned electronic content. The permission control means controls each acquirer who has acquired an electronic security issued for the electronic content to allow the acquirer to use the electronic content, on the condition that it is determined that the acquirer has the right to use the electronic content (for example, Figure 38(A), etc.). <5th Disclosure> [Technical field] The fifth disclosure concerns an agent system in which agents autonomously conduct economic activities. [Background technology] Conventionally, there have been systems in which AI agents autonomously process tasks (for example, Patent Document 1). [Prior art document] [Patent Document 1] Japanese Unexamined Patent Publication No. 2033-126606 [Overview of the prefecture] [Problems the invention aims to solve] One approach is to assign tasks involving economic activity to such agents, thereby enabling them to generate revenue.

[0323] However, simply having agents handle tasks involving economic activity had the drawback that it was difficult to conduct sufficient economic activity compared to a corporation in the physical world.

[0324] The fifth disclosure was conceived in light of these circumstances, and its purpose is to enable agents to conduct sufficient economic activities when they autonomously engage in economic activities. [Means for solving the problem] The following lists the solutions described in the fifth disclosure, with corresponding portions of the embodiments and drawings inserted in parentheses.

[0325] (1) An agent system in which agents autonomously carry out economic activities (e.g., S415, S416, etc.), A digital securities issuance means (e.g., S418, etc.) that issues the agent's own digital securities (e.g., security tokens, etc.), A primary trading instrument (e.g., S430-S441, etc.) that trades digital securities issued by the aforementioned digital securities issuance instrument in the primary market, The system also includes a secondary trading means (for example, S455 to S466, etc.) for trading the digital securities acquired by the primary trading means in the secondary market.

[0326] With this configuration, just like a corporation in the physical world, the agent's own digital securities are issued and traded in the primary and secondary markets, making it easier for the agent to conduct sufficient economic activities when operating autonomously.

[0327] (2) The above (1) further includes smart contract means (for example, S430-S441, S455-S466, etc.) that automatically perform transactions by one or both of the primary transaction means and the secondary transaction means using a smart contract.

[0328] (3) In (1) or (2) above, an agent is included as a trading entity in either or both of the primary trading means and the secondary trading means (for example, Figures 44, 46, etc.).

[0329] (4) The above (1) to (3) further includes dividend distribution means (for example, S445 to S452, etc.) for distributing dividends to holders of the digital securities.

[0330] (5) The above (1) to (4) further includes a means for publishing (for example, S470 to S473, etc.) for calculating and publishing the trading price of one unit of digital security for each AI issued by a digital security.

[0331] (6) The above (5) further comprises a transaction condition determination means (for example, S430, S431, S455, S456, etc.) that determines the transaction conditions of one or both of the primary transaction means and the secondary transaction means based on the content published by the public disclosure means. <Disclosure No. 6> [Technical field] The sixth disclosure concerns a service delivery system that utilizes users' reading content. [Background technology] Previously, some systems estimated the reader's level of understanding and difficulty based on the operation history of the device displaying the ebook, and recommended books that matched the reader's level of understanding. (For example, Patent Document 1). [Prior art document] [Patent Document 1] Japanese Unexamined Patent Publication No. 2011-215679 [Overview of the prefecture] [Problems the invention aims to solve] However, this underlying technology merely analyzes the reader's situation while reading and recommends books that are suitable for them. Reading is a source of knowledge for readers, and there has been no service that utilizes this source of knowledge until now.

[0332] The sixth disclosure was conceived in light of these circumstances, and its purpose is to provide a service that utilizes the reading content itself, which is the source of users' knowledge. [Means for solving the problem] The following lists the solutions described in the sixth disclosure, with corresponding portions of the embodiments and drawings inserted in parentheses.

[0333] (1) A service provision system that utilizes the user's reading content (for example, content 113, etc.), A means of accumulating reading content for each user (for example, a shared DB80, etc.), An analysis means (for example, analysis phases 116, S293, S303, etc.) for analyzing the reading content stored in the storage means and extracting the user's intellectual factors, The system includes a service provision means (for example, utilization phases 117, S293, S303, etc.) that provides services utilizing the user's intellectual factors based on the results analyzed by the aforementioned analysis means.

[0334] This configuration allows us to provide services that effectively utilize the user's reading content.

[0335] (2) In (1) above, the service provision means includes persuasive communication generation means (e.g., S293, etc.) that generates communication means (e.g., presentation, etc.) for persuading the user by applying the psychology of persuasion to the intellectual factors of the user to be persuaded.

[0336] (3) In (1) or (2) above, the service provision means includes a means for selecting a user who has intellectual factors suitable for carrying out a particular project (e.g., S303).

[0337] The embodiments disclosed herein should be considered in all respects to be illustrative and not restrictive. The scope of this invention is indicated by the claims rather than by the foregoing description, and all modifications within the meaning and scope equivalent to the claims are intended to be included. [Explanation of Symbols]

[0338] 1 Various AI groups, 3 Publisher groups, 4 IPFS, 50 Internet, 51 Server, 54 User terminal, 72 AI registration DB, 73 Content DB, 74 Content rights DB, 75 AI development and operation company, 77 Electronic certificate for content usage rights, 78 Content rights stakeholders, 86 Blockchain, 99 Node, 99b Node, 82 AI platform system, 106 File server, 112 Security token, 113 Reading content utilization system, 777 Security token DB, 987 Knowledge sharing DB, 998 Like icon, 999 Hyperlink

Claims

1. A service delivery system in which artificial intelligence provides services to users, The system includes a reference content usage control means that controls how the reference content, which is the content that the artificial intelligence referenced in order to provide the aforementioned service, can be made available to the user. The aforementioned reference content usage control means is: A means for determining whether a user has acquired a paid usage right that permits the use of the aforementioned reference content, The system includes an allowance control means that controls whether a user is permitted to use the reference content, provided that the user is determined to have acquired the right of use by the right of use determination means, A service provision system in which the consideration paid by a user who has acquired the aforementioned usage rights is given to the person who holds the rights associated with the aforementioned reference content.

2. The aforementioned reference content usage control means includes a presentation control means that performs control over presenting specific information identifying the reference content to the user, The service provision system according to claim 1, wherein the permitting control means controls the use of the reference content identified by the specific information on the condition that the user who performed the access operation to the specific information is a user who has been determined by the usage right determination means to have acquired the usage right.

3. The aforementioned reference content usage control means includes a presentation control means that performs control over presenting specific information identifying the reference content to the user, The aforementioned tolerance control means is An access determination means for determining whether an operation to access the presented specific information has been performed by a user who has been determined to have acquired the right to use the information by the aforementioned right to use determination means, The service provision system according to claim 1, wherein, on the condition that the access determination means determines that an access operation has been performed, control is performed to allow the user to use the reference content identified by the specific information.

4. A service delivery system in which artificial intelligence provides services to users, Presentation control means that controls the presentation of specific information to the user that identifies the reference content, which is the content that the artificial intelligence used as a reference in order to provide the aforementioned service. The system includes a purchase page display control means that controls the display of the purchase page for the reference content identified by the aforementioned specific information, A service provision system in which the consideration paid by a user who purchases reference content identified by the specified information from the displayed purchase page is given to the person who holds the rights associated with the content.

5. A service delivery system in which artificial intelligence provides services to users, The aforementioned artificial intelligence provides a service provision means for providing a service to a user that references the content, The service provision means includes a reference content usage control means that controls how the reference content, which is the content that the artificial intelligence referenced when providing the service, can be used by the user. The aforementioned service provision means is A means for determining whether a user has acquired a paid usage right that permits the use of the aforementioned reference content, A service permission means that permits the provision of the service to a user on the condition that the user has been determined to have acquired the right of use by the right of use determination means, A service provision system in which the consideration paid by a user who has acquired the aforementioned usage rights is given to the person who holds the rights associated with the aforementioned reference content.

6. The service further includes a reference content designation means for specifying content that a user seeking to receive services from the aforementioned service provision means should refer to, The service provision means provided above provides the user with a service that takes into account the content specified by the user. The service provision system according to claim 5, wherein the service permission means permits the provision of the service to the user on the condition that the user has acquired a paid usage right that permits the use of content specified by the user.

7. The aforementioned reference content includes books, The service provision system according to claim 1, wherein the rights associated with the aforementioned reference content include the copyright of the aforementioned book.

8. The service provision system according to claim 2, wherein the permitting means includes viewing permitting means that permits a user who has selected the specific information to view the reference content identified by the specific information, on the condition that the user has acquired the usage rights determined by the usage rights determination means.

9. The system further includes a purchase page display control means that controls the display of the purchase page for the reference content identified by the specified information when a user who has not acquired the aforementioned usage rights selects the specified information. The service provision system according to claim 2, wherein a user can acquire the right to use the reference content when the user purchases the reference content.

10. The service provision system according to claim 1, further comprising means for granting usage rights to a content transferee when a user who has acquired the aforementioned usage rights transfers content that is permitted to be used by said usage rights.

11. A service provision system according to any one of claims 1 to 8, further comprising a content provision means that allows the artificial intelligence to provide content, on the condition that the artificial intelligence is an artificial intelligence that presents the specific information identifying the aforementioned reference content to the user by a presentation control means that controls the presentation of the specific information identifying the aforementioned reference content to the user.

12. A service delivery method in which artificial intelligence provides services to users, This includes a usage control step that controls how to make reference content, which is content that the artificial intelligence used as a reference for providing the aforementioned service, available to the user. The aforementioned usage control step is: A usage rights determination step that determines whether the user has acquired paid usage rights that permit the use of the aforementioned reference content, The system includes an allowance control step that controls whether a user is allowed to use the reference content, provided that the user is determined to have acquired the right to use the reference content in the right to use determination step. A service provision method in which the consideration paid by the user who has acquired the aforementioned usage rights is given to the person who holds the rights associated with the aforementioned reference content.

13. A program that is used to provide services to users using artificial intelligence and is executed on a computer, To the aforementioned computer, The artificial intelligence performs a usage control step to control how to make the reference content, which is the content that the artificial intelligence used as a reference for providing the aforementioned service, available to the user. The aforementioned usage control step is: The system includes the step of providing a service to a user that allows the use of the aforementioned reference content, provided that the user has acquired a paid usage right that permits the use of the aforementioned reference content. Furthermore, the computer is made to perform a step in which it accepts an operation in which a user pays a fee to acquire the usage rights. A program in which the consideration paid by the user who acquired the aforementioned usage rights is granted to the person who holds the rights associated with the aforementioned reference content.

14. A server used in a service delivery system where artificial intelligence provides services to users, Equipped with a processor and memory, The aforementioned processor, In order to provide the aforementioned service, the artificial intelligence performs a reference content usage control process to control how the reference content, which is the content it referenced, can be made available to the user. The aforementioned reference content usage control process is: A usage rights determination process that determines whether the user has acquired a paid usage right that permits the use of the aforementioned reference content, The process includes an allowance control process that controls whether a user is allowed to use the reference content, provided that the user is determined to have acquired the right of use by the right of use determination means, A server on which the consideration paid by the user who acquired the aforementioned usage rights is granted to the person who holds the rights associated with the aforementioned reference content.

Images