Information processing device, authentication server, system, information processing method, and program

Abstract

This technology provides an improved reliability of the truthfulness of results generated by AI-generated data processing. [Solution] The information processing device is an information processing device that holds generation results generated by generation AI (Artificial Intelligence) processing, and comprises an acquisition means for acquiring authentication information relating to the authenticated truthfulness, which is the result of authentication of the truthfulness of the generation results, from an external source, and a holding control means for linking and holding the authenticated truthfulness based on the authentication information to the generation results.

Description

Technical Field

[0001] The present invention relates to generative AI.

Background Art

[0002] In recent years, with the spread of generative AI (Artificial Intelligence), an environment has been gradually established where even individuals can easily generate large amounts of various types of data (text, images, videos, voices, 3D models, etc.). It is considered that the creation of data by such generative AI will continue to be utilized in the future.

[0003] On the other hand, the generative results generated by generative AI may include information not based on facts or information with significantly modified facts. Therefore, information indicating the reliability of the generative results is required.

[0004] Regarding this technology, in Patent Document 1, in order to determine whether a video has been edited, a technique for identifying the origin of a video is disclosed by comparing the visual and audio features extracted by inputting a target video into a neural network with the features of known videos.

[0005] Also, in Patent Document 2, a technique for determining the authenticity of characters generated by the model during learning in order to perform character generation using an adversarial network model is disclosed.

Prior Art Documents

Patent Documents

[0006]

Patent Document 1

Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0007] The technology described in Patent Document 1 records the source of the generation results by the generation AI in metadata, but it does not address the degree of truthfulness. Furthermore, the technology in Patent Document 2 did not provide sufficient reliability regarding the truthfulness of the generation results by the generation AI. The aforementioned technologies failed to provide a highly reliable degree of truthfulness for the generation results of the generation AI processing.

[0008] Therefore, in order to solve the above problems, the present invention provides a technology that can improve the reliability of the truthfulness of the results generated by generation AI processing. [Means for solving the problem]

[0009] To solve this problem, for example, the information processing apparatus of the present invention has the following configuration. That is, An information processing device that holds the generation results generated by generation AI (Artificial Intelligence) processing, An acquisition means for obtaining authentication information regarding the authenticated truthfulness, which is the result of authentication of the truthfulness of the generated result, from an external source. A retention control means for storing the authenticated truthfulness based on the aforementioned authentication information in association with the generation result, It is equipped with. [Effects of the Invention]

[0010] According to the present invention, the reliability of the truthfulness of the results generated by the generation AI can be improved. [Brief explanation of the drawing]

[0011] [Figure 1] Conceptual diagram of a generative AI system according to the first embodiment. [Figure 2] A block diagram showing the module configuration of the client according to the first embodiment. [Figure 3] A block diagram showing the module configuration of the authentication server according to the first embodiment. [Figure 4] A block diagram showing the hardware configuration of the information processing device according to the embodiment. [Figure 5]Sequence diagram of the generative AI system according to the first embodiment. [Figure 6] Flowchart showing the processing procedure of the authentication server according to the first embodiment. [Figure 7] Conceptual diagram of the generative AI system according to the second embodiment. [Figure 8] Block diagram showing the module configuration of the client according to the second embodiment. [Figure 9] Block diagram showing the module configuration of the authentication server according to the second embodiment. [Figure 10] Sequence diagram of the generative AI system according to the second embodiment. [Figure 11] Conceptual diagram of the generative AI system according to the third embodiment. [Figure 12] Block diagram showing the module configuration of the client according to the third embodiment. [Figure 13] Block diagram showing the module configuration of the authentication server according to the third embodiment. [Figure 14] Sequence diagram of the generative AI system according to the third embodiment. [Figure 15] Flowchart showing the processing procedure of the authentication server according to the third embodiment.

Mode for Carrying Out the Invention

[0012] Hereinafter, embodiments will be described in detail with reference to the accompanying drawings. Note that the following embodiments do not limit the invention according to the claims. Although a plurality of features are described in the embodiments, not all of these plurality of features are essential to the invention, and the plurality of features may be arbitrarily combined. Further, in the accompanying drawings, the same or similar configurations are denoted by the same reference numerals, and redundant descriptions are omitted.

[0013] (Authentication by the Client and the Authentication Server in the First Embodiment) First, an overview of the environment in which the information processing device according to the first embodiment is used will be described. In this embodiment, an evaluation value called the truthfulness score is calculated for the generation results by the generation AI (Artificial Intelligence) processing, and the authenticated truthfulness score, which is the authentication result of the truthfulness score, is stored along with the generation results. For example, the generation AI processing generates news content and AI newscasters in news programs in the mass media industry, or AI talents and AI actors in the entertainment industry as generation results, but the generation results are not particularly limited. Here, "the degree to which the generation results by the generation AI represent the extent to which they are based on fact" will be defined as "truthfulness score" from now on. The embodiment maintains the truthfulness score in a state where it can be verified. As a result, for example, by notifying the user which parts of a news video featuring an AI newscaster are based on fact, this embodiment can realize a situation in which users can confidently watch news videos featuring AI newscasters. An example of applying the information processing device according to the first embodiment in such a situation will be described.

[0014] (Usage form) Figure 1 is a conceptual diagram of a generative AI system according to the first embodiment. As shown in Figure 1, the generative AI system according to the first embodiment includes a client 101 and an authentication server 106.

[0015] In the generation AI system according to the first embodiment, client 101 generates an image as a generation result through generation AI processing and sends information about the generation AI processing, including the generation result and the truthfulness of the generation result, to the authentication server 106. Client 101 obtains the truthfulness authenticated by the authentication server 106 (hereinafter also referred to as the authenticated truthfulness) and stores the generation result 105, which is the generated image, the authenticated truthfulness 107, which is the result of truthfulness authentication, and identification information 109 indicating the authentication server 106 that authenticated the truthfulness, linked together. The authenticated truthfulness is an example of authentication information. The processing in this embodiment will be described in detail below.

[0016] Client 101 inputs the base image 102, which is the image on which the generation is based, and the generation parameters 103 into the generation model 104 to obtain the generation result 105. Next, Client 101 requests the authentication server 106 to calculate the truth value, which is an evaluation value of the generation result 105, along with information about the generation AI processing, such as the base image 102 and the generation result 105, and to authenticate the calculation result.

[0017] The authentication server 106 receives the information and requests sent from the client 101 and authenticates the truthfulness of the generated result 105 based on the information regarding the generated AI processing. If the truthfulness meets the predetermined authentication criteria, the authentication server 106 sends it to the client 101 as authenticated truthfulness 107. The client 101 associates the generated result 105, the received authenticated truthfulness 107, and the identification information 109 indicating the authentication server 106, and stores them in the database 108.

[0018] Figure 2 is a block diagram showing the module configuration of client 101 according to the first embodiment. Client 101 includes a generation instruction unit 201, a generation unit 202, an evaluation value acquisition unit 203, an evaluation unit 204, an authentication request unit 205, an authentication result receiving unit 206, a holding control unit 207, and a holding unit 208.

[0019] The generation instruction unit 201 issues a generation instruction to the generation unit 202 based on information including the base image 102, generation parameters 103, and generation model 104.

[0020] The generation unit 202 receives a generation instruction from the generation instruction unit 201, inputs the base image 102 and generation parameters 103 to be processed into the generation model 104, executes generation AI processing, and generates and outputs the generation result 105.

[0021] The evaluation value acquisition unit 203 instructs the evaluation unit 204 to calculate the truthfulness score. The evaluation value acquisition unit 203 obtains the truthfulness score for the generated result 105 from the evaluation unit 204.

[0022] The evaluation unit 204 calculates the truthfulness score based on information related to the generation AI processing. For example, the evaluation unit 204 compares the base image 102 and the generated result 105 on a pixel-by-pixel basis to calculate the difference, and calculates the truthfulness score based on the ratio of the number of pixels with a difference (or discrepancy) to the total number of pixels.

[0023] The authentication request unit 205 sends an authentication request to the authentication server 106 to have the truthfulness calculated by the evaluation unit 204 authenticated. At this time, the authentication request unit 205 also sends predetermined information necessary for authentication to the authentication server 106 in addition to the truthfulness. The predetermined information is, for example, at least one of the following: information related to the generation AI processing and an ID indicating the client.

[0024] The authentication result receiving unit 206 is an example of an acquisition method and receives the authenticated truth value 107 transmitted from the authentication server 106. The authenticated truth value 107 is the result of the authentication of the truth value of the generated result.

[0025] The retention control unit 207 controls the retention unit 208 to associate and retain the generated result 105, the authenticated truthfulness score 107, and the identification information 109 indicating the authentication server 106.

[0026] The storage unit 208, under control from the storage control unit 207, associates the generated result 105, the authenticated truthfulness score 107, and the identification information 109 indicating the authentication server 106, and stores them in the database 108.

[0027] It should be noted that the means for linking and storing the generated result 105, the authenticated truth score 107, and the identification information 109 indicating the authentication server 106 are not limited to this embodiment. For example, the authenticated truth score 107 and the identification information 109 indicating the authentication server 106 may be embedded in the metadata of the generated result 105. In such a case, since the other two pieces of information can be confirmed in a single file of the generated result 105, the reliability of the evaluation value can be demonstrated even when sharing the generated result 105 with users who do not have access to the database 108.

[0028] Figure 3 is a block diagram showing the module configuration of the authentication server 106 according to the first embodiment. The authentication server 106 includes an authentication request receiving unit 301, an authentication unit 302, and an authentication result transmission unit 303.

[0029] The authentication request receiving unit 301 receives the authentication request and predetermined information transmitted from the authentication request unit 205.

[0030] The authentication unit 302 verifies the validity of the truthfulness received based on predetermined information received by the authentication request receiving unit 301, and creates an authentication result based on the verification result. The detailed processing procedure of the authentication unit 302 will be described later.

[0031] The authentication result transmission unit 303 transmits the authentication result created by the authentication unit 302 to the client 101.

[0032] (Hardware configuration) Figure 4 is a block diagram showing the hardware configuration of an information processing device according to an embodiment. In this embodiment, the client 101 and the authentication server 106 may both be information processing devices having these hardware components. The information processing device is, for example, a computer. The information processing device includes a CPU 401, a bus 402, a ROM 403, a RAM 404, an external memory 405, an input unit 406, a display unit 407, and a communication interface 408.

[0033] The CPU 401 controls various devices connected to the bus 402 and performs information processing. CPU 401 is an abbreviation for Central Processing Unit and is a type of processor.

[0034] The information processing device may have other processors such as an MPU (Micro Processing Unit), GPU (Graphics Processing Unit), NPU (Neural Processing Unit), and QPU (Quantum Processing Unit) in place of or in addition to the CPU 401. One or more processors, including the CPU 401, read computer programs (also called programs) stored in the ROM 403 or external memory 405, load them into the RAM 404, and execute them, thereby realizing some or all of the modules of the information processing device of the client 101 and authentication server 106 shown in Figures 2 and 3. The information processing device may also have multiple processors of the same type, with each processor realizing a different function.

[0035] Furthermore, some or all of the modules of the client and authentication server information processing devices may be implemented using one or more circuits, such as ASICs (Application Specific Integrated Circuits) and PLDs (Programmable Logic Devices) including FPGAs (Field Programmable Gate Arrays).

[0036] ROM403 stores the BIOS (Basic Input Output System) program and the boot program. ROM is an abbreviation for Read Only Memory. ROM403 may be non-volatile memory.

[0037] RAM404 is used as the main memory of CPU401. RAM404 is an abbreviation for Random Access Memory and is a high-speed read and write memory. RAM404 functions as a working area when CPU401 executes programs.

[0038] External memory 405 stores programs to be processed by the information processing device. External memory 405 may be a non-volatile storage device such as an HDD (Hard Disk Drive) or an SSD (Solid State Drive).

[0039] The input unit 406 receives input such as instructions and information from the user and processes it to output to the CPU 401. The input unit 406 may be a keyboard, mouse, touchpad, or touch panel.

[0040] The display unit 407 outputs the calculation results of the information processing device to the display device according to instructions from the CPU 401. The display device may be a liquid crystal display, a projector, an LED indicator, etc., and the type is not limited. LED is an abbreviation for Light Emitting Diode.

[0041] Bus 402 connects the CPU 401, RAM 404, ROM 403, external memory 405, input unit 406, display unit 407, and communication interface 408 to each other so that they can communicate with one another.

[0042] The communication interface 408 is an interface for communicating with other information processing devices and is connected to a network. In this embodiment, the client 101 and the authentication server 106 are connected via the communication interface 408 to communicate with each other through the network.

[0043] (Processing sequence) Figure 5 is a sequence diagram of the generation AI system according to the first embodiment. Specifically, Figure 5 shows the processing sequence between the client 101 and the authentication server 106 according to the first embodiment.

[0044] In step S701, the generation instruction unit 201 of the client 101 issues a generation instruction to the generation unit 202 based on information including the base image 102, generation parameters 103, and generation model 104.

[0045] In step S702, the generation unit 202 receives instructions from the generation instruction unit 201, generates the generation result 105, and outputs it.

[0046] In step S703, the evaluation unit 204 calculates the truthfulness score, which is an evaluation value for the calculated generation result. The evaluation value acquisition unit 203 acquires the truthfulness score for the generation result calculated by the evaluation unit 204.

[0047] In step S704, the authentication request unit 205 sends an authentication request to the authentication server 106 to have the acquired truthfulness score before authentication authenticated. At this time, the authentication request unit 205 may also send predetermined information necessary for authentication to the authentication server 106 in addition to the truthfulness score before authentication. The predetermined information includes, for example, the base image 102 and the generated result 105.

[0048] When the authentication request receiving unit 301 receives an authentication request, the authentication unit 302 verifies the validity of the truthfulness of the generated result and, according to the verification result, generates an authentication result that includes the authenticated truthfulness. If the authentication unit 302 rejects authentication, it may include a statement of rejection in the authentication result.

[0049] In step S705, the authentication result transmission unit 303 sends a response to the client 101 that includes the authentication result created by the authentication unit 302.

[0050] On client 101, the authentication result receiving unit 206 receives the authentication result, including the authenticated truthfulness score, as a response. The authentication result receiving unit 206 outputs the acquired response to the holding control unit 207.

[0051] In step S706, the retention control unit 207 controls the retention unit 208 to associate and retain the generated result 105, the received authenticated truthfulness score 107, and the identification information 109 indicating the authentication server 106. As a result, the retention unit 208 retains the generated result 105, the received authenticated truthfulness score 107, and the identification information 109 indicating the authentication server 106 in association.

[0052] (Detailed explanation of authentication methods) Figure 6 is a flowchart showing the processing procedure of the authentication server 106 according to the first embodiment.

[0053] In step S501, the authentication request receiving unit 301 obtains information to be used for authentication from the predetermined information it has received. For example, the authentication request receiving unit 301 obtains the base image 102, the generated result 105, and the truthfulness score before authentication.

[0054] In step S502, the authentication unit 302 calculates a truth value to compare with the received truth value based on the information acquired in step S501. For example, the authentication unit 302 compares the base image 102 and the generated result 105 on a pixel-by-pixel basis and calculates the truth value based on the ratio of the number of pixels with differences to the total number of pixels.

[0055] In step S503, the authentication unit 302 determines whether to authenticate the received pre-authentication truth value based on the verification results obtained in step S502. For example, the authentication unit 302 compares the truth value calculated in step S502 with the received pre-authentication truth value, and if the absolute value of the difference between the two truth values ​​is less than or equal to a predetermined threshold (or smaller than the threshold), it permits authentication; otherwise, it rejects authentication.

[0056] In step S504, the authentication unit 302 creates an authentication result based on the determination result obtained in step S503. For example, if the authentication unit 302 grants authentication in step S503, it uses the truthfulness score received in step S501 as an authenticated truthfulness score of 107, and combines it with a message notifying that authentication has been granted to form the authentication result. If the authentication unit 302 rejects authentication in step S503, it combines the truthfulness score calculated in step S502 with a message notifying that authentication has been rejected to form the authentication result.

[0057] In step S505, the authentication result transmission unit 303 transmits the authentication result, including the authenticated truthfulness score generated by the authentication unit 302, to the client 101.

[0058] By performing the above steps, the authentication process by the authentication unit 302 is completed.

[0059] (effect) The first embodiment described above can improve the reliability of the truthfulness of the generated results by associating and storing the authenticated truthfulness obtained from the external authentication server 106 with the generated results. Furthermore, the first embodiment allows users to view the generated results with peace of mind by notifying them of the authenticated truthfulness.

[0060] Furthermore, the first embodiment stores a link between the generated result produced by the generation AI processing, its evaluation value, which is the truthfulness score, and the identification information 109 of the authentication server 106 that authenticated the truthfulness score. This allows the first embodiment to demonstrate the reliability of the evaluation value of the generated result. In addition, according to the configuration of the first embodiment, the user can calculate the truthfulness score on the client side and only request authentication from the authentication server when authentication is required. With this configuration, the truthfulness score of the generated result can be easily checked, and authentication processing can be performed only when necessary.

[0061] (Experimental Example 1-1: The authentication server evaluates the data.) In the first embodiment, the client 101 had the evaluation unit 204. However, the authentication server 106 may have the evaluation unit 204 instead of the client 101.

[0062] In this case, the authentication result transmission unit 303 of the authentication server 106 uses the evaluation value calculated by the evaluation unit 204 of the authentication server 106 as the authenticated truthfulness, stores the authenticated truthfulness in a storage unit or other storage device, and transmits it as an authentication result including the authenticated truthfulness. Next, the evaluation value acquisition unit 203 acquires the authenticated truthfulness, which is the evaluation value received by the authentication result receiving unit 206 of the client 101. In this case, the processing in steps S502 and S503 can be omitted. This is because the evaluation value was calculated by the authentication server 106 itself, so verification is unnecessary.

[0063] This allows the calculation and authentication of evaluation values ​​to be completed within the authentication server 106, thereby increasing the reliability of the authentication results. Furthermore, the authentication server 106 can calculate evaluation values ​​using its own methods. This is an effective implementation for authentication servers that want to keep their evaluation value calculation methods and authentication criteria closed.

[0064] (Variation 1-2: The generation server performs generation, evaluation, and authentication requests) In the first embodiment, the client 101 had the generation unit 202. However, instead of the client 101, a third information processing device acting as a generation server may have the generation unit 202. Furthermore, the generation server may have an evaluation unit 204, an authentication request unit 205, and an authentication result receiving unit 206. In this case, the generation AI system according to this embodiment consists of the client 101, the generation server, and the authentication server 106.

[0065] In this case, client 101 sends a generation instruction to the generation server. Upon receiving the generation instruction, the generation unit 202 generates and outputs the generation result 105. Next, the generation server calculates an evaluation value based on the evaluation unit 204 and sends an authentication request to the authentication server 106 based on the authentication request unit 205. The authentication server creates an authentication result in the same manner as in the first embodiment and sends it to the generation server. The generation server sends the generation result 105 and the authentication result received by the authentication result receiving unit 206 to client 101.

[0066] This allows generation and evaluation processes to be executed on the generation server, eliminating the need for client 101 to possess generation or evaluation means. This is an effective embodiment when client 101 does not have sufficient resources to compute the generation AI processing. It is also an effective embodiment when companies or other organizations want to keep the details of the generation model confidential when providing generation AI as a service.

[0067] (Triplication 1-3: The authentication server issues a certificate of evaluation value.) In the first embodiment, the authentication unit 302 created an authentication result that included an evaluation value and a message notifying the authentication result. However, the authentication unit 302 may also perform additional processing on the message notifying the evaluation value and authentication result to obtain the authentication result. For example, the authentication unit 302 may obtain a digital certificate of the evaluation value, which contains a message notifying the evaluation value and authentication result, as the authentication result.

[0068] In this case, the authentication server 106 may create and store in advance the private key and public key used to digitally sign the digital certificate of the evaluation value. The authentication server 106 may also make the public key publicly available to third parties via an appropriate medium such as the internet. Next, the authentication unit 302 creates data containing the evaluation value and a message notifying the authentication result, digitally signs it using the private key, and creates a digital certificate of the evaluation value based on the data notifying the authentication result and the digital signature, which is the authentication result.

[0069] This makes the certifier information for the evaluation value clear, and allows third parties to detect tampering with the certifier information and certification results, thus improving the reliability of the evaluation value for the AI ​​generation results. This method is effective when the generation results are made public to an unspecified number of people via the internet, etc.

[0070] Furthermore, the authentication server 106 may publish a digital certificate of its public key based on a publicly known PKI (Public Key Infrastructure) mechanism. In this case, if the authentication server 106 is a root certificate authority, the digital certificate of its public key may be signed with a private key created by the authentication server 106. On the other hand, if the authentication server 106 is not a root certificate authority, a higher-level certificate authority may sign the certificate.

[0071] (Variation 1-4: Behavior when tampering is detected) In the first embodiment, an embodiment was described in which the generation AI processing is performed normally by client 101. However, it is conceivable that a malicious user's client may send false information to the authentication server 106 and fraudulently obtain authentication. Therefore, the first embodiment may have a mechanism to prevent tampering.

[0072] In this case, the client 101 records the input and output information in a log in an immutable format at each timing of the generation and evaluation processes, and sends this log along with the authentication request. At this time, the authentication unit 302 performs verification using the log sent by the client 101 in step S502. The authentication unit 302 may also reject authentication if the log information is not included when the authentication request is received. In addition, the retention control unit 207 may disable the retention unit 208 if authentication is rejected.

[0073] This allows the generating AI system to prevent tampering with authentication results, thereby improving the reliability of the evaluation values ​​for the AI's generated results.

[0074] (Variations 1-5: Variations in the calculation method of evaluation values) In the first embodiment, the evaluation unit 204 used pixel-level differences to calculate the truthfulness. However, the evaluation unit 204 may calculate the truthfulness using any calculation method, as long as the truthfulness is information indicating the degree to which the base image has been modified by the generating AI.

[0075] For example, the evaluation unit 204 may calculate the truthfulness score based on the similarity between the base image 102 and the generated result 105. Specifically, the evaluation unit 204 may calculate the truthfulness score based on the similarity of the edge images between the base image 102 and the generated result 105, or on the similarity that also takes color difference into consideration. The evaluation unit 204 may also calculate the truthfulness score based on at least one of the image quality evaluation indices, such as mean squared error (MSE), PSNR, and SSIM. Furthermore, if information is attached to at least one of the input data used in the generation AI processing, the generation model used in the generation AI processing, the prompts used in the generation AI processing, and the training data of the generation model used in the generation AI processing, the evaluation unit 204 may use this attached information to calculate the truthfulness score. If the history of the input data used in the generation AI processing is included in the additional information, the evaluation unit 204 may use this additional information to calculate the truthfulness score. For example, if the information indicating the history of the input data includes at least one of the following: the amount and number of modifications to the input data by past generation AI processing, the presence and intensity of noise added to the input data, and the presence or absence of a digital watermark, the evaluation unit 204 may calculate the truthfulness based on at least one of this information. In addition, if there is information linking at least one of the generation model and prompts with the evaluation value of the generation result, the evaluation unit 204 may calculate the truthfulness based on this information.

[0076] This allows the evaluation unit 204 to calculate truthfulness in various ways, enabling it to record evaluation results from diverse perspectives on the generated results. As a result, the generating AI system can further improve the reliability of the evaluation values ​​for the generated results of the generating AI.

[0077] (Second embodiment: Generation of results by an authenticated application) In the first embodiment, the authentication server authenticated the evaluation value of the generation result based on information regarding the AI ​​generation process performed by the client. In this embodiment, the authentication server authenticates the application for performing the AI ​​generation process and publishes it as an authenticated application. The client performs the AI ​​generation process using this authenticated application. The information processing device according to the second embodiment in this configuration will now be described. In this embodiment, the authenticated application can be said to be associated with authentication information indicating that it has been authenticated, and can also be said to be an example of authentication information.

[0078] (Usage form) Figure 7 is a conceptual diagram of the generative AI system according to the second embodiment.

[0079] Figure 7 shows a situation where the authentication server 106 publishes an authenticated generation AI application (hereinafter also referred to as the authenticated application), and the client 101 performs generation AI processing using the authenticated application. The generation AI system of the second embodiment will be described in detail below.

[0080] The authentication server 106 stores and publishes the generated AI application that meets the predetermined authentication criteria as an authenticated application in the application database 601. The client 101 retrieves any authenticated application from the application database 601 and starts it. The client 101 executes the process from generation to authentication on the application and stores the resulting artifact 602 in the database 108.

[0081] (composition) The generation AI system according to this embodiment consists of a client 101, an authentication server 106, and an authenticated application. Compared to the generation AI system according to the first embodiment, this configuration has the generation unit and evaluation unit located in the authenticated application instead of the client 101.

[0082] Figure 8 is a block diagram showing the module configuration of client 101 according to the second embodiment. Client 101 includes an authenticated application acquisition unit 801, an authenticated application verification request unit 802, a generation instruction unit 201, a generation unit 202, an evaluation value acquisition unit 203, an evaluation unit 204, an output product unit 803, a holding control unit 207, and a holding unit 208. Note that the authenticated application verification request unit 802, the generation instruction unit 201, the generation unit 202, the evaluation value acquisition unit 203, the evaluation unit 204, and the output product unit 803 may be the configuration of an authenticated application.

[0083] The authenticated application acquisition unit 801 retrieves a list of authenticated applications from the application database 601. From the list, the authenticated application acquisition unit 801 downloads an arbitrary authenticated application to the client 101 based on instructions from the user or other source. The authenticated application acquisition unit 801 then launches the acquired authenticated application.

[0084] The authenticated application verification request unit 802 requests the authentication server 106 to verify whether the authenticated application launched by the authenticated application acquisition unit 801 has been illegally modified since the authentication by the authentication server 106. At this time, the authenticated application verification request unit 802 may also send predetermined information necessary for verification to the authentication server 106. The predetermined information may be, for example, at least one of the hash values ​​of the application and the library.

[0085] The output unit 803 outputs the generated result 105, the authenticated truthfulness score 107, and identification information 109 indicating the authentication server 106.

[0086] The other modules are the same as in the first embodiment. However, in this embodiment, since both generation and evaluation are performed on a verified and certified application, the truthfulness calculated by the evaluation unit 204 can be treated as the certified truthfulness 107 from the beginning.

[0087] Figure 9 is a block diagram showing the module configuration of the authentication server 106 according to the second embodiment. The authentication server 106 includes an AI application generation authentication unit 901, an authenticated application publication unit 902, an authenticated application verification request receiving unit 903, an authenticated application verification unit 904, and an authenticated application verification result transmission unit 905.

[0088] The generation AI application authentication unit 901 verifies whether the generation AI application meets predetermined authentication criteria and designates those that meet the criteria as authenticated applications. For example, the generation AI application authentication unit 901 authenticates a generation AI application as an authenticated application by verifying whether the method for calculating the evaluation value of the generation result meets predetermined conditions set by the authentication server 106, and whether the processing on the application is protected from tampering. The generation AI application authentication unit 901 stores the authenticated applications in the application database 601.

[0089] The authenticated application publishing unit 902 publishes the authenticated application in a form that can be used by designated clients. For example, the authenticated application publishing unit 902 may publish the authenticated application in a downloadable format on a website on the internet so that it can be obtained by an unspecified number of people, or it may publish the authenticated application as an installation disc so that it can be obtained only by purchasers.

[0090] The authenticated application verification request receiving unit 903 receives the verification request and predetermined information transmitted from the authenticated application verification request unit 802.

[0091] The authenticated application verification unit 904 verifies the validity of the application based on predetermined information received by the authenticated application verification request receiving unit 903, and creates a verification result based on the verification result. If additional information is required during the application validity verification, the authenticated application verification unit 904 may instruct the client 101 to send the additional information.

[0092] The authenticated application verification result transmission unit 905 transmits the verification result created by the authenticated application verification unit 904 to the client 101.

[0093] (Processing sequence) Figure 10 is a sequence diagram of the generation AI system according to the second embodiment. Specifically, Figure 10 is a diagram showing the processing sequence between the client 101, the authentication server 106, the application database and 601 according to the second embodiment.

[0094] In step S1000a, the authenticated application acquisition unit 801 of client 101 requests an authenticated application from the application database 601.

[0095] In step S1000b, the application database 601 sends a response to client 101 containing the authenticated application in response to a request from client 101.

[0096] In step S1001, the authenticated application acquisition unit 801 launches the downloaded authenticated application.

[0097] In step S1002, the authenticated application verification request unit 802 sends an application verification request to the authentication server 106.

[0098] In the authentication server 106, the authenticated application verification request receiving unit 903 receives a verification request and outputs the verification request to the authenticated application verification unit 904. Based on the verification request, the authenticated application verification unit 904 verifies the authenticated application and generates a verification result.

[0099] In step S1003, the authenticated application verification result transmission unit 905 sends the verification result to the client 101 as a response to the verification request.

[0100] In step S1004, the generation instruction unit 201 of the client 101 issues a generation instruction to the generation unit 202 based on information including the base image 102, generation parameters 103, and generation model 104.

[0101] In step S1005, the generation unit 202 receives a generation instruction from the generation instruction unit 201, generates the generation result 105, and outputs it.

[0102] In step S1006, the evaluation unit 204 calculates the truthfulness score, which is an evaluation value for the generated result. In this embodiment, since both generation and evaluation are performed on a verified and certified application, the truthfulness score calculated by the evaluation unit 204 can be treated as the certified truthfulness score 107 from the beginning. The evaluation value acquisition unit 203 acquires the truthfulness score calculated by the evaluation unit 204 as the certification result.

[0103] In step S1007, the output unit 803 outputs the generated result 105, the authenticated truthfulness score 107, and identification information 109 indicating the authentication server 106.

[0104] In step S1008, the retention control unit 207 controls the retention unit 208 to retain the generated result 105, the authenticated truthfulness 107, and the identification information 109 indicating the authentication server 106, in association with each other.

[0105] (effect) According to the second embodiment described above, the client 101 performs the procedures from generation AI processing to evaluation of the generation result and authentication of the evaluation value on an authenticated application authenticated by the authentication server 106. As a result, the second embodiment can prevent tampering during the procedure and can record the reliability of the authenticated truthfulness, which is the evaluation value for the generation result, as more reliable. Furthermore, since the second embodiment verifies the authenticated application, the reliability of the truthfulness of the generation result can be further improved.

[0106] According to the configuration of the second embodiment, users can perform the entire process from generation to authentication simply by preparing an authenticated application. This configuration allows the second embodiment to consolidate the tools for generation AI processing, making it easier to create the environment for generation.

[0107] (Variation 2-1: Additional verification and authentication are performed) The authenticated application that executes the generation AI processing according to the second embodiment is protected so that the process from generation to authentication on the application cannot be tampered with from the outside. Therefore, the second embodiment performs verification once when the authenticated application is launched, and does not perform application verification or evaluation value authentication thereafter. However, the second embodiment may additionally perform application verification and evaluation value authentication. For example, the second embodiment may verify the application before and after generation, and may authenticate the evaluation value as in the first embodiment. In this case, the second embodiment can more effectively prevent unauthorized modification of the application and tampering with the authentication results.

[0108] (Variation 2-2: Verification omitted) In the second embodiment described above, an example was given in which the authentication server 106 verifies an authenticated application. However, in this modified example, the truthfulness of the generated result produced by an unverified authenticated application may be retained as the authenticated truthfulness.

[0109] (Third embodiment: Generation by a certified model, evaluation value = suitability of AI training data) In this embodiment, the authentication server 106 authenticates the generation model used for generation AI processing and publishes it as an authenticated model. The client 101 performs generation AI processing using this authenticated model. In this embodiment, the authenticated model can be said to be associated with authentication information, which is information indicating that it has been authenticated, and can also be said to be an example of authentication information. A third embodiment of this configuration will be described.

[0110] (Usage form) Figure 11 is a conceptual diagram of a generative AI system according to the third embodiment.

[0111] Figure 11 shows a situation where the authentication server 106 publishes an authenticated generation model (hereinafter also referred to as the authenticated model), and the client 101 performs generation AI processing using that authenticated model. The generation AI system of the third embodiment will be described in detail below.

[0112] The authentication server 106 publishes a generation model that meets the predetermined authentication criteria as an authenticated model 1102 in the model database 1101. The client 101 retrieves an arbitrary authenticated model 1102 from the model database 1101 along with an authenticated truth score 1120. The authenticated truth score 1120 associated with an authenticated model 1102 is an example of a model truth score. In the generation AI processing, the client 101 performs generation using the authenticated model 1102, and stores the obtained generation result 105, the authenticated truth score 107, and the identification information 109 indicating the authentication server 106 in the database 108.

[0113] In the model database 1101, each authenticated model 1102 is associated with its authentication result, the authenticated truth score 1120. For example, an authenticated model 1102 associated with an authenticated truth score 1120 showing a high value (e.g., "90") means that it hardly alters the base image 102. Authenticated models 1102 associated with such high authenticated truth scores 1120 are models that handle processes that modify the base image 102 relatively little, such as noise reduction and super-resolution. On the other hand, an authenticated model 1102 associated with an authenticated truth score 1120 showing a low value (e.g., "15") means that it significantly alters the base image 102, and these are models that handle processes such as style transfer and image generation of non-existent people. The authenticated model 1102 may output the authenticated truth score 1120 it is associated with along with the output of the generated result.

[0114] (composition) The generation AI system according to this embodiment includes a client 101 and an authentication server 106.

[0115] Figure 12 is a block diagram showing the module configuration of client 101 according to the third embodiment. Client 101 includes a certified model acquisition unit 1201, a generation instruction unit 201, a generation unit 202, a certified model verification request unit 1202, an evaluation value acquisition unit 203, an evaluation unit 204, a holding control unit 207, and a holding unit 208.

[0116] The authenticated model acquisition unit 1201 retrieves a list of authenticated models from the model database 1101 and downloads an arbitrary authenticated model from that list to the client 101. The authenticated model acquisition unit 1201 may download the authenticated model along with its authenticated truth score 1120. The authenticated model acquisition unit 1201 then starts the downloaded authenticated model.

[0117] The authenticated model verification request unit 1202 requests the authentication server 106 to verify whether the authenticated model 1102 has been illegally altered since its authentication by the authentication server 106. At this time, the authenticated model verification request unit 1202 also sends predetermined information necessary for verification to the authentication server 106. This predetermined information includes, for example, the hash value of the authenticated model 1102 and output information obtained when predetermined verification data is input to the authenticated model 1102.

[0118] Furthermore, the evaluation unit 204 according to this embodiment calculates the authenticated truth score 107 of the generated result 105 based on the evaluation value, which is the authenticated truth score 1120 associated with the authenticated model 1102. Detailed processing procedures will be described later.

[0119] The other modules are the same as in the first embodiment.

[0120] Figure 13 is a block diagram showing the module configuration of the authentication server 106 according to the third embodiment. The authentication server 106 includes a generation model authentication unit 1301, an authenticated model publication unit 1302, an authenticated model verification request receiving unit 1303, an authenticated model verification unit 1304, and an authenticated model verification result transmission unit 1305.

[0121] The generation model authentication unit 1301 checks whether the generation model meets predetermined authentication criteria and authenticates the generation model that meets the criteria as an authenticated model. For example, the generation model authentication unit 1301 checks what the characteristics of the generation model are, how much the output data is modified when authentication data is input, what the structure of the generation model is, and so on, and authenticates the generation model.

[0122] The authenticated model publication unit 1302 publishes authenticated models in a form that can be used by a specified client. For example, the authenticated model publication unit 1302 may publish authenticated models in a downloadable format on a website on the internet so that they can be obtained by an unspecified number of people, or it may publish authenticated models as an installation disc so that they can be obtained only by purchasers.

[0123] The authenticated model verification request receiving unit 1303 receives the verification request and predetermined information transmitted from the authenticated model verification request unit 1202.

[0124] The authenticated model verification unit 1304 determines whether the authenticated model 1102 meets the predetermined authentication criteria set by the authentication server 106, based on predetermined information received by the authenticated model verification request receiving unit 1303, and creates a verification result based on that result. The detailed processing procedure of the authenticated model verification unit 1304 will be described later.

[0125] The authenticated model verification result transmission unit 1305 transmits the verification results created by the authenticated model verification unit 1304 to the client 101.

[0126] (Processing sequence) Figure 14 is a sequence diagram of the generation AI system according to the third embodiment. Specifically, Figure 14 is a diagram showing the processing sequence between the client 101, the authentication server 106, and the model database 1101 according to the third embodiment.

[0127] In step S1401a, the authenticated model acquisition unit 1201 of the client 101 requests information about the authenticated model 1102 from the authentication server 106.

[0128] In step S1401b, the authenticated model publishing unit 1302 of the authentication server 106 returns information about the authenticated model to be published and sends it to the client 101.

[0129] In step S1402a, the authenticated model acquisition unit 1201 sends the authenticated model information obtained from the authentication server 106 to the model database 1101 to request an authenticated model.

[0130] In step S1402b, the model database 1101 sends the authenticated model 1102 corresponding to the request to the client 101 as a response. The model database 1101 may send the authenticated model along with the authenticated truth value 1120 to the client 101. The authenticated model acquisition unit 1201 acquires the authenticated model 1102 corresponding to the request from the model database 1101 along with the authenticated truth value 1120 and activates it.

[0131] In step S1403, the generation instruction unit 201 issues a generation instruction to the generation unit 202 based on information including the base image 102, generation parameters 103, and the authenticated model 1102.

[0132] In step S1404, the authenticated model verification request unit 1202 requests the authentication server 106 to verify whether the acquired authenticated model 1102 has been illegally altered since the time of authentication by the authentication server 106.

[0133] On the authentication server 106, the authenticated model verification request receiving unit 1303 receives a verification request and outputs it to the authenticated model verification unit 1304. The authenticated model verification unit 1304 verifies the authenticated model in response to the verification request and generates a verification result.

[0134] In step S1405, the authenticated model verification result transmission unit 1305 sends the verification result created by the authenticated model verification unit 1304 to the client 101 as a response to the request.

[0135] In step S1406, the generation unit 202 receives instructions from the generation instruction unit 201 and generates and outputs the generation result 105 using the authenticated model 1102. Note that the generation unit 202 does not have to generate a generation result if the verification result is not authenticated.

[0136] In step S1407, the evaluation unit 204 calculates the truth score, which is an evaluation value for the generated result. Since the generated result evaluated by the evaluation unit 204 is the result of a verified and certified model, the truth score calculated by the evaluation unit 204 can be treated as the certified truth score of 107. In other words, the certified truth score of the certified model, 1120, corresponds to the certified truth score of the generated result, 107. The evaluation value acquisition unit 203 acquires the certified truth score of 107 calculated by the evaluation unit 204.

[0137] In step S1408, the retention control unit 207 controls the retention unit 208 to retain the generated result 105, the authenticated truthfulness 107, and the identification information 109 indicating the authentication server 106, in association with each other.

[0138] (Detailed explanation of evaluation methods) In this embodiment, the authenticated model 1102 used in the generation AI processing is associated with an authenticated truth score 1120. Therefore, the evaluation unit 204 in this embodiment calculates the authenticated truth score 107 of the generation result 105 based on the authenticated truth score 1120 associated with the authenticated model 1102. Specifically, since the truth score 1120 associated with the authenticated model 1102 is "90", the evaluation unit 204 calculates "90" as the authenticated truth score 107 of the generation result.

[0139] (Detailed description of certified model validation methods) Figure 15 is a flowchart showing the processing procedure of the authentication server 106 according to the third embodiment.

[0140] In step S1501, the authenticated model verification request receiving unit 1303 obtains information to be used for verification from the predetermined information received. The predetermined information is, for example, at least one of the hash value of the authenticated model 1102 and the output information obtained when predetermined verification data is input to the authenticated model 1102.

[0141] In step S1502, the authenticated model verification unit 1304 verifies, based on the information obtained in step S1501, whether the authenticated model 1102 of client 101 has been illegally altered. For example, the authenticated model verification unit 1304 compares the hash value of the authenticated model 1102 sent by client 101 with the hash values ​​of each model recorded in the model database 1101, and stores the result if there is a match. If additional information is needed during the verification of the authenticated model 1102, the authenticated model verification unit 1304 may instruct client 101 to send the additional information.

[0142] In step S1503, the authenticated model verification unit 1304 determines whether the authenticated model to be verified passes or fails. For example, based on the verification results obtained in step S1502, the authenticated model verification unit 1304 determines whether the authenticated model 1102 of client 101 meets the predetermined authentication criteria set by the authentication server 106. For example, if the authenticated model verification unit 1304 determines in step S1502 that the hash values ​​match, it can determine that the authenticated model 1102 of client 101 has not been modified, and therefore may determine that the authenticated model to be verified passes.

[0143] In step S1504, the authenticated model verification unit 1304 creates a verification result based on the judgment result obtained in step S1503. For example, if it is determined to pass in step S1503, the verification result is a message notifying that the authenticated model 1102 of client 101 meets the authentication criteria of the authentication server 106. If it is determined to fail in step S1503, the verification result is a message notifying that the authenticated model 1102 of client 101 does not meet the authentication criteria of the authentication server 106.

[0144] In step S1505, the authenticated model verification result transmission unit 1305 transmits the verification result to the client 101.

[0145] By performing the above steps, the authentication server 106 will terminate the verification process.

[0146] (effect) According to the third embodiment described above, client 101 performs generation AI processing using a generation AI model authenticated by authentication server 106. This makes it easier for client 101 to predict the evaluation value of the generation result in advance, depending on which authenticated model is selected. Therefore, when client 101 has preferences or constraints regarding the evaluation value of the generation result, the user can efficiently perform generation AI processing.

[0147] (Variation 3-1: Verification with input and output of a certified model) In the third embodiment, an example was shown in which a hash value is used to verify the authenticated model 1102 in the authenticated model verification unit 1304. However, the method by which the authenticated model verification unit 1304 verifies the authenticated model 1102 is not limited to the method of this embodiment, as long as it can confirm whether the authenticated model 1102 satisfies the predetermined authentication criteria defined by the authentication server 106.

[0148] For example, the authenticated model verification unit 1304 may retrieve a model with the same name as the authenticated model 1102 from the model database 1101 and input verification data and generation parameters into that model. The authenticated model verification unit 1304 may then compare the obtained output with the information sent by the client 101 and create a verification result based on the result.

[0149] This allows the authenticated model verification unit 1304 to verify the model based on the internal processing of the authenticated model 1102, making it easier to have a broader range of authentication criteria for the model compared to comparing by hash value matching. Specifically, the authenticated model verification unit 1304 calculates the similarity between the output when verification data is input and a predetermined output. The authenticated model verification unit 1304 may determine that the model has passed if the similarity is above a predetermined threshold. This method is effective when allowing a certain level of modification to the authenticated model, such as transfer learning or fine-tuning.

[0150] (Variation 3-2: Calculation of evaluation values ​​based on a certified model) In the third embodiment, in the model database 1101, each authenticated model 1102 was associated with a single authenticated truth score. The evaluation unit 204 calculated "90" as the authenticated truth score 107 for the generated result 105 because the authenticated truth score associated with the authenticated model 1102 was "90". However, the amount of modification to the generated result 105 may increase or decrease depending on the generation parameters and prompts given to the authenticated model 1102. Therefore, the truth scores associated with each authenticated model in the model database 1101 may have a range. Furthermore, when the evaluation unit 204 calculates the authenticated truth score of the generated result 105 based on the truth score associated with the authenticated model 1102, it may also take other information into account when calculating the authenticated truth score.

[0151] For example, suppose there is a certified model in which the base image 102 is not significantly altered under normal use, but is altered to a certain extent only when extremely strong noise reduction is applied. In this case, the model database 1101 may assign a truth score of 70 to 90 to the above certified model. At this time, the evaluation unit 204 may calculate a separate truth score based on the generation parameters 103 of the same certified model, in addition to the truth score information of 70 to 90 assigned to the above certified model.

[0152] This method allows the evaluation unit 204 to calculate the certified truthfulness of the generated results in a manner that reflects the actual usage of the certified model 1102. As a result, this modified example can calculate the evaluation value for the generated AI's results in a more realistic manner, thereby increasing the reliability of the evaluation value, which is the certified truthfulness of the generated results.

[0153] (Variation 3-3: Behavior when tampering is detected) In the third embodiment, an embodiment was described in which the generation AI processing is performed normally by client 101. However, it is conceivable that a malicious client could send false information to the authentication server 106 and fraudulently bypass the verification of the authenticated model 1102. Therefore, a mechanism to prevent tampering may be added to the third embodiment.

[0154] For example, when client 101 first calculates the hash value of authenticated model 1102, this hash value may be recorded in a log in an immutable format. From thereafter, client 101 may continue to calculate the hash value of authenticated model 1102 at regular intervals as background processing, and if a value different from the hash value in the log is obtained, the verification result given to authenticated model 1102 may be invalidated. In addition, if the verification result is invalidated, the retention control unit 207 may invalidate the retention unit 208 or the corresponding information of the retention unit 208.

[0155] By using the method described above, this modification can prevent tampering with the authentication results and further improve the reliability of the evaluation value, which is the authenticated truthfulness of the results generated by the AI ​​generation process.

[0156] (Variation 3-4: Verification omitted) In the third embodiment described above, an example was given in which the authentication server 106 verifies the authenticated model. However, in this modified example, the truthfulness of the generated result produced by an unverified authenticated model may be retained as the authenticated truthfulness.

[0157] (Other embodiments) The embodiments described above may be combined as appropriate. Furthermore, the system may be configured to allow the user to select any of the combined embodiments.

[0158] The present invention can also be realized by supplying a program that implements one or more of the functions of the above-described embodiments to a system or device via a network or storage medium, and by having one or more processors in the computer of that system or device read and execute the program. Furthermore, the present invention can also be realized by a circuit (e.g., an ASIC) that implements one or more functions.

[0159] The disclosures herein include the following information processing devices, authentication servers, systems, information processing methods, and programs. (Item 1) An information processing device that holds the generation results generated by generation AI (Artificial Intelligence) processing, An acquisition means for obtaining authentication information regarding the authenticated truthfulness, which is the result of authentication of the truthfulness of the generated result, from an external source. A retention control means for storing the authenticated truthfulness based on the aforementioned authentication information in association with the generation result, An information processing device characterized by comprising: (Item 2) The information processing device according to item 1, characterized in that, in the generation AI processing, it comprises a generation means having a generation model linked to authentication information and an authenticated model that outputs a generation result. (Item 3) The information processing device according to item 2, characterized by comprising an evaluation means for calculating the truthfulness of the generation result generated by the certified model as the certified truthfulness. (Item 4) The model truth score, which indicates the truthfulness of the aforementioned generative model, is associated with it. The information processing device according to item 2 or 3, characterized in that it has an evaluation means for calculating the certified truthfulness of the generation result generated by the certified model based on the model truthfulness. (Item 5) The information processing apparatus according to any one of items 1 to 4, characterized by comprising an evaluation means for calculating the truthfulness of the generated result based on the difference between the input data that is the target of the generation AI processing and the generated result. (Item 6) The information processing apparatus according to any one of items 1 to 5, characterized by comprising an evaluation means for calculating the truthfulness of the generated result based on the similarity between the input data to be processed by the generation AI process and the generated result. (Item 7) An information processing apparatus according to any one of items 1 to 6, characterized by comprising an evaluation means for calculating the truthfulness of the generated result based on at least one of a generative model used in the generative AI processing, a prompt used in the generative AI processing, and training data of the generative model used in the generative AI processing. (Item 8) The information processing apparatus according to any one of items 1 to 7, characterized by comprising an evaluation means for calculating the truthfulness of the generated result based on information regarding the history of the input data used in the generation AI processing. (Item 9) The information processing apparatus according to any one of items 1 to 8, characterized in that it comprises an evaluation means for calculating the truthfulness of the generated result based on at least one of the number of times and the amount of modification the input data used in the generation AI processing has been processed by the generation AI in the past. (Item 10) The information processing apparatus according to any one of items 1 to 9, characterized by comprising an evaluation means for calculating the truthfulness of the generated result based on at least one of noise and a digital watermark applied to the input data used in the generation AI processing. (Item 11) The information processing apparatus according to any one of items 1 to 10, characterized in that the acquisition means acquires at least one of the following as authentication information: the authenticated truthfulness, the authenticated application for executing the generation AI processing, and the authenticated model used in the generation AI processing. (Item 12) The information processing apparatus according to any one of items 1 to 11, characterized in that it comprises a generation means having an authenticated application for which the application that performs the generation AI processing has been authenticated. (Item 13) The information processing apparatus according to any one of items 1 to 12, characterized by comprising a holding means for holding the generated result and the authenticated truthfulness based on the control of the holding control means. (Item 14) Authentication means for generating authentication information regarding the authenticated truthfulness, which is the result of authentication of the truthfulness of the generated result produced by the generation AI processing, A transmission means for transmitting the aforementioned authentication information to an external party, An authentication server characterized by having the following features. (Item 15) The authentication server according to item 14, characterized in that the authentication means generates authentication information including an authenticated truth value, which is authenticated based on the generation result and the truth value obtained from an external source. (Item 16) The authentication server according to item 14 or item 15, characterized in that the authentication means generates authentication information including an authenticated application that has authenticated the application for executing the generation AI processing. (Item 17) The authentication server according to any one of items 14 to 16, characterized in that the authentication means generates authentication information including an authenticated model that authenticates the generation model used for the generation AI processing. (Item 18) The authentication server according to any one of items 14 to 17, characterized by having an evaluation means for calculating the truthfulness of the generation result based on the information relating to the generation result. (Item 19) An authentication server according to any one of items 14 to 18, characterized by having a means for holding information for generating the aforementioned authentication information. (Item 20) The information processing device described in item 1, The authentication server that generates the aforementioned authentication information, A system characterized by comprising the following features. (Item 21) An information processing method for storing the generation results generated by generative AI (Artificial Intelligence) processing, The acquisition step involves obtaining authentication information regarding the authenticated truthfulness, which is the result of authenticating the truthfulness of the generated result, from an external source. A retention control step that stores the authenticated truthfulness based on the authentication information linked to the generation result, An information processing method characterized by comprising: (Item 22) A program to cause a computer to function as one of the information processing devices described in any one of items 1 through 13.

[0160] The invention is not limited to the embodiments described above, and various modifications and variations are possible without departing from the spirit and scope of the invention. Accordingly, claims are attached to disclose the scope of the invention. [Explanation of Symbols]

[0161] 101...Client, 104...Generating model, 105...Generating result, 106...Authentication server, 107...Authentication truthfulness, 202...Generation unit, 204...Evaluation unit, 206...Authentication result receiving unit, 207...Holding control unit, 302...Authentication unit, 303...Authentication result transmission unit, 801...Authentication-certified application acquisition unit, 901...Generating AI application authentication unit, 1102...Authentication-certified model, 1201...Authentication-certified model acquisition unit, 1301...Generating model authentication unit.

Claims

1. An information processing device that holds the generation results generated by generation AI (Artificial Intelligence) processing, An acquisition means for obtaining authentication information regarding the authenticated truthfulness, which is the result of authentication of the truthfulness of the generated result, from an external source. A retention control means for storing the authenticated truthfulness based on the aforementioned authentication information in association with the generation result, An information processing device characterized by comprising:

2. The information processing apparatus according to claim 1, characterized in that the generation AI processing includes a generation means having a generation model linked to authentication information and an authenticated model that outputs a generation result.

3. The information processing apparatus according to claim 2, further comprising an evaluation means for calculating the truthfulness of the generation result generated by the certified model as the certified truthfulness.

4. The model truth score, which indicates the truthfulness of the aforementioned generative model, is associated with it. The information processing apparatus according to claim 2, further comprising an evaluation means for calculating the certified truthfulness of the generated result generated by the certified model based on the model truthfulness.

5. The information processing apparatus according to claim 1, further comprising an evaluation means for calculating the truthfulness of the generated result based on the difference between the input data that is the target of the generation AI processing and the generated result.

6. The information processing apparatus according to claim 1, further comprising an evaluation means for calculating the truthfulness of the generated result based on the similarity between the input data, which is the target of the generation AI processing, and the generated result.

7. The information processing apparatus according to claim 1, further comprising an evaluation means for calculating the truthfulness of the generated result based on at least one of the generative model used in the generative AI processing, the prompt used in the generative AI processing, and the training data of the generative model used in the generative AI processing.

8. The information processing apparatus according to claim 1, further comprising an evaluation means for calculating the truthfulness of the generated result based on information regarding the history of the input data used in the generation AI processing.

9. The information processing apparatus according to claim 1, further comprising an evaluation means for calculating the truthfulness of the generated result based on at least one of the number of times and the amount of modification the input data used in the generation AI processing has been processed by the generation AI in the past.

10. The information processing apparatus according to claim 1, further comprising an evaluation means for calculating the truthfulness of the generated result based on at least one of noise and a digital watermark applied to the input data used in the generation AI processing.

11. The information processing apparatus according to claim 1, wherein the acquisition means acquires at least one of the following as authentication information: the authenticated truthfulness, the authenticated application for executing the generation AI processing, and the authenticated model used in the generation AI processing.

12. The information processing apparatus according to claim 1, characterized in that it comprises a generation means having an authenticated application for which the application that performs the generation AI processing has been authenticated.

13. The information processing apparatus according to claim 1, further comprising a holding means for holding the generated result and the authenticated truthfulness based on the control of the holding control means.

14. Authentication means for generating authentication information regarding the authenticated truthfulness, which is the result of authentication of the truthfulness of the generated result produced by generation AI processing, A transmission means for transmitting the aforementioned authentication information to an external party, An authentication server characterized by having the following features.

15. The authentication server according to claim 14, characterized in that the authentication means generates authentication information including an authenticated truth value, which is authenticated based on the generation result and the truth value obtained from an external source.

16. The authentication server according to claim 14, characterized in that the authentication means generates authentication information including an authenticated application that has authenticated the application for executing the generation AI processing.

17. The authentication server according to claim 14, characterized in that the authentication means generates authentication information including an authenticated model which authenticates the generation model used for the generation AI processing.

18. The authentication server according to claim 14, characterized in that it has an evaluation means for calculating the truthfulness of the generation result based on the information relating to the generation result.

19. The authentication server according to claim 14, characterized in that it has a holding means for holding information for generating the aforementioned authentication information.

20. The information processing apparatus according to claim 1, The authentication server that generates the aforementioned authentication information, A system characterized by comprising the following features.

21. An information processing method for holding generation results generated by artificial intelligence (AI) processing, The acquisition step involves obtaining authentication information regarding the authenticated truthfulness, which is the result of authenticating the truthfulness of the generated result, from an external source. A retention control step that stores the authenticated truthfulness based on the authentication information linked to the generation result, An information processing method characterized by comprising:

22. A program for causing a computer to function as one of the means of an information processing device according to any one of claims 1 to 13.

Images