JP2025104457A - Learning data rights management system - Google Patents

Abstract

To provide a rights management system for learning data used for machine learning models such as generative AI.SOLUTION: A rights management system allocates a learning data ID to learning data to be used for generating a generative AI model for generating a new content or new data, and receives registration of rights information of learning data to which the learning data ID is allocated. Registration of the generative AI model that is generated using licensed learning data is received from a model developer, and model registration information associated with the learning data ID of the used learning data, is generated. License applications for the registered generative AI model are received from business operators, usage fees corresponding to the rights information of the learning data associated with the licensed generative AI model, are calculated for the business operators who obtain business revenues through a content or data generated by the registered generative AI model, and the usage fees deposited from the business operators are aggregated and a dividend is calculated for each learning data provider.SELECTED DRAWING: Figure 1

Description

An embodiment of the present invention relates to a technology for managing rights to training data used in machine learning models such as generative AI models.

The generation of machine learning models such as AI models and businesses using machine learning models are thriving. In particular, in recent years, technology that uses artificial intelligence known as generative AI to generate new content and new data has been attracting attention.

Generative AI can learn patterns and other information based on training data and generate new content and new data. This generative AI is a machine learning model built using machine learning techniques such as deep learning.

Publisher: National Diet Library Research and Legislative Reference Bureau, Author: Takuji Amemiya, Economy, Trade and Industry Division, Title: "Trends and Issues of NFTs: Focusing on Content Holder Protection and Consumer Protection", Publication Date: April 4, 2023 (No. 1232), URL: https://dl.ndl.go.jp/view/prepareDownload?itemId=info:ndljp/pid/12767935

The purpose is to provide a rights management system for learning data.

The rights management system of the embodiment performs rights management of learning data used in a generative AI model that generates new content or new data. The rights management system has a learning data management unit that assigns a learning data ID to the learning data of the learning data provider, accepts registration of rights information for the learning data to which the learning data ID is assigned, and accepts a license application for the learning data from the model developer, a model management unit that accepts registration of a generative AI model generated using the licensed learning data, generates model registration information linked to the learning data ID of the used learning data, and accepts a license application for the registered generative AI model from a business, and a usage fee management unit that calculates a usage fee according to the rights information of the learning data linked to the licensed generative AI model for a business that earns business revenue through content or data generated by the generative AI model, and calculates a distribution amount for each learning data provider by aggregating the usage fees paid by the business.

1 is a schematic explanatory diagram of a rights management system according to a first embodiment; 1 is a diagram showing a system configuration and functional blocks of a right management system according to a first embodiment; FIG. 11 is a diagram showing a processing flow centered on a learning data provider in the right management system of the first embodiment. FIG. 2 is a diagram showing a processing flow centered on a model developer in the right management system of the first embodiment. FIG. 2 is a diagram showing a process flow centered on the business operator in the right management system of the first embodiment. 1 is a diagram illustrating a distribution process flow and a distribution mechanism for a training data provider in the right management system of the first embodiment. FIG.

The following describes the embodiment with reference to the drawings.

Machine learning models are generated through a learning process using training data. Machine learning models can be broadly divided into two types.

The first machine learning model is an AI model that organizes and classifies information to make predictions and inspections, and examples of this include image recognition models and voice recognition models. The characteristic of this first machine learning model is that it automatically performs predetermined actions and outputs appropriate answers, such as prediction results and inspection results for input data, based on the learned data.

On the other hand, the second type of machine learning model is a generative AI model that generates (creates) content and data, and there are various types such as text generation AI models, voice generation AI models, image generation AI models, and video generation AI models. Generative AI models are characterized by newly generating content such as sentences (text), voice, music, images, and videos.

In recent years, the emergence of generative AI has sparked a flurry of opinions and issues about how training data should be handled. For example, a voice generation AI that has been trained using the voice data of a voice actor (actor, narrator) who appears in audio or video works only as a voice can generate new audio content that relies on the voice actor's voice data. Regardless of the debate over whether the content is created by a person or by a generative AI, it cannot be said that those who hold the rights to the training data are unrelated to the revenue earned by a voice generation AI that uses their own voice data as training data.

Because new content and new data generated by generative AI in this way depend on the training data used to build the generative AI, unregulated use of the training data is not an appropriate situation from the perspective of protecting those who have rights to the training data. There is an urgent need to create a platform that protects those who have rights to the training data; in other words, a platform that can effectively connect the parties involved when using generative AI to create new content and businesses.

The rights management system of the embodiment provides a rights management platform that centrally manages training data providers, model developers, and businesses that conduct business using generative AI models, and realizes a mechanism for profit sharing for training data that contributes to generative AI models that generate business revenue.

From another perspective, businesses may incur risks when using generative AI models that use training data of unknown origin. For example, if copyrighted work is used in the training data, permission may be required. Therefore, if it is possible to clearly understand what training data was used to develop a generative AI model, risks can be avoided and the quality based on the training data and the value of the generative AI model can be promoted. Model developers can also clearly state the origin of the training data to businesses and promote it to business partners. Through a rights management platform, it is possible to improve the development of generative AI models and the value of businesses that use generative AI models.

First Embodiment
1 to 6 are diagrams for explaining the first embodiment.

Figure 1 is a schematic diagram of a rights management system for training data used in machine learning models, including generative AI models. The rights management system includes a management device 100. The management device 100 provides a rights management platform for use by training data providers, model developers, and businesses.

In the following explanation, a generative AI model will be used as an example of a machine learning model. The learning data is data used in the generative AI model, such as sentences (text data), audio data, music data, image data, and video data, and the rights management platform of this embodiment performs rights management of the learning data used in the learning process (model generation process) of the generative AI model, regardless of the type or content of the learning data.

A person who has rights to training data is called a training data provider. For example, if the training data is voice data, the training data provider is an individual who provides the voice or an organization such as a company to which that individual belongs. For example, this would include an individual such as a voice actor or actor, or a company to which a voice actor or actor belongs.

A model developer is a developer of a machine learning model that performs learning processing using the learning data provided by the learning data provider and builds and generates a generative AI model. There are no particular limitations on the type or use of the generative AI model registered in the rights management platform, as long as it is a generative AI model that uses the learning data registered in the rights management platform. Note that a model developer can develop not only a generative AI model, but also a computer system that includes a generative AI model and provide it to a business operator. A model developer is an individual or an organization such as a company.

An operator is a person who conducts a business by receiving a generative AI model from a model developer. For example, an operator may purchase a generative AI model from a model developer or purchase the right to use a computer system including a generative AI model provided by a model developer, and provide content or data generated by the generative AI model or a service using the content or data generated by the generative AI model. An operator may also enter into a business agreement with a model developer and request the development of a generative AI model. An operator is an individual or a company or other organization that operates a profit-making business and earns revenue by operating a generative AI model or a computer system including a generative AI model.

Training data providers, model developers, and businesses register as users in preparation for using the rights management system (management device 100).

The learning data provider registers the learning data in the rights management system. The rights management system identifies the registered learning data. Specifically, it assigns a learning data ID to the learning data. It then accepts registration of rights information for the learning data to which the learning data ID has been assigned. With the learning data identified and the rights information registered, the learning data is ready for licensing.

The learning data can be registered as multiple data groups, i.e., a dataset. When a dataset made up of multiple learning data groups is registered, a learning data ID is assigned so that the dataset and each of the multiple learning data included in the dataset can be identified. The learning data does not have to be registered as a dataset group; for example, it can be controlled so that a single learning data can be registered.

Model developers can use functions such as searching for training data registered in the rights management system, viewing rights information attached to training data, and applying for a license. The rights management system accepts license applications for training data from model developers and performs the license processing. Model developers can obtain licensed training data from the rights management system.

The model developer uses the acquired learning data to generate a generative AI model. The model developer registers the generative AI model generated using the learning data in the rights management system. The rights management system accepts the registration of the generative AI model, and generates model registration information linked to the learning data ID of the used learning data. In other words, the generated generative AI model is linked to the used learning data, and the generative AI model is identified. This makes it possible to license the generative AI model.

The model developer can obtain the training data in two ways: in the first case, the training data is downloaded from the management device 100 via the rights management system, and in the second case, the training data is obtained without going through the rights management system. In the second case, the training data can be obtained directly from the training data provider, for example.

In the first case, since the learning data is downloaded from the rights management system, the generative AI model is generated using the learning data that has been licensed for use. In the second case, since the learning data is not provided via the rights management system, the generative AI model may be generated using learning data that has not been licensed for use on the rights management system. In this case, the model developer can later apply for a license for the learning data used to generate the generative AI model. For example, when registering the generative AI model, the model developer can search for the learning data used and link the learning data registered in the rights management system to the generative AI model. Even in the second case, the model developer can access the rights management system and apply for a license for the relevant learning data before generating the generative AI model using the learning data. In this case, the generative AI model will be generated using the licensed learning data.

Businesses conduct individual transactions with model developers. Businesses acquire desired generative AI models from model developers, and obtain the right to use computer systems that include generative AI models provided by model developers. Businesses submit license applications to the rights management system for generative AI models to be used in their business. The rights management system accepts license applications from businesses for registered generative AI models, and links the business to the generative AI model. This links the business that uses the generative AI model to the training data.

Here, we will explain the licenses for use of the training data and the generated AI model. The license for use of the training data is a process that links the training data with the model developer. In other words, the license for use of the training data leaves a history of which training data the model developer has used. In this case, the form of the license is arbitrary, and may be, for example, a mechanism that applies a publicly known license agreement, or a mechanism in which the model developer authenticates (permits) the use of the training data registered in the rights management system. Therefore, a license application for use of the training data is an act by which the model developer offers to use (have used) the training data registered in the rights management system, and the process of linking the model developer who has offered to use the training data with the relevant training data (training data provider) is the license process.

Similarly, the license for a generated AI model is a process that links a model developer with a business operator and is a process for keeping a history of which generated AI model the business operator is using to conduct business. The format of the license for a generated AI model is also arbitrary, and may be, for example, a mechanism that applies a publicly known license agreement. Therefore, a license application for a generated AI model is an act by a business operator offering to use (have used) a generated AI model registered in a rights management system, and the process of linking the business operator that has offered to use the generated AI model with the relevant generated AI model is the license process.

Next, the rights management system calculates the usage fee to be paid by the business operator according to the rights information of the learning data linked to the licensed generative AI model, and performs a process of billing the business operator for the usage fee. At this time, the business operator registers (reports) business information (revenue-related information) including business revenue related to the licensed generative AI model to the rights management system. The rights management system can calculate the usage fee to be billed to the business operator based on the reported business revenue. The business operator pays the billed usage fee. The rights management system tallys up the usage fees paid by the business operators and calculates distribution amounts for each learning data provider. The calculated distribution amounts are deposited into the bank account of the learning data provider.

In this way, the rights management system provides a rights management platform that centrally manages training data providers, model developers, and businesses, and can realize a mechanism for sharing profits from training data that contributes to generative AI models that generate business revenue.

Figure 2 is a diagram showing the system configuration and functional blocks of the rights management system of this embodiment. The rights management system is configured to include a management device 100. The management device 100 includes a communication device 110, a control device 120, and a storage unit 130. Training data providers, model developers, and businesses connect to the management device 100 from their respective user terminals via a network such as an IP network, and use the rights management platform.

The control device 120 includes a user interface control unit 121, a user management unit 122, a learning data management unit 123, a model management unit 124, and a usage fee management unit 125.

The user interface control unit 121 is a processing unit that provides various screens for using the rights management system to each user terminal of the training data provider, model developer, and business operator, and controls the screens in response to the operation of the user terminal.

The user management unit 122 provides a user registration function through a user registration screen. The user management unit 122 includes a registration control unit 1221. The registration control unit 1221 stores the user information input for the learning data provider through the learning data provider registration screen in the storage device 130 for each learning data provider. The user information of the learning data provider includes a learning data provider ID, a login password, registrant information (name, address, contact information, email address, etc.), organization information (organization name, address, contact information, email address, etc.), a registration type (type indicating the learning data provider), transfer information (transfer information for usage fees, etc.), etc.

Furthermore, the registration control unit 1221 stores the user information entered for the model developer through the model developer registration screen in the storage device 130 for each model developer. The user information for the model developer includes the model developer ID, login password, registrant information (name, address, contact information, email address, etc.), organizational information (organization name, address, contact information, email address, etc.), registration type (type indicating the model developer), etc.

The registration control unit 1221 also stores the user information entered for the business through the business registration screen in the storage device 130 for each business. The user information for the business includes the business ID, login password, registrant information (name, address, contact information, email address, etc.), organization information (organization name, address, contact information, email address, etc.), registration type (type indicating the business), and fee payment information (credit card information, account transfer information, etc.).

Next, the learning data management unit 123 includes a learning data registration control unit 1231, a learning data rights information management unit 1232, a learning data identity management unit 1233, and a learning data usage permission management unit 1234.

The learning data registration control unit 1231 assigns a learning data ID to the learning data uploaded to the management device 100 via the learning data registration screen and stores it in the storage device 130.

The learning data right information management unit 1232 controls the input of right information through the user terminal of the learning data provider, and stores the input right information in the storage device 130 in association with the learning data ID. The right information can include permission for use, permission range, unavailable content information, and rate setting information. The permission range is information that specifies the purpose and range of use of the generated AI model generated using the learning data. For example, it is possible to set the purpose of use of the learning data such that the generated AI model is permitted when used for non-commercial purposes but not permitted when used for commercial purposes, or to set the use such that the generated AI model is permitted when used for in-house training. The unavailable content information is information that specifies the scope of provision and age restrictions of content and data generated by the generated AI model, or services provided using the generated AI model. For example, it is possible to set restrictions such as not permitting application to content and data provided to children under the age of 12, or to the generated AI model used for services.

The rate setting information is information that specifies the usage fee for the training data. The usage fee can be specified arbitrarily, but it is possible to set, for example, a usage fee based on the number of training data used in the training process of the generated AI model, a usage fee based on the sales price of the generated AI model, a usage fee based on the profits obtained by the business operator from the business using the generated AI model, etc.

The learning data identity management unit 1233 performs identity check information generation processing and identity check processing for the learning data. For example, it generates a hash value of the learning data using a hash function based on a predetermined algorithm. This hash value is managed in association with the learning data ID, and is used to confirm (prove) that the learning data is genuine learning data that has not been tampered with or edited. For example, when there is an identity check request from the model developer, the learning data identity management unit 1233 can generate a hash value of the learning data received from the model developer's user terminal, compare it with the registered hash value, and provide the model developer with the identity check result.

The training data identity management unit 1233 can also provide the model developer's user terminal with an identity check information generation program including a hash function based on a predetermined algorithm. The model developer executes the provided identity check information generation program on the training data to generate a hash value. The model developer transmits the generated hash value from the user terminal to the management device 100. The training data identity management unit 1233 compares the received hash value with the registered hash value, performs identity check processing, and provides the result to the user terminal.

This learning data identity check function is a function that determines whether the learning data is learning data registered in the rights management system, and is used by the model developer to confirm the authenticity of the learning data, as well as to confirm whether learning data obtained from a source other than the rights management system is learning data registered in the rights management system. In preparation for generating a generative AI model, the model developer can check in advance whether the learning data is registered. If the learning data obtained from a source other than the rights management system is registered learning data, the model developer can apply for a license to use it from the rights management system.

The learning data usage permission management unit 1234 performs usage permission processing for the learning data registered in the rights management system. The learning data usage permission management unit 1234 accepts a usage permission application for the registered learning data from the user terminal of the model developer, and performs usage permission processing for each learning data ID.

The learning data management unit 123 centrally manages the learning data attribute information, including the learning data ID, identity check information, learning data provider ID, rights information ID, and description of the data set, for each piece of learning data. The learning data management unit 123 can then control the downloading of the learning data that has been approved for use to the user terminal of the model developer.

As mentioned above, the learning data may include a dataset made up of multiple data, but since the dataset has a large data volume, it may be difficult to upload it to the management device 100. Therefore, the learning data itself may not be uploaded to the management device 100, and only the registration information of the learning data may be input to the management device 100.

Specifically, the learning data management unit 123 controls the above-mentioned identity check information generation program managed by the learning data identity management unit 1233 so that it can be provided to the user terminal of the learning data provider. The learning data provider then executes the provided identity check information generation program on the learning data to generate a hash value. The learning data provider transmits the generated hash value to the management device 100. The learning data registration control unit 1231 can assign a learning data ID to the received hash value, register it, and generate learning data attribute information.

In this case, the model developer can obtain the training data by means other than the management device 100. For example, the model developer can obtain the training data directly from the training data provider, or through a download site used by the training data provider. When the model developer obtains the training data through a route other than the management device 100, the model developer can access the management device 100, check the identity of the acquired training data through an identity check function provided by the training data identity management unit 1233, and apply for permission to use the acquired training data through a license function provided by the training data license management unit 1234.

The model management unit 124 is composed of a model registration control unit 1241, a model identity management unit 1242, and a model usage permission management unit 1243.

The model registration control unit 1241 assigns a model ID to each generated AI model generated using the learning data registered in the management device 100, and generates model registration information linking the learning data ID of the used learning data with the model ID. Specifically, the model developer registers the generated AI model generated using the learning data registered in the rights management system in the rights management system. The model registration control unit 1241 accepts model property information through the generated AI model registration screen and assigns a model ID. Then, in conjunction with the registration of the generated AI model, an identity check information generation program including a hash function for generating identity check information is provided to the user terminal. The model developer executes the identity check information generation program for the source data of the generated AI model including the model property information, and generates a hash value. The generated hash value is input together with the model property information through the generated AI model registration screen.

The model registration control unit 1241 generates model registration information including a model ID, a model developer ID, model property information, and identity check information (hash value), and stores it in the storage device 130.

Here, the model property information will be described. The model property information includes the learning data ID and the number of pieces of data used in the learning process of the generating AI model, and can be configured to include any property information of the generating AI model in addition to the learning data ID and the number of pieces of data. The model registration control unit 1241 can be configured to generate model registration information including identity check information generated for the source data of the generating AI model including the model property information. In this way, the hash value that checks the identity of the generating AI model includes the learning data ID and the number of pieces of data used, so that even if the generating AI model uses the same learning data ID, if the number of pieces of data is different, different hash values (individual model registration information) are generated. The rights management system registers and manages the generating AI model, the learning data ID, and the number of pieces of data in an inseparable state.

The model identity management unit 1242 provides a function similar to that of the learning data identity management unit 1233 of the learning data management unit 123. In other words, it provides a function to determine whether or not a generative AI model is a generative AI model registered in a rights management system, and is used by businesses using a generative AI model to confirm whether the generative AI model is registered in the rights management system. Businesses can check in advance whether a generative AI model is registered or not in preparation for conducting a business using a generative AI model. If the generative AI model to be used is a registered generative AI model, the business submits a license application to the rights management system.

The model license management unit 1243 performs a license process for the business operator to use the generated AI model. The model license management unit 1243 accepts a license application for the generated AI model registered from the business operator's user terminal, and performs a license process that links the model ID with the business operator.

The usage fee management unit 125 includes a business information reception unit 1251, a usage fee calculation unit 1252, a usage fee payment management unit 1253, and a usage fee distribution management unit 1254.

The business information receiving unit 1251 provides a function for receiving revenue-related information of a business that uses a generated AI model that has been licensed for use from the user terminal of the business. The usage fee calculation unit 1252 calculates the usage fee according to the rights information linked to the learning data based on the revenue-related information. Specifically, the usage fee to be paid by the business is calculated based on the rate setting information included in the rights information and the business revenue from using the generated AI model included in the revenue-related information. The usage fee payment management unit 1253 performs billing processing for the usage fee to be paid by the business. The billing processing is processing for requesting payment of the usage fee from the business. Note that the system may be configured to include payment processing such as credit settlement or account transfer associated with the billing. The usage fee distribution management unit 1254 tallys up the usage fees paid by the business and calculates the distribution amount for each learning data provider. The usage fee management unit 125 performs processing for paying the calculated distribution amount to the transfer destination of the learning data provider.

The storage device 130 stores user information 1301, learning data 1302, rights information 1303, learning data attribute information 1304, learning data usage permission information 1305, model registration information 1306, model usage permission information 1307, business information 1308, usage fee billing information 1309, and usage fee distribution information 1310.

Figure 3 shows the processing flow centered on the learning data provider in the rights management system.

As shown in FIG. 3, the learning data provider accesses the management device 100 and performs user registration (S301). The management device 100 assigns a learning data provider ID and performs a registration process for the input user information (S101). The learning data provider logs in based on the learning data provider ID through the user terminal (S302), and the management device 100 performs a login authentication process (S102).

After login authentication, the learning data provider performs a registration process for the learning data (S303). The learning data provider uploads the learning data to the management device 100 through the user terminal. The management device 100 performs a process for generating identity check information for the uploaded learning data (S103). The management device 100 assigns a learning data ID to the uploaded learning data, associates the learning data ID with the identity check information, and stores the learning data (S104). Along with uploading the learning data, the learning data provider inputs rights information for the learning data to be registered (S304). The management device 100 performs a registration process for the input rights information (S105), assigns a rights information ID to the input rights information, and associates it with the learning data ID. The rights information associated with the learning data ID is stored (S106). The management device 100 generates learning data attribute information through the learning data registration process and the rights information registration process, and stores it in the storage device 130.

Figure 4 shows the process flow centered on the model developer in the rights management system.

As shown in FIG. 4, the model developer accesses the management device 100 and performs user registration (S401). The management device 100 assigns a model developer ID and performs registration processing of the input user information (S111). The model developer logs in based on the model developer ID through a user terminal (S402), and the management device 100 performs login authentication processing (S112).

After login authentication, the model developer can search for learning data (S403). The management device 100 performs a search process based on the search conditions specified by the model developer, and provides the search results to the model developer's user terminal (S113).

The management device 100 controls the rights information of the searched learning data so that it can be viewed (S114), and the model developer checks the rights information through the user terminal (S404).

The model developer applies for a license to use the desired learning data (S405). The management device 100 accepts the license application, performs a license process, and links the requested learning data to the model developer (S115). The management device 100 controls the licensed learning data so that it can be downloaded, and provides the learning data to the model developer's user terminal based on a learning data acquisition request from the model developer (S406) (S116).

The model developer develops and generates a generative AI model using the learning data registered in the management device 100 (S407). The model developer registers the generated generative AI model in the management device 100. At this time, the model developer records the learning data ID and number of pieces of data of the used learning data in the model property information of the generated generative AI model.

The model developer acquires the identity check information generation program from the management device 100 (S409, S117), and executes the identity check information generation program on the user terminal for source data including model property information in which the learning data ID and the number of data are recorded, to generate identity check information (S410).

The model developer inputs identity check information and model property information through the model registration screen (S411). The management device 100 assigns a model ID, generates model registration information including identity check information and model property information (S118), and saves it (S119).

Figure 5 shows the process flow centered on the business operator in the rights management system.

As shown in FIG. 5, the business operator accesses the management device 100 and performs user registration (S501). The management device 100 assigns a business operator ID and performs registration processing of the input user information (S131). The business operator logs in based on the business operator ID through the user terminal (S502), and the management device 100 performs login authentication processing (S132).

After login authentication, the business operator can search for a generated AI model (S503). The management device 100 performs a search process based on the search conditions specified by the business operator, and provides the search results to the business operator's user terminal (S133).

The business operator applies for a license to use the desired generated AI model (S404). The management device 100 accepts the license application and generates usage fee billing information based on the rights information of the learning data linked to the generated AI model (S134). The management device 100 provides the generated usage fee billing information to the business operator's user terminal, and the business operator expresses its intention to agree to the license through the user terminal (S505). The management device 100 performs a license process based on the business operator's agreement to the license, and links the requested generated AI model to the business operator (S135).

The usage fee billing information is, for example, usage fee billing information for each business operator ID and model ID, which is generated based on the rights information of the learning data linked to the generated AI model. The usage fee billing information can be configured to include revenue-related information registered by the business operator, usage fee billing history, and usage fee payment history (deposit history).

The business uses the registered machine learning model to conduct business and create revenue-related information related to usage fee billing information (S506). The business registers (reports) the created revenue-related information to the management device 100 (S507).

The management device 100 calculates the usage fee according to the revenue-related information based on the usage fee billing information (S136). The management device 100 performs billing processing for the calculated usage fee (S137), and the business operator confirms the billing details (S508).

The business operator performs the payment process for the invoiced usage fee, with the operator who operates the management device 100 (rights management platform) as the payee (S509). As described above, the business operator may appropriately apply known methods such as credit card payment or account transfer to pay the usage fee. The management device 100 confirms the payment of the usage fee from the business operator (S138), creates a payment history (payment history), and updates the usage fee billing information (S139).

Figure 6 is a diagram explaining the distribution process flow and distribution mechanism for training data providers in the rights management system.

As shown in FIG. 6, the management device 100 performs a usage fee distribution process at regular intervals (S151). Specifically, the management device 100 aggregates usage fee billing information (usage fee billing log) (S152) and distributes the aggregated usage fee by learning data ID used in the generating AI model (S153). In other words, the distribution amount is calculated by learning data ID. The management device 100 performs a process of paying the calculated distribution amount to the transfer destination of the corresponding learning data provider (S154). After the payment process is completed, a notification of the payment is sent to the user terminal of the learning data provider (S155).

As a rights management platform, the management device 100 identifies the learning data of the learning data provider and links it to the generated AI model generated using the learning data. Then, by linking the generated AI model to a business operator that actually conducts a business using the generated AI, it is possible to link the business operator's business using the generated AI model to the learning data provider. This makes it possible to form a route for distributing the business operator's business profits to the learning data provider.

As shown in Figure 6, the model developer and the business operator may belong to the same organization. In this case, the model developer and the business operator will each register as separate users, and the business operator and the training data provider will be linked after obtaining the training data usage license and model usage license described above.

The rights management system of this embodiment can provide a rights management platform that can effectively connect the parties involved when creating new content or business using generative AI.

Through the rights management platform, training data providers, model developers, and businesses that use generative AI models can be managed in a centralized manner, and profits can be distributed to training data providers for the training data that contributes to generative AI models that generate business revenue.

Although the embodiment has been described above, the rights management system is not limited to generative AI models, but can also be applied to the first machine learning model described above. In other words, the training data may contribute to the machine learning model not only to business revenue from new content or new data generated by the generative AI model, but also to business revenue from the results of predictions and judgments made by machine learning models such as conventional image recognition AI models and voice recognition AI models. Therefore, regardless of whether the machine learning model generates new content or new data, it is possible to protect those who have rights to the training data.

Furthermore, each function constituting the rights management system can be realized by one computer device, or distributed across multiple computer devices. Furthermore, the memory area that stores the learning data and can provide the learning data to the model developer's user terminal may be configured to be provided separately from this rights management system. For example, the rights management system may be constructed in cooperation with an external storage device that provides a data service for storing and downloading data.

In addition, the training data identity management unit 1233 and the model identity management unit 1242 may be configured as a single identity management unit, and configured to provide the functions of generating identity check information for both the training data and the machine learning model including the generated AI model, providing an identity check information generation program, and matching with registered identity check information.

In addition, each function that constitutes the rights management system can be realized by a program, and computer programs prepared in advance to realize each function are stored in an auxiliary storage device. A control unit such as a CPU reads the program stored in the auxiliary storage device into a main storage device, and the control unit executes the program read into the main storage device, thereby operating the functions of each unit.

The above program can also be provided to a computer in a state in which it is recorded on a computer-readable recording medium. Examples of computer-readable recording media include optical disks such as CD-ROMs and Blu-ray (registered trademark) Disc Rewritable, phase-change optical disks such as DVD-ROMs, magneto-optical disks such as MO (Magneto Optical), magnetic disks such as floppy (registered trademark) disks and hard disks, and memory cards such as SD memory cards and USB flash drives. Also included as recording media are hardware devices such as integrated circuits (IC chips such as ROM and RAM) that are specially designed and configured for the purpose of the present invention.
Furthermore, the present invention, including the above-mentioned program, is not limited to being executed on the architecture of a von Neumann-type computer, but may be executed on the architecture of a so-called non-von Neumann-type computer, such as a neurocomputer based on the mechanism of neural circuits in the brain or a quantum computer that applies quantum mechanics to information processing.

Although an embodiment of the present invention has been described, this embodiment is presented as an example and is not intended to limit the scope of the invention. This new embodiment can be embodied in various other forms, and various omissions, substitutions, and modifications can be made without departing from the gist of the invention. These embodiments and their modifications are included in the scope and gist of the invention, and are included in the scope of the invention and its equivalents described in the claims.

100 Management device 110 Communication device 120 Control device 121 User interface control unit 122 User management unit 1221 Registration control unit 123 Learning data management unit 1231 Learning data registration control unit 1232 Learning data right information management unit 1233 Learning data identity management unit 1234 Learning data usage permission management unit 124 Model management unit 1241 Model registration control unit 1242 Model identity management unit 1243 Model usage permission management unit 125 Usage fee management unit 1251 Business information reception unit 1252 Usage fee calculation unit 1253 Usage fee payment management unit 1254 Usage fee distribution management unit 130 Storage device 1301 User information 1302 Learning data 1303 Rights information 1304 Learning data attribute information 1305 Learning data usage permission information 1306 Model registration information 1307 Model usage permission information 1308 Business information 1309 Usage fee billing information 1310 Royalty distribution information

Claims (12)

A rights management system for learning data used in a generative AI model that generates new content or new data,
a learning data management unit that assigns a learning data ID to the learning data of the learning data provider, receives registration of rights information for the learning data to which the learning data ID is assigned, and receives a license application for the learning data from the model developer;
a model management unit that accepts registration of a generative AI model generated using the licensed learning data, generates model registration information linked to the learning data ID of the used learning data, and accepts a license application for the registered generative AI model from a business operator;
a usage fee management unit that calculates usage fees according to the rights information of the learning data linked to the licensed generating AI model for businesses that earn business revenue through the content or data generated by the generating AI model, and that aggregates the usage fees paid by the businesses to calculate distribution amounts for each learning data provider;
13. A rights management system comprising: The learning data management unit
a learning data registration control unit that assigns a learning data ID to the learning data uploaded from the terminal of the learning data provider and registers the learning data in a predetermined storage area;
a learning data right information management unit that controls input of right information through a terminal of a learning data provider and registers the input right information in a predetermined storage area in association with a learning data ID;
a learning data license management unit that accepts a license application for the registered learning data from a model developer's terminal and performs a license process for each learning data ID;
The rights management system according to claim 1 , wherein the learning data management unit controls the learning data that has been permitted for use so that the learning data can be downloaded to a terminal of a model developer. The learning data management unit
a learning data identity management unit that executes an identity check information generation program based on a hash function configured by a predetermined algorithm for the uploaded learning data to generate identity check information;
3. The rights management system according to claim 2, wherein the learning data registration control unit registers the learning data by linking the learning data ID with identity check information. The learning data management unit
a training data identity management unit that provides a training data provider's terminal with an identity check information generation program based on a hash function configured with a predetermined algorithm;
a learning data registration control unit that receives identity check information generated by executing an identity check information generation program on the learning data from a terminal of a learning data provider, assigns a learning data ID to the identity check information, and registers the identity check information in a predetermined storage area;
a learning data right information management unit that controls input of right information through a terminal of a learning data provider and registers the input right information in a predetermined storage area in association with a learning data ID;
The rights management system according to claim 1, further comprising a learning data license management unit that accepts a license application for registered learning data from a model developer's terminal and performs license processing for each learning data ID. The rights management system according to claim 3 or 4, characterized in that the training data identity management unit receives training data or identity check information of the training data from each terminal of the training data provider, the model developer and/or the business operator, compares identity check information generated by executing an identity check information generation program on the received training data or the received identity check information with registered identity check information, and outputs the identity check result. The model management unit
A model registration control unit that assigns a model ID to each generated AI model generated using the learning data and generates model registration information including the learning data ID and the model ID of the used learning data;
A model license management unit that accepts a license application for the registered generated AI model from the business operator's terminal and performs license processing for each model ID;
3. The rights management system according to claim 1, further comprising: The model management unit
A model identity management unit is further provided for providing an identity check information generation program based on a hash function configured by a predetermined algorithm to a terminal of a model developer,
The rights management system described in claim 6, characterized in that the model registration control unit receives identity check information generated by executing an identity check information generation program targeting source data of a generated AI model from a model developer's terminal, and generates model registration information including a learning data ID, a model ID, and identity check information. The generated AI model holds model property information including the learning data ID and the number of pieces of learning data used,
The rights management system described in claim 7, characterized in that the model registration control unit accepts identity check information generated for source data of a generated AI model including model property information from a model developer's terminal, and generates model registration information. The rights management system of claim 7, characterized in that the model identity management unit receives identity check information generated for the source data of the generation AI model from each terminal of the training data provider, model developer, and/or business operator, compares the received identity check information with registered identity check information, and outputs the identity check result. The fee management unit
A business information receiving unit that receives revenue-related information related to business revenue obtained through content or data generated by the licensed generation AI model from a business operator's terminal;
A fee calculation unit that calculates a fee according to the rights information of the learning data associated with the licensed generated AI model based on the revenue-related information;
a fee payment management unit that processes billing for fees paid by businesses;
a fee distribution management unit that aggregates the fees received from the businesses, calculates distributions for each learning data provider, and pays the distributions to the bank accounts of the learning data providers;
2. The rights management system of claim 1, further comprising: A program executed by a computer that performs rights management of learning data used in a generative AI model that generates new content or new data,
a first function of assigning a learning data ID to the learning data of the learning data provider, accepting registration of rights information for the learning data to which the learning data ID is assigned, and accepting an application for permission to use the learning data from the model developer;
A second function of accepting registration of a generative AI model generated using the licensed learning data, generating model registration information linked to the learning data ID of the used learning data, and accepting a license application for the registered generative AI model from the business operator;
A third function of calculating a usage fee according to the rights information of the learning data linked to the licensed generating AI model for a business operator who obtains business revenue through the content or data generated by the generating AI model, aggregating the usage fees paid by the business operator, and calculating distributions for each learning data provider;
A program to achieve this. A system for managing rights to learning data, comprising:
a learning data management unit that assigns a learning data ID to the learning data of the learning data provider, receives registration of rights information for the learning data to which the learning data ID is assigned, and receives a license application for the learning data from the model developer;
a model management unit that accepts registration of a machine learning model generated using the licensed learning data, generates model registration information linked to a learning data ID of the used learning data, and accepts a license application for the registered machine learning model from a business operator;
a fee management unit that calculates fees to be paid by businesses according to the rights information of the training data linked to the licensed machine learning model, and that aggregates the fees received from the businesses to calculate distributions to each training data provider;
13. A rights management system comprising:

Images