Information processing system, information processing method, and information processing program

Abstract

A processor in an information processing system according to the present invention inputs input data to at least one generative trained model among a plurality of generative trained models which have been trained for respective copyright holders in advance on the basis of learning data provided from the copyright holders. The processor thereby acquires content which corresponds to the input data and has been output from the generative trained model, and outputs the content.

Description

Information processing system, information processing method, and information processing program

[0001] This disclosure relates to an information processing system, an information processing method, and an information processing program.

[0002] Japanese Patent Publication No. 2024-052226 discloses a character modification system that can generate characters that accurately reflect user preferences while eliminating the complexity of character creation. This character modification system has a configuration that performs a component determination process to determine which components of a base character to change based on player-related information already stored or player input information entered by the player, and a modified character generation process that generates a modified character in which the components have been modified from the base character based on the determined components (see, for example, the [Abstract] of Japanese Patent Publication No. 2024-052226).

[0003] Furthermore, Japanese Patent Publication No. 7398723 discloses a technology that assists users in creating images. Specifically, Japanese Patent Publication No. 7398723 discloses an illustration background image generation system in which an illustration background image generation server and multiple user terminals are connected via a network. This illustration background image generation server includes a UI data transmission unit that transmits UI data for generating a UI including multiple tags to a user terminal, a tag receiving unit that receives one or more tags selected by the user from the user terminal, a prompt creation unit that creates a prompt including elements corresponding to the tags, an AI image generation unit that generates an image based on the prompt, and an image output unit that outputs the image.

[0004] Japanese Patent Publication No. 7329293 discloses a technology that assists in preparing visual content suitable as a record of a character's activities. Specifically, the program disclosed in Japanese Patent Publication No. 7329293 causes a computer to function as a means for obtaining feature information representing the characteristics of visual content predicted to be suitable for a specific character by providing a first model input based on profile information set for a specific character to a first model that has learned the relationship between the profile information of multiple reference characters and the characteristics of the visual content that each of the multiple reference characters has made publicly available; a means for obtaining candidate content that is visual content that features the appearance of a specific character and has characteristics corresponding to the second model input by providing a second model input based on the feature information to a second model that has learned to generate images that feature the appearance of a specific character as the subject; and a means for outputting the candidate content.

[0005] Incidentally, content is sometimes generated using so-called generative AI (Artificial Intelligence). In this case, when predetermined input data is provided to the generative AI, content corresponding to that input data is output from the generative AI. Examples of such content include text, images, or videos.

[0006] Generative AI is pre-generated based on collected training data. In this case, the generative AI may be trained using training data for which multiple copyright holders each hold copyright. However, if a generative AI is trained using training data provided by multiple different copyright holders, that generative AI may be undesirable from the perspective of copyright of the training data.

[0007] For example, user A owns the copyright to educational content C. Ａ And learning content C, whose copyright is held by another user B. Ｂ Based on this, let's consider the case where we train a single learning model to create a generative AI. In this case, the generative AI is the learning content C Ａ and learning content C ＢBecause it is learned based on the content C Ａ The style and content C Ｂ The output may sometimes be a mix of different styles.

[0008] For example, if the generative AI is an image-generating AI or a video-generating AI, the AI ​​may output an anime character that is a mix of an anime character appearing in content copyrighted by User A and an anime character appearing in content copyrighted by User B.

[0009] However, since users A and B, who are copyright holders, own the copyright to their own content, it is conceivable that they would not like or would not permit the output of content from a generative AI that mixes their copyrighted content with distinctive content from others (e.g., anime characters), as mentioned above. On the other hand, for example, a copyright holder might tolerate the output of such content from a generative AI, even if their copyrighted content is mixed with general content (e.g., just a landscape image), because their own content would not be excessively affected.

[0010] While the above-mentioned documents disclose technologies for creating content such as anime characters or images, they do not disclose how to manage content used for training generative AI.

[0011] Therefore, even if the technologies disclosed in the above-mentioned documents are used, there is a problem in that when creating content using generative AI, it is not possible to manage the content used for training the generative AI.

[0012] This disclosure is made in view of the above circumstances and aims to provide an information processing system, an information processing method, and an information processing program that can generate content using a generative AI while managing content used for training a generative AI.

[0013] To achieve the above objective, the first aspect of the present disclosure is an information processing system comprising a processor, wherein the processor inputs input data to at least one of a plurality of generative trained models, each pre-trained based on training data provided by a copyright holder, and each of the plurality of generative trained models pre-trained for each copyright holder, thereby acquiring content corresponding to the input data and content output from the generative trained model, and outputting the content.

[0014] Furthermore, the second aspect of this disclosure is an information processing method in which a computer performs a process in which input data into at least one of a plurality of generative trained models, each pre-trained based on training data provided by a copyright holder, and each of the plurality of generative trained models pre-trained for each copyright holder, thereby acquiring content corresponding to the input data and outputting the content from the generative trained model.

[0015] Furthermore, the third aspect of this disclosure is an information processing program that causes a computer to perform a process that involves inputting input data into at least one of a plurality of generative trained models, each pre-trained based on training data provided by a copyright holder, and each of the plurality of generative trained models pre-trained for each copyright holder, thereby acquiring content corresponding to the input data and outputting the content from the generative trained model.

[0016] According to this disclosure, the effect is that content used for training generative AI can be managed while simultaneously generating content using generative AI.

[0017] This figure shows an example of the schematic configuration of the information processing system of this embodiment. This figure shows an example of the functional configuration of the server of this embodiment. This figure illustrates the training content data of this embodiment. This figure shows an example of a generative trained model of this embodiment. This is a schematic block diagram of the computers that function as each device of the information processing system. This is an explanatory diagram illustrating the processing performed by the information processing system of this embodiment. This is an explanatory diagram illustrating the processing performed by the information processing system of this embodiment.

[0018] The embodiments will be described in detail below with reference to the drawings.

[0019] <System Configuration of the Information Processing System> Figure 1 is a block diagram showing the information processing system 10 of the embodiment. As shown in Figure 1, the information processing system 10 of the embodiment includes a server 12, a user terminal 14A operated by a first user, a user terminal 14B operated by a second user, and a user terminal 14C operated by a third user. The server 12, user terminals 14A, 14B, and 14C are connected by a network 15, such as the Internet. In the following, when referring to user terminals 14A, 14B, and 14C collectively, they will simply be referred to as user terminal 14. User terminal 14 is, for example, a smartphone or a personal computer. Although three user terminals 14 are shown in Figure 1, there may be more user terminals 14.

[0020] Figure 2 is a diagram showing an example of the functional configuration of the server 12 in the embodiment. As shown in Figure 2, the server 12 includes a reception unit 20, a learning data storage unit 22, a control unit 24, a generative trained model storage unit 26, and an output unit 28.

[0021] The reception unit 20 receives various data transmitted from the user terminal 14.

[0022] The learning data storage unit 22 stores learning data provided by a plurality of copyright holders. FIG. 3 is a diagram for explaining the data stored in the learning data storage unit 22. As shown in FIG. 3, for example, learning content data C Ｘ provided by copyright holder X, and learning content data C Ｙ provided by copyright holder Y, and learning content data C Ｚ provided by copyright holder Z are stored in the learning data storage unit 22. Note that the learning content data C Ｘ and the learning content data C Ｙ and the learning content data C Ｚ are partitioned by partition P so that their storage areas are separated and data contamination does not occur. In addition, the learning data storage unit 22 also stores common learning content data C. This common learning content data C is content data for which consent of the copyright holder has been obtained regarding the learning and use of the generated trained model.

[0023] Each of the copyright holders X, Y, and Z may not want a generated trained model to be generated based on the learning content data they own and the learning content data owned by others. As described above, for example, the copyright holder may not want content in which the content they own and the characteristic content owned by others are mixed to be generated by the generated trained model. Also, from the perspective of copyright management, it is not preferable for a generated trained model to be learned based on a plurality of contents for which consent of the copyright holder has not been obtained.

[0024] Therefore, in the present embodiment, a generated trained model for each copyright holder is created. Specifically, based on the learning content data provided by each of the copyright holders, a generated trained model for each copyright holder is generated. For example, based on the above learning content data C Ｘ a generated trained model M Ｘ of copyright holder X is created. Also, based on the above learning content data CＹ Based on this, the pre-trained generative model M of copyright holder Y Ｙ The above learning content data C is created. Ｚ Based on this, the pre-trained generative model M of copyright holder Z Ｚ This is created. Furthermore, in this embodiment, when creating a generative trained model, the above generative trained model is trained based on common training content data C, which is general content for which consent has been obtained from other copyright holders. This results in a generative trained model M for each copyright holder. Ｘ , M Ｙ , M Ｚ This is generated. Note that copyright holders X, Y, and Z are so-called IP (Intellectual Property) holders.

[0025] The control unit 24 controls the operation of the server 12. Specifically, the control unit 24 generates a generative trained model for each copyright holder by training a training model using known machine learning techniques based on training data provided by multiple copyright holders stored in the training data storage unit 22. The generative trained model is a so-called generative AI. An example of a generative trained model is a generative AI that generates text, images, or videos. When some input data (e.g., a prompt) is input to this generative trained model, it outputs content data corresponding to that input data. The control unit 24 may also generate a generative trained model for each copyright holder by fine-tuning a generative AI that has already been trained (for example, the Llama (Language Large Models Meta AI) model provided as open source) using the training content data for each copyright holder as described above. The control unit 24 then stores each of the generative trained models for each copyright holder in the trained model storage unit 26. Furthermore, if the generative pre-trained model is an LLM (Large Language Model), the generative pre-trained model for each copyright holder in this embodiment is trained based on training content data separated by partitions, and therefore can also be called a partitioned LLM.

[0026] The trained model storage unit 26 stores each of the generative trained models for each copyright holder. Figure 4 shows an example of a generative trained model for each copyright holder. As shown in Figure 4, for example, the training content data C whose copyright is held by copyright holder X Ｘ A pre-trained generative model M learned based on this Ｘ And, the learning content data C, whose copyright is held by copyright holder Y. Ｙ A pre-trained generative model M learned based on this Ｙ And, the learning content data C, whose copyright is held by copyright holder Z. Ｚ A pre-trained generative model M learned based on this Ｚ and are stored.

[0027] The control unit 24 inputs the input data transmitted from the user terminal 14 into at least one of several pre-trained generative models, each pre-trained for each copyright holder, thereby acquiring content that corresponds to the input data and is output from the generative model.

[0028] For example, the input data transmitted from the user terminal 14 includes identification data and a prompt for selecting a generative trained model. In this case, the control unit 24 selects the target generative trained model from a plurality of generative trained models stored in the trained model storage unit 26 based on the identification data included in the input data.

[0029] For example, the identification data included in the input data is an ID for identifying a pre-trained generative model. Alternatively, for example, the identification data included in the input data may be text data. For example, if the input data is "Please output a townscape image in the style of Studio G," then "Studio G" represents the copyright holder, and the control unit 24 selects a pre-trained generative model corresponding to that copyright holder and generates content using that pre-trained generative model. For example, if the input data is "Please output a townscape image in the style of Anime R," then the control unit 24 selects a pre-trained generative model corresponding to the copyright holder of "Anime R" and generates content using that pre-trained generative model. To realize the above processing, for example, known natural language processing techniques and artificial intelligence techniques are used.

[0030] The control unit 24 then inputs the input data to the selected target generative trained model, thereby obtaining the content output from the generative trained model. The input data may include the input content and a prompt. In this case, the control unit 24 inputs the input content and the prompt to the generative trained model, thereby generating content that combines the input content and the training data used to train the generative trained model.

[0031] For example, when the input data includes an image and the prompt is "Please convert the attached image into an anime R-style image", the style of the training content data when the generative pre-trained model is trained is reflected in the image, which is the content to be input. Therefore, when the content to be input is input into the generative pre-trained model, content synthesized from the content to be input and the training content data is output.

[0032] The output unit 28 transmits the content obtained by the control unit 24 to the user terminal 14.

[0033] In addition, each time the content generated by using the generative pre-trained model is used (for example, viewed), the fee paid to the copyright owner who provided the training data of the used generative pre-trained model may be increased. In this case, for example, the control unit 24 may associate data for recording the number of times of use with each generative pre-trained model of each copyright owner stored in the pre-trained model storage unit 26 every time content is generated using the generative pre-trained model.

[0034] The server 12 and the user terminal 14 of the information processing system 10 can be realized by, for example, the computer 70 shown in FIG. 5. The computer 70 includes a CPU 71, a memory 72 as a temporary storage area, and a non-volatile storage unit 73. The computer 70 also includes an input / output interface (I / F) 74 to which input / output devices etc. (not shown) are connected, and a read / write (R / W) unit 75 that controls reading and writing of data to and from a recording medium. The computer 70 also includes a network I / F 76 connected to a network such as the Internet. The CPU 71, the memory 72, the storage unit 73, the input / output I / F 74, the R / W unit 75, and the network I / F 76 are connected to each other via a bus 77.

[0035] The storage unit 73 can be implemented using a Hard Disk Drive (HDD), Solid State Drive (SSD), flash memory, etc. The storage unit 73, as a storage medium, stores a program for making the computer 70 function. The CPU 71 reads the program from the storage unit 73, loads it into memory 72, and sequentially executes the processes contained in the program.

[0036] <Operation of Information Processing System 10> Next, the operation of the information processing system 10 in this embodiment will be described. The server 12 of the information processing system 10 executes the learning processing routine shown in Figure 6.

[0037] In step S100, the control unit 24 reads out a plurality of learning content data stored in the learning data storage unit 22.

[0038] In step S102, the control unit 24 creates a generative trained model for each copyright holder based on the training content data for each copyright holder read in step S100.

[0039] In step S104, the control unit 24 stores the pre-trained generation model for each copyright holder, which was created in step S102, into the pre-trained model storage unit 26.

[0040] The trained model storage unit 26 stores the trained model for each copyright holder, and when input data is output from the user terminal 14, the information processing system 10 executes the information processing routine shown in Figure 7.

[0041] In step S200, the reception unit 20 receives the input data output from the user terminal 14.

[0042] In step S202, the control unit 24 selects a target generative trained model from a plurality of generative trained models stored in the trained model storage unit 26 based on the identification data included in the input data received in step S200.

[0043] In step S204, the control unit 24 inputs the prompt portion of the input data received in step S200 to the generative trained model selected in step S202, thereby causing the generative trained model to output content.

[0044] In step S206, the output unit 28 outputs the content obtained in step S204 as a result.

[0045] As described above, the information processing system according to the embodiment inputs input data to at least one of a plurality of generative trained models, each pre-trained for each copyright holder, which are pre-trained based on training data provided by copyright holders. The system then acquires content corresponding to the input data and output content output from the generative trained model. This allows for the management of content used for training the generative AI while simultaneously generating content using the generative AI.

[0046] This disclosure is not limited to the embodiments described above, and various modifications and applications are possible without departing from the gist of this disclosure.

[0047] For example, the various processes that the CPU reads and executes in the above embodiment may be executed by various processors other than the CPU. Examples of such processors include PLDs (Programmable Logic Devices) whose circuit configuration can be changed after manufacturing, such as FPGAs (Field-Programmable Gate Arrays), and dedicated electrical circuits that have a circuit configuration specifically designed to execute a particular process, such as ASICs (Application Specific Integrated Circuits). Furthermore, the various processes may be executed by one of these various processors, or by a combination of two or more processors of the same or different types (for example, multiple FPGAs, and a combination of a CPU and an FPGA). More specifically, the hardware structure of these various processors is an electrical circuit that combines circuit elements such as semiconductor elements.

[0048] Furthermore, although the above embodiment describes a configuration in which the program is pre-stored (installed) on storage, the invention is not limited to this. The program may be provided in a form stored on a non-transitor storage medium such as a CD-ROM (Compact Disk Read Only Memory), DVD-ROM (Digital Versatile Disk Read Only Memory), or USB (Universal Serial Bus) memory. Alternatively, the program may be provided in a form downloaded from an external device via a network.

[0049] Although the processing performed by each device in the above embodiment has been described as software processing performed by executing a program, it may also be hardware processing. Alternatively, it may be a combination of both software and hardware processing. Furthermore, the program stored in the ROM may be stored and distributed on various storage media.

[0050] Furthermore, although the above embodiment describes a case where a server 12, which is an example of an information processing device, generates content, it is not limited to this. Some or all of the processing performed by the server 12 in the above embodiment may be performed by a terminal such as a smartphone carried by the user.

[0051] All documents, patent applications, and technical standards described herein are incorporated by reference to the same extent as if each individual document, patent application, and technical standard were specifically and individually noted to be incorporated by reference.

[0052] (Note) The following is a note regarding the nature of this disclosure.

[0053] (Note 1) An information processing system comprising a processor, wherein the processor inputs input data to at least one of a plurality of generative trained models, each pre-trained based on training data provided by a copyright holder and pre-trained for each copyright holder, thereby acquiring content corresponding to the input data and outputting the content from the generative trained model, and outputting the content. (Note 2) The information processing system according to Note 1, wherein the input data is a prompt, and the generative trained model outputs the content corresponding to the input prompt. (Note 3) The information processing system according to Note 1 or Note 2, wherein the input data includes identification data for selecting the generative trained model, and the processor selects a target generative trained model from a plurality of generative trained models based on the identification data included in the input data, and acquires content outputting from the generative trained model by inputting the input data to the target generative trained model. (Note 4) The information processing system according to Note 3, wherein the input data is data including the content to be input and a prompt, and the processor inputs the content to be input and the prompt to the generative trained model to generate the content by synthesizing the content to be input and the training data used when the generative trained model was trained. (Note 5) The information processing system according to Note 2, wherein each time the content generated by using the generative trained model is used, the fee paid to the copyright holder who provided the training data for the generative trained model used is increased.(Note 6) An information processing method in which a computer performs a process of inputting input data into at least one of a plurality of generative trained models, each pre-trained based on training data provided by the copyright holder, and each pre-trained for each copyright holder, thereby acquiring content corresponding to the input data and outputting the content from the generative trained model. (Note 7) An information processing program for causing a computer to perform a process of inputting input data into at least one of a plurality of generative trained models, each pre-trained based on training data provided by the copyright holder, and each pre-trained for each copyright holder, thereby acquiring content corresponding to the input data and outputting the content from the generative trained model.

Claims

1. An information processing system comprising a processor, wherein the processor inputs input data to at least one of a plurality of generative trained models, each pre-trained based on training data provided by a copyright holder and pre-trained for each copyright holder, thereby acquiring content corresponding to the input data and outputting the content from the generative trained model, and outputting the content.

2. The information processing system according to claim 1, wherein the input data is a prompt, and the generative trained model outputs the content corresponding to the input prompt.

3. The information processing system according to claim 1 or 2, wherein the input data includes identification data for selecting the generative trained model, the processor selects a target generative trained model from a plurality of generative trained models based on the identification data included in the input data, and obtains content output from the generative trained model by inputting the input data to the target generative trained model.

4. The information processing system according to claim 3, wherein the input data is data including the content to be input and a prompt, and the processor generates the content by synthesizing the content to be input and the training data used when training the generative system trained model by inputting the content to be input and the prompt to the generative system trained model.

5. The information processing system according to claim 2, wherein each time the content generated by using the generative trained model is used, the fee paid to the copyright holder who provided the training data for the generative trained model used is increased.

6. An information processing method in which a computer performs a process of inputting input data into at least one of a plurality of generative trained models, each pre-trained based on training data provided by the copyright holder, and each of the plurality of generative trained models pre-trained for each copyright holder, thereby obtaining content corresponding to the input data and outputting the content from the generative trained model.

7. An information processing program for causing a computer to perform a process that involves inputting input data into at least one of several generative trained models, each pre-trained based on training data provided by the copyright holder, and each pre-trained for each copyright holder, thereby acquiring content corresponding to the input data and outputting the content from the generative trained model.

Images