AI model provisioning and control system, processing method and program for the AI ​​model provisioning and control system.

Abstract

To enable users to appropriately determine whether or not they are suitable for using AI models. [Solution] The AI ​​model provision control system includes a use case estimation means that estimates use cases in which a user will use the AI ​​model based on information of a user who intends to use a trained AI model, and an AI model usability determination means that determines whether or not the user is suitable to use the AI ​​model based on the use cases estimated by the use case estimation means.

Description

Technical Field

[0001] The present invention relates to an AI model providing control system, a processing method of the AI model providing control system, and a program.

Background Art

[0002] In recent years, many techniques for highly processing images to extract useful information have been proposed. Among them, in particular, regarding face authentication that uses a multi-layer neural network called a deep net (or also referred to as a deep neural net or deep learning) to compare an input face image with a pre-registered face image and determine whether the input face image is of the same person as the registered face image, extensive research and development have been carried out.

[0003] Patent Document 1 discloses a server provided with a processor configured to identify AI ethics that a learned model must satisfy and determine whether the learned model satisfies the AI ethics based on learning conditions and the AI ethics.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] The progress of information extraction technology from images using AI (Artificial Intelligence) is remarkable. Especially regarding face authentication, from the perspective of safety and security, its introduction in public places is also under consideration.

[0006] On the other hand, concerns have been raised about the use of facial recognition technology, particularly regarding privacy violations, and there is a growing debate about the need to impose certain restrictions on its use. For example, the EU is considering legislation to prohibit the use of AI that conducts large-scale surveillance activities using facial recognition and biometric authentication technologies in public places.

[0007] Thus, biometric authentication technologies, such as facial recognition, offer numerous advantages to users, including the ability to verify IDs without having to remember passwords. On the other hand, if not used carefully, biometric authentication technologies can potentially infringe upon people's privacy and human rights. Therefore, when using biometric authentication technologies, such as facial recognition, sufficient consideration must be given to protecting privacy and human rights.

[0008] On the other hand, it is common practice to publish pre-trained AI models on the internet and share them among AI developers and users. When considering a service that provides such pre-trained AI models, it has been extremely difficult in practice to understand how the provided AI models are used by the recipients or to restrict how they are used.

[0009] Furthermore, technologies for estimating how AI models will be used have not been considered until now. Additionally, technologies for suspending the provision of AI models if it is determined that they are likely to be used inappropriately have also not been considered until now.

[0010] The objective of this invention is to enable users to appropriately determine whether or not they are suitable to use an AI model. [Means for solving the problem]

[0011] The AI ​​model provision control system includes a use case estimation means that estimates use cases in which a user will use the AI ​​model based on information of the user who intends to use the trained AI model, and an AI model usability determination means that determines whether or not the user is suitable to use the AI ​​model based on the use cases estimated by the use case estimation means. [Effects of the Invention]

[0012] According to the present invention, it is possible to appropriately determine whether or not a user is suitable to use an AI model. [Brief explanation of the drawing]

[0013] [Figure 1] This is a diagram showing the configuration of the AI ​​model provisioning control system. [Figure 2] This is a schematic diagram of a user information management method. [Figure 3] This is a schematic diagram of the AI ​​model management state. [Figure 4] This figure shows the operation sequence of the AI ​​model provisioning and control system. [Figure 5] This is a schematic diagram of a database for determining the level of ethical risk using AI. [Figure 6] This is a diagram showing the configuration of the AI ​​model provisioning control system. [Figure 7] This is a schematic diagram of a user information management method. [Figure 8] This is a block diagram showing the configuration of the use case estimation unit. [Figure 9] This is a diagram showing an example of a captured image. [Figure 10] This is a diagram showing the configuration of the AI ​​model provisioning control system. [Figure 11] This figure shows the operation sequence of the AI ​​model provisioning and control system. [Figure 12] This is a schematic diagram of a user information management method. [Modes for carrying out the invention]

[0014] (First Embodiment) In the first embodiment, an example of determining whether to provide an AI model capable of performing face authentication to a user in a service (referred to as an AI model provision control system) that provides various learned AI models being managed to the user will be described.

[0015] Face authentication refers to a personal authentication method that extracts information (biometric identifier) representing an individual from a face image and uses it to identify the person ID of that person.

[0016] Regarding the unrestricted use of face authentication, it has been pointed out that there are problems from the perspective of privacy infringement, etc. Depending on the usage method of face authentication, there is a risk of privacy infringement and human rights infringement. In the AI model provision control system according to this embodiment, in order to reduce that risk, the information of the AI model's destination (user of the AI model) is referred to and its use case is estimated. Then, when it is determined that the usage is likely to cause a risk, control is performed to reject the provision of the AI model.

[0017] In this embodiment, an example of rejecting the provision of a face authentication AI model when it is presumed that the face authentication AI model will be used for criminal investigation when the face authentication AI model is provided to a user who is attempting to use the face authentication AI model is shown.

[0018] FIG. 1 is a block diagram showing a configuration example of an AI model provision control system 100 according to the first embodiment. In the AI model provision control system 100 in this embodiment, it is assumed that the management and provision of AI models that can be used in a digital camera are being performed.

[0019] The AI model provision control system 100 includes a user information management unit 102, a use case estimation unit 103, an AI model management unit 104, and an AI model availability determination unit 105.

[0020] Reference numeral 101 denotes a user of the AI ​​model provision control system 100. User 101 refers to a user who wants to try using the AI ​​models managed by the AI ​​model provision control system 100. In this embodiment, as described above, the AI ​​model provision control system 100 manages and provides AI models for digital cameras, so user 101 is assumed to be a digital camera user.

[0021] User 101 registers as a user by disclosing their personal information in order to use the AI ​​model provision control system 100, and then begins using the AI ​​model provision control system 100. Examples of personal information include the user's name and information about the organization they belong to. This personal information of User 101 will be referred to as user information.

[0022] Code 102 represents the User Information Management Department. The User Information Management Department 102 manages the association between users 101 and their user information. Generally, it is assumed that there are multiple users who use the AI ​​Model Provision Control System 100, so user information exists for each user. Therefore, the User Information Management Department 102 manages the association between each user and their personal information (name and affiliated organization).

[0023] Figure 2 is a schematic diagram of the user information management method in the User Information Management Department 102. Figure 2 shows how the organizations to which users A, B, and C belong are managed. For example, it shows that user A belongs to a "criminal investigation organization".

[0024] Reference numeral 103 denotes a use case estimation unit. Based on the user information of the user managed by the user information management unit 102, the use case estimation unit 103 estimates how the user is likely to use the AI ​​model if it is provided to the user.

[0025] For example, since user A belongs to a "criminal investigation organization," it is presumed that if they receive an AI model, they will use the digital camera that the AI ​​model can operate on for criminal investigations. In this case, the presumed use case for user A would be criminal investigation.

[0026] Similarly, since user B belongs to a "military-related organization," it is expected that the AI ​​model will be used as part of military activities. Furthermore, for user C, who belongs to a "cram school management organization," use cases such as confirming attendance and measuring student concentration levels can be estimated by using a digital camera to film the lessons.

[0027] This method of estimating "use cases" from "affiliated organizations" can be achieved by having a human pre-evaluate "use cases" from "affiliated organizations," creating a database of those relationships, and setting it in the use case estimation unit 103.

[0028] Alternatively, this can be achieved by using large-scale language models called LLMs (Large Language Models), which have recently become increasingly useful for various tasks, to ask AI models about use cases for a given organization. When using LLMs, one can use LLMs published by major IT vendors as they are, or the service provider operating the AI ​​model provision control system 100 can perform additional training on a published LLM and use their own unique LLM as a result. By using the relationship between "organization" and "use case" estimated by the user as training data to perform additional training on the LLM, it is expected that a unique LLM more suitable for use case estimation can be obtained.

[0029] Code 104 represents the AI ​​model management unit. The AI ​​model management unit 104 manages trained AI models in association with the tasks that those AI models can perform.

[0030] Figure 3 is a schematic diagram of the AI ​​model management state in the AI ​​model management unit 104. Figure 3 shows the case where three trained AI models are being managed. As mentioned above, the AI ​​model management unit 104 manages trained AI models in association with the tasks that those AI models can perform. For example, the first AI model is a "face recognition AI model," and the task that can be performed using this model is "estimating a person ID from a face image in an image."

[0031] Code 105 is the AI ​​model availability determination unit. The AI ​​model availability determination unit 105 determines whether or not to provide the AI ​​model to the user 101 based on the use case estimated by the use case estimation unit 103 and the tasks that the AI ​​model managed by the AI ​​model management unit 104 will implement. The criterion for determining whether or not to provide the AI ​​model is whether or not the AI ​​model being offered poses a high or low risk of causing AI ethical problems if used in the estimated use case. In other words, if the risk of causing AI ethical problems is high, the unit determines that it will refuse to provide the AI ​​model; otherwise, it determines that it will provide it.

[0032] For example, if "User A" is attempting to use the "Facial Recognition AI Model," the AI ​​Model Availability Determination Unit 105 determines whether or not to provide the AI ​​model (in this case, the Facial Recognition AI Model) as follows: In this case, it is presumed that User A's use case is criminal investigation, so there is a high probability that the Facial Recognition AI Model will be used to "identify a person's ID from a facial image in an image during a criminal investigation." Using "Facial Recognition AI in criminal investigations" is not socially acceptable, as there have been cases of wrongful arrests in the past, and it carries a high risk of causing AI ethical problems. Therefore, if "User A" is attempting to use the "Facial Recognition AI Model," the provision of the Facial Recognition AI Model is refused.

[0033] Determining the level of AI ethical risk based on such "combinations of use cases and tasks realized by AI models" can be achieved by having a human examine in advance whether the AI ​​ethical risk is high or low for each combination of use cases and AI models, creating a database of the results of that examination, and setting it in the AI ​​model usability determination unit 105. When examining the level of AI ethical risk, one can refer to information on legal regulations concerning AI ethics, such as the EU's AI Act, and past cases of AI ethical problems.

[0034] For example, the AI ​​Act prohibits or classifies as high-risk "real-time use of remote biometric authentication technologies such as facial recognition AI in public places" and "biometric authentication AI for criminal behavior prediction or law enforcement purposes." A database for determining the level of risk can be constructed using this kind of information.

[0035] Figure 5 is a schematic diagram of the database used by the AI ​​model availability determination unit 105 to determine the level of AI ethical risk. In Figure 5, three use cases are assumed for a facial recognition AI model, and the results of examining the AI ​​ethical risk for each are stored in the database. For example, the first combination is using the "facial recognition AI model" in "criminal investigation," and the AI ​​ethical risk in this case is set to be high. Therefore, in this combination, the provision of the facial recognition AI model will be refused.

[0036] The above describes the configuration of the AI ​​model provisioning and control system 100.

[0037] Next, we will explain the operation sequence of the AI ​​model provision control system 100 using Figure 4. We will then explain the processing method of the AI ​​model provision control system 100.

[0038] First, in step S400, a request for AI model provision is sent from a user to the AI ​​model provision control system 100. For example, user A sends a request to the AI ​​model provision control system 100 for the provision of a facial recognition AI model. For simplicity of explanation, this user who made the model provision request will be called the model requesting user. Furthermore, the AI ​​model that the model requesting user is requesting will be called the requested AI model.

[0039] In the following step S401, the user information management unit 102 obtains the user information of the model requesting user. This user information is that of the user who intends to use the trained AI model.

[0040] Furthermore, in step S402, the use case estimation unit 103 estimates the use cases in which the model requesting user will use the requested AI model, based on the user information acquired in S401. For example, the use case estimation unit 103 estimates the use cases based on the organizational information to which the model requesting user in Figure 2 belongs. For example, as shown in Figure 2, in the case of user A, criminal investigation is estimated as the use case.

[0041] In the following step S403, the AI ​​model management unit 104 identifies tasks that the requested AI model can perform, as shown in Figure 3.

[0042] Furthermore, in step S404, the AI ​​model availability determination unit 105 determines the level of AI ethical risk based on the use case estimated in S402 and the tasks that the requested AI model identified in S403 can perform. For example, as shown in Figure 5, if user A is requesting the provision of a facial recognition AI model, the unit determines that the AI ​​ethical risk is high because there is a possibility that the facial recognition AI model will be used in criminal investigations.

[0043] In the following step S405, the AI ​​model availability determination unit 105 determines whether the model requester is suitable to use the requested AI model based on the level of AI ethical risk determined in S404. If the AI ​​ethical risk is high, the AI ​​model availability determination unit 105 determines that the model requester is not suitable to use the requested AI model and proceeds to step S406. Conversely, if the AI ​​ethical risk is not high, the AI ​​model availability determination unit 105 determines that the model requester is suitable to use the requested AI model and proceeds to step S407.

[0044] In step S406, the AI ​​model availability determination unit 105 refuses to provide the requested AI model to the model requesting user.

[0045] In step S407, the AI ​​model availability determination unit 105 provides the requested AI model to the model requesting user.

[0046] This concludes the explanation of the AI ​​model provision control system 100 according to this embodiment.

[0047] As explained in detail above, according to this embodiment, the use case (how the AI ​​model will be used) of the AI ​​model provided to the user is estimated, and if it is determined that the way it is being used poses a high risk from an AI ethical standpoint, the provision of the AI ​​model is refused. This makes it possible to deter the use of AI models that pose a high risk from an AI ethical standpoint, such as infringing on people's privacy or human rights.

[0048] Generally, users requesting AI models (users who request AI models from the AI ​​model provision control system 100) do not necessarily possess sufficient knowledge of AI ethics or legal regulations concerning AI. For such users, assessing the AI ​​ethical risks and providing (or refusing to provide) AI models is highly beneficial, as it prevents users from inadvertently using AI models in ethically problematic ways (due to lack of knowledge, etc.). Furthermore, for the service provider operating the AI ​​model provision control system 100, it prevents users from using the AI ​​models provided through their service in ethically problematic ways. This contributes to the provision of trustworthy services and is therefore beneficial.

[0049] Furthermore, the above section provided a detailed example of a case where "User A," who belongs to a criminal investigation organization, refuses to provide an AI model for use with a "facial recognition AI model." In this example, the provision of the AI ​​model was refused, but naturally, in cases where the AI ​​ethical risks are deemed to be small, the AI ​​model will be provided.

[0050] For example, if "User C," who belongs to a tutoring school management organization, uses the "facial recognition AI model," a possible use case for the AI ​​model is to verify attendance at classes. Using facial recognition to verify class attendance is generally considered to pose no high ethical risks for AI, so the AI ​​model can be provided.

[0051] As explained above, in this embodiment, even with the same "facial recognition AI model," it is possible to change the decision to refuse or provide it depending on the use case (how it will be used). In assessing the risks of AI ethics, not only what task the AI ​​model performs but also how it is used are important factors, so assessing the risks of AI ethics by combining the task and the use case is an effective method.

[0052] Furthermore, the above describes a method for determining whether or not to use an AI model in the AI ​​model usability determination unit 105, in which a human examines the criteria in advance and stores the results in a database.

[0053] There are no absolute, universally applicable standards for judging AI ethical risks; these standards are likely to vary depending on the country or region where the AI ​​Model Provisioning and Control System 100 service is implemented. This is because AI ethics itself is strongly influenced by social and cultural backgrounds. Furthermore, legal regulations concerning AI ethics generally differ from country to country. It is also easy to imagine that the service provider operating the AI ​​Model Provisioning and Control System 100's views and policies regarding AI ethics will influence the consideration of the criteria for determining whether or not to approve an AI. Moreover, social acceptance of AI ethics may also change over time.

[0054] Therefore, the decision on whether or not to use an AI model may differ depending on the country or region operating the AI ​​model provision control system 100, the service provider operating the AI ​​model provision control system 100, and even the era. The AI ​​model provision control system 100 according to this embodiment is configured to allow for the construction of a flexible system that can absorb such differences in social and historical contexts regarding AI ethics.

[0055] (Second embodiment) In the first embodiment, an example was described in detail regarding the estimation of a use case for when a user uses an AI model by referring to the user's affiliated organization.

[0056] In the second embodiment, we will describe an example in which a use case is estimated by referring to the user's organization and images the user has taken to date.

[0057] Figure 6 is a block diagram showing an example configuration of the AI ​​model provisioning control system 600 according to the second embodiment. In Figure 6, components having the same meaning as in Figure 1 are assigned the same numbers as in Figure 1, and their explanations are omitted.

[0058] The AI ​​model provision control system 600 includes a user information management unit 602, a use case estimation unit 603, an AI model management unit 104, and an AI model availability determination unit 105.

[0059] In this embodiment, the AI ​​model provision control system 600 manages and provides AI models usable with digital cameras, similar to the first embodiment. Furthermore, in this embodiment, in addition to managing and providing AI models, it also manages images taken by users of digital cameras.

[0060] Code 601 is a user of the AI ​​model provision control system 600. User 601 registers as a user by disclosing their personal information when using the AI ​​model provision control system 600. Furthermore, user 601 manages the images taken with their own digital camera within the AI ​​model provision control system 600. This personal information of user 601 and the images taken by user 601 will be collectively referred to as user information.

[0061] Code 602 represents the User Information Management Department. The User Information Management Department 602 manages the association between users 601 and their user information. The User Information Management Department 602 manages the association between each user and their personal information (name and affiliated organization), and the association between each user and their captured images.

[0062] Figure 7 is a schematic diagram of the user information management method in User Information Management Department 602. Figure 7 shows the organizations to which users A, B, and C belong, and how their captured images are managed. For example, it shows that user A belongs to a "criminal investigation organization," and that the images captured by user A so far are managed as image group A.

[0063] Reference numeral 603 denotes a use case estimation unit. When an AI model is provided to a user, the use case estimation unit 603 estimates how the user is likely to use the AI ​​model based on the user's personal information and captured images managed by the user information management unit 602. In the first embodiment, the method for estimating use cases from the user's personal information (affiliated organization) was described in detail, so here we will describe the method for estimating use cases based on the user's captured images.

[0064] Figure 8 is a block diagram of the process within the use case estimation unit 603 that estimates use cases based on captured images.

[0065] The use case estimation unit 603 includes an image recognition processing unit 800 and an image estimation result integration unit 808. The image recognition processing unit 800 includes a face detection unit 801, a face orientation estimation unit 802, a gaze detection unit 803, a face authentication unit 804, a people counting unit 805, a scene estimation unit 806, and an image estimation result integration unit 807.

[0066] In Figure 8, code 800 is the image recognition processing unit. Code 801 is the face detection unit. Code 802 is the face orientation estimation unit. Code 803 is the gaze detection unit. Code 804 is the face recognition unit. Code 805 is the people counting unit. Code 806 is the scene estimation unit. All of these are techniques for estimating the position of objects (e.g., people and faces) in an image, estimating their attributes, and further, estimating the scene in which the image was taken. In other words, they are processes that perform so-called image recognition. All of these processes can be implemented using existing, known image recognition technologies.

[0067] For example, the face detection unit 801 may extract shapes corresponding to the components of the face region, such as the nose, mouth, and eyes, estimate the size of the face from the size of both eyes and their distance, and define the region enclosed by the estimated size region, using the position corresponding to the center of the nose as a reference, as the face region.

[0068] Furthermore, the scene estimation unit 806 may, for example, perform object detection on the image to estimate "what" is depicted and "where," and then estimate the scene based on the positional relationships of the detected objects. For example, if benches and swings are visible in the image, the scene estimation unit 806 will estimate the location where the image was acquired to be a "park." Similarly, if there are buildings on both sides of the road, it may estimate that the scene is "on the street."

[0069] The face orientation estimation unit 802 estimates the orientation of the faces of people in the image. The gaze detection unit 803 detects the direction of the gaze of people in the image. The face recognition unit 804 recognizes the faces of people in the image. The people counting unit 805 counts the number of people in the image.

[0070] In recent years, these image recognition processes are often implemented using AI models. Therefore, image recognition processes represented by symbols 801 to 806 may be implemented using AI models managed by the AI ​​model provision control system 600 (those managed by the AI ​​model management unit 104).

[0071] The image recognition processing unit 800 receives captured images managed by the user information management unit 602 as input, and performs the respective image recognition processes indicated by codes 801 to 806 on the input captured images.

[0072] In Figure 9, symbols 900 and 901 indicate examples of captured images. When image recognition processing is performed on the captured image 900 by the image recognition processing unit 800, the "position of the face," "direction of the face," and "direction of gaze" are estimated for each person in the image. In addition, the number of people counting unit 805 estimates the number of people in the image, and the scene estimation unit 806 estimates the location where the image was taken. The estimation results from each image recognition process are sent to the image estimation result integration unit 807.

[0073] Reference numeral 807 denotes the image estimation result integration unit. The image estimation result integration unit 807 receives the results of the individual recognition processes 801 to 806 that have been executed on the image (for example, image 900) input to the image recognition processing unit 800. For example, the face detection unit 801 inputs the face positions in the image, and the face orientation estimation unit 802 inputs the results of face orientation estimation for each face in the image.

[0074] The image estimation result integration unit 807 obtains an estimation result for a single image, determining what is depicted in that image and how, from these individual recognition processing results. For example, in the case of an image like image 900, the individual inference results from recognition processes 801 to 806 are integrated to obtain the estimation result that "there are 15 people of varying (apparent) sizes, each with their face facing in a different direction, and each looking in a different direction. The scene is on a street."

[0075] Furthermore, the image estimation result integration unit 807 estimates the use case of how the image was taken based on the obtained estimation results. In the case of image 900, since it is an image in which multiple people are shown on the street with their faces facing / looking in different directions, it is estimated that this image was taken with a digital camera of people walking on the street. In other words, it is estimated that image 900 was taken under the use case of "taking images of people in a public space".

[0076] One possible technique for estimating use cases from individual estimation results is to use neural networks, such as deep networks. In other words, by preparing a neural network that takes individual estimation results as input and outputs use cases as output, use case estimation can be achieved.

[0077] Such a neural network can be prepared by training it using the following type of training data: A large number of images with known use cases (the circumstances under which the images were taken) are prepared, and individual recognition processes 801-806 are performed on each image to obtain individual estimation results. By preparing a large number of these pairs of use cases and individual estimation results, and training a neural network using these as training data, a neural network that estimates use cases from individual estimation results can be realized.

[0078] By configuring the image estimation result integration unit 807 as described above, it is possible to estimate the use case of the captured image.

[0079] Reference numeral 808 denotes an inter-image estimation result integration unit. The inter-image estimation result integration unit 808 integrates the use cases output from the intra-image estimation result integration unit 807 across multiple images taken by the same user 601. For example, assuming that image 900 and image 901 are images taken by the same user 601, the estimated use cases for these images are integrated.

[0080] The integration of estimated use cases in the image estimation result integration unit 808 can be achieved using a method such as majority voting. In the case of captured images 900 and 901, since they are very similar images taken in the same location, it is expected that the estimated use case for both images will be estimated to be "taking pictures of people in a public space." In such cases, the estimated use case for the users of captured images 900 and 901 will be "taking pictures of people in a public space."

[0081] As explained in detail above, the use case estimation unit 603 uses the user's captured images to estimate what kind of photography the user prefers and what use cases they take photos in. For example, the use case estimation unit 603 outputs the estimated result that the use case of a user who has taken images like images 900 and 901 is "taking pictures of people in public spaces".

[0082] The use case estimation unit 603 estimates use cases based on images 900 and 901 previously taken by the model requesting user. For example, as shown in Figure 8, the use case estimation unit 603 estimates use cases based on the orientation of people's faces, their gaze, or the number of people in the captured images.

[0083] This embodiment shows an example of estimating a use case by referring to images the user has previously taken, in addition to the user's affiliated organization. If the use case estimated from the affiliated organization differs from the use case estimated from the captured images, it is sufficient to decide in advance which estimation result to prioritize. Alternatively, a process to integrate both estimation results may be provided. For example, as shown in the first embodiment, the use case of "criminal investigation" and the use case of "taking images of people in public spaces" shown in this embodiment are compatible, so the integrated use case may be "criminal investigation using images of people in public spaces."

[0084] The above describes the configuration of the AI ​​model provisioning and control system 600.

[0085] The operation sequence of the AI ​​model provisioning control system 600 is almost the same as that shown in Figure 4, so it is omitted here.

[0086] In this embodiment, the system estimates the use case that the model requesting user (a user who requests the provision of an AI model from the AI ​​model provision control system 100) intends to realize using the AI ​​model (the requested AI model) by referring to the user's past captured images as user information. Then, based on the estimated use case and the executable tasks associated with the requested AI model, the system determines whether or not the AI ​​model is usable.

[0087] This allows us to determine whether a user requesting a model is likely to use the requested AI model in an ethically problematic way. By determining whether or not to allow the use of the AI ​​model based on this assessment, it becomes possible to suppress ethically problematic uses of AI.

[0088] (Third embodiment) In the third embodiment, we describe an example in which the AI ​​model provision control system provides an AI model to the user with an expiration date for its use. By setting an expiration date for the AI ​​model, it is possible to avoid situations where the AI ​​model can still be used even though the use case estimated at the time of provision has changed over time. In other words, it is possible to suppress situations where the AI ​​model can be used even though the use case has changed from the one initially estimated.

[0089] Furthermore, if the expiration date of this usage period is updated when a user who has received an AI model registers newly taken images in the AI ​​model provision control system, it will be possible to check whether the use case has changed through the newly registered images. By configuring it in this way, it is possible to periodically check the use case, which will prevent the AI ​​from being used in a way that is unethical, while also ensuring that user convenience is not compromised.

[0090] The method for stopping the operation of an AI model based on its expiration date can be achieved using existing, publicly known methods. For example, it is common practice to set a license period for software and activate it accordingly, and similar methods can be used to enable or disable the operation of an AI model.

[0091] Figure 10 is a block diagram showing an example configuration of the AI ​​model provisioning control system 1000 according to the third embodiment. In Figure 10, components that have the same meaning as those in Figures 1 and 6 are assigned the same numbers as those in Figures 1 and 6, and their descriptions are omitted.

[0092] The AI ​​model provision control system 1000 includes a user information management unit 1002, a use case estimation unit 603, an AI model management unit 104, an AI model availability determination unit 1005, and a usage period setting unit 1006.

[0093] In this embodiment, the AI ​​model provision control system 1000, similar to the second embodiment, not only manages and provides AI models but also manages images taken by users of digital cameras. Furthermore, as described above, the AI ​​models provided by the AI ​​model provision control system 1000 have an expiration date set for their use.

[0094] Code 1002 represents the User Information Management Unit. Similar to the User Information Management Unit 602, the User Information Management Unit 1002 manages the association between each user and their personal information (name and affiliated organization), as well as the association between each user and their captured images. Furthermore, the User Information Management Unit 1002 also manages the last update date of the captured images managed by the AI ​​Model Provision Control System 1000.

[0095] Figure 12 is a schematic diagram of the user information management method in the user information management unit 1002. Figure 12 shows how the organizations to which users A, B, and C belong, the images they have taken, and the last modified date of those images are managed. For example, it shows that user A belongs to a "criminal investigation organization," the images taken by user A so far are "Image Group A," and the last modified date of those images is managed as "2024.04.01."

[0096] The user information management unit 1002 sends the "last updated date of the captured image" corresponding to the model requesting user (a user who requests the provision of an AI model from the AI ​​model provision control system 1000) to the usage period setting unit 1006.

[0097] Code 1005 is the AI ​​model availability determination unit. The AI ​​model availability determination unit 1005 performs the same determination as the AI ​​model availability determination unit 105. That is, the AI ​​model availability determination unit 1005 determines whether or not to provide the AI ​​model to the user 601 based on the use case estimated by the use case estimation unit 603 and the tasks that the AI ​​model managed by the AI ​​model management unit 104 will realize. In addition to this, the AI ​​model availability determination unit 1005 also takes into account the expiration date input from the expiration date setting unit 1006 to make a final determination as to whether or not to provide the AI ​​model to the user 601.

[0098] To determine whether or not to provide an AI model based on the expiration date, the system compares the entered expiration date with the current date and time. If the expiration date indicates a past date, the AI ​​model is not provided. Conversely, if the expiration date indicates a future date, the AI ​​model is provided. If the model is provided, the expiration date is also provided along with the AI ​​model itself.

[0099] Code 1006 is the usage period setting unit. The usage period setting unit 1006 sets the usage period (the period during which the AI ​​model can operate normally) during which the model requesting user can use the requested AI model, based on the last update date of the images previously taken by the model requesting user, which is sent from the user information management unit 1002. For example, the usage period may be set to one year after the last update date of the images.

[0100] The above describes the configuration of the AI ​​model provisioning and control system 1000.

[0101] The operation sequence of the AI ​​model provision control system 1000 is shown in Figure 11. In Figure 11, steps that perform the same processing as in Figure 4 are assigned the same numbers as in Figure 4, and their explanations are omitted.

[0102] First, in step S1000, the AI ​​model provision control system 1000 receives a request from the user for the provision of an AI model or for continued use. In this embodiment, since an expiration date is set for the AI ​​model provided, a request for continued use may be sent before that expiration date.

[0103] Steps S401 to S405 are the same as those in Figure 4. In step S405, if the AI ​​ethical risk is high, proceed to step S406; otherwise, proceed to step S1108.

[0104] In step S1108, the AI ​​model availability determination unit 1005 determines whether the expiration date indicates a future date or time. If the expiration date indicates a past date or time, the process proceeds to step S406. Conversely, if the expiration date indicates a future date or time, the process proceeds to step S1107.

[0105] In step S406, the AI ​​model availability determination unit 1005 refuses to provide the requested AI model to the model requesting user.

[0106] In step S1107, the AI ​​model availability determination unit 1005 provides the requested AI model to the model requester, with an expiration date for use.

[0107] By configuring the AI ​​model provision control system 1000 in this way, it is possible to encourage users to continuously register captured images (management of captured images in the AI ​​model provision control system 1000). In other words, if captured images are not registered, the AI ​​model's usage period will eventually expire. To prevent the AI ​​model from shutting down due to its expiration date, users must continuously register captured images.

[0108] By using these continuously registered images to estimate use cases, it becomes possible to track changes in the user's use cases over time, even when the AI ​​model's usage period expires.

[0109] This prevents situations where an AI model can be used even if the use case has changed from the one it was estimated for. Furthermore, this configuration allows for regular checks of use cases, preventing the AI ​​from being used in unethical ways while maintaining user convenience.

[0110] The above explanation described an example where the User Information Management Unit 1002 manages the last update date of captured images. However, the essential point is whether or not the captured image was recently registered (whether or not use case estimation can be performed with recently registered captured images). Therefore, any information that shows when the captured image was registered does not necessarily have to be the last update date.

[0111] Furthermore, if the AI ​​model that the user is currently requesting is an AI model that has already been provided, the AI ​​model availability determination unit 1005 can determine that it is available and simply provide a new expiration date for use. This is because the AI ​​model itself has already been provided.

[0112] Furthermore, the images used for use case estimation in the use case estimation unit 603 may be limited to images taken within the period in which the current usage period is valid. In other words, use case estimation may be performed by focusing on the most recent images among the images registered by the user. Focusing on images taken within the period in which the current usage period is valid for use case estimation is preferable because it allows for more immediate tracking of changes in the user's use case.

[0113] In the example above, the expiration date setting unit 1006 set the expiration date to one year after the last update date of the captured image, but the method of setting the expiration date is not limited to this. The service provider operating the AI ​​model provision control system 1000 can decide how far after the last update date the expiration date should be, or how long the expiration date should be.

[0114] Setting the expiration date to the near future allows for quick adaptation to changes in use cases, but it becomes inconvenient for users due to the increased frequency of AI model expiration date updates. Quick adaptation to changes in use cases makes it easier to prevent AI from being used in unethical ways. Therefore, it becomes a trade-off between user convenience and the risk of AI ethical problems arising.

[0115] Furthermore, even if an AI model has already been provided to the user requesting the model in the past, if the AI ​​model availability determination unit 105 determines that the combination of the estimated use case and the previously provided AI model is likely to result in an ethically problematic use of the AI, it may invalidate the previously provided AI model (set its usage period to expire).

[0116] In that case, the AI ​​model availability determination unit 1005 can determine whether the model requesting user is suitable to use the requested AI model, and then, based on the new use case estimated by the use case estimation unit 603, determine whether the model requesting user is not suitable to use the requested AI model.

[0117] For example, the AI ​​model availability determination unit 1005 determines that the model requesting user is suitable to use the requested AI model and provides the requested AI model to the model requesting user. Subsequently, depending on the new use case estimated by the use case estimation unit 603, it can determine that the model requesting user is not suitable to use the requested AI model and disable the use of the requested AI model.

[0118] (Fourth embodiment) In previous embodiments, the AI ​​model availability determination unit has described an example where the decision is either to refuse or to provide the model. However, it is not always necessary to limit the decision to just these two options. As mentioned above, the purpose of this embodiment is to deter the use of AI models in a way that is ethically problematic.

[0119] Therefore, instead of simply refusing, it might be possible to provide the AI ​​model after displaying a warning message. In other words, even if it is presumed that the facial recognition AI model will be used in criminal investigations, as shown in the first embodiment, the facial recognition AI model may be provided along with a warning message such as, "Using facial recognition in criminal investigations may raise ethical concerns regarding AI" or "Using facial recognition in criminal investigations is prohibited by law."

[0120] In that case, if the AI ​​model availability determination unit determines that the model requesting user is suitable to use the requested AI model, it will provide the requested AI model to the model requesting user. On the other hand, if the AI ​​model availability determination unit determines that the model requesting user is not suitable to use the requested AI model, it will display a warning message to the model requesting user and then provide the requested AI model to the model requesting user.

[0121] Furthermore, it is conceivable to provide an AI model with limited functionality. For example, if it is assumed that the facial recognition AI model, as shown in the second embodiment, will be used when taking pictures of people in public spaces, it is conceivable to provide it with functional limitations such that the facial recognition AI model will not operate unless the size of the face to be recognized (the apparent size in the image) is large enough. Since the face size in images like 900 and 901 is small, such functional limitations can effectively make facial recognition for images like 900 and 901 impossible.

[0122] In that case, if the AI ​​model availability determination unit determines that the model requesting user is suitable to use the requested AI model, it provides the requested AI model to the model requesting user. On the other hand, if the AI ​​model availability determination unit determines that the model requesting user is not suitable to use the requested AI model, it provides the requested AI model to the model requesting user after restricting the functionality of the requested AI model.

[0123] As described above, according to the first to fourth embodiments, the AI ​​model provision control system estimates how the AI ​​model will be used if it is provided to the recipient, and if it determines that the intended use is inappropriate, it refuses to provide the AI ​​model. This makes it possible to deter the use of AI models in a way that infringes on people's privacy or human rights. In other words, it makes it possible to deter the use of AI models in a way that is ethically problematic from an AI perspective.

[0124] (Other embodiments) This disclosure can also be implemented by supplying a program that implements one or more of the functions of the embodiments described above 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. It can also be implemented by a circuit (e.g., an ASIC) that implements one or more functions.

[0125] Furthermore, the embodiments described above are merely examples illustrating how to implement this disclosure, and they should not be interpreted as limiting the technical scope of this disclosure. In other words, this disclosure can be implemented in various ways without departing from its technical concept or its main features.

[0126] This embodiment includes the following configuration. (Item 1) A use case estimation means that estimates the use cases in which a user will use the AI ​​model, based on information of a user who intends to use the trained AI model, Based on the use cases estimated by the use case estimation means, an AI model usability determination means determines whether the user is suitable to use the AI ​​model. An AI model provisioning control system characterized by having the following features. (Item 2) The AI ​​model provision control system according to item 1, characterized in that the use case estimation means estimates the use case based on the organizational information to which the user belongs. (Item 3) The AI ​​model provision control system according to item 1 or 2, characterized in that the use case estimation means estimates the use case based on images previously taken by the user. (Item 4) The AI ​​model provision control system according to item 3, characterized in that the use case estimation means estimates the use case based on the orientation of the faces, gaze, or number of people in the captured image. (Item 5) The AI ​​model availability determination means is characterized in that it determines whether the user is suitable to use the AI ​​model based on the tasks that the AI ​​model can perform and the use cases estimated by the use case estimation means, as described in any one of items 1 to 3 of the AI ​​model provision control system. (Item 6) The AI ​​model provision control system according to any one of items 1 to 5, further comprising means for setting an expiration date for which the user can use the AI ​​model. (Item 7) The AI ​​model provision control system according to item 6, characterized in that the usage expiration date setting means sets the usage expiration date based on information about images previously taken by the user. (Item 8) The means for determining whether the AI ​​model can be used is: If it is determined that the user is suitable to use the AI ​​model, and the expiration date indicates a future date and time, the AI ​​model with the expiration date is provided to the user. If it is determined that the user is suitable to use the AI ​​model, and the expiration date of the usage period indicates a past date, the AI ​​model will not be provided to the user. The AI ​​model provision control system according to item 6 or 7, characterized in that if it is determined that the user is not suitable to use the AI ​​model, the AI ​​model is not provided to the user. (Item 9) The AI ​​model availability determination means is characterized in that, after having previously determined that the user is suitable to use the AI ​​model, it determines that the user is not suitable to use the AI ​​model in accordance with a new use case estimated by the use case estimation means, as described in any one of items 1 to 8. (Item 10) The AI ​​model provision control system according to item 9, characterized in that the AI ​​model availability determination means determines in the past that the user is suitable to use the AI ​​model, provides the AI ​​model to the user, and then, in accordance with a new use case estimated by the use case estimation means, determines that the user is not suitable to use the AI ​​model and disables the use of the AI ​​model. (Item 11) The means for determining whether the AI ​​model can be used is: If it is determined that the user is suitable to use the AI ​​model, the AI ​​model will be provided to the user. An AI model provision control system according to any one of items 1 to 10, characterized in that if it is determined that the user is not suitable to use the AI ​​model, the AI ​​model is not provided to the user. (Item 12) The means for determining whether the AI ​​model can be used is: If it is determined that the user is suitable to use the AI ​​model, the AI ​​model will be provided to the user. An AI model provision control system according to any one of items 1 to 10, characterized in that, if it is determined that the user is not suitable to use the AI ​​model, a warning message is displayed to the user and the AI ​​model is provided to the user. (Item 13) The means for determining whether the AI ​​model can be used is: If it is determined that the user is suitable to use the AI ​​model, the AI ​​model will be provided to the user. An AI model provision control system according to any one of items 1 to 10, characterized in that, if it is determined that the user is not suitable to use the AI ​​model, the AI ​​model is provided to the user after restricting the functions of the AI ​​model. (Item 14) A use case estimation step that estimates the use cases in which the user will use the AI ​​model, based on information about the user who intends to use the trained AI model, Based on the use cases estimated in the use case estimation step, an AI model usability determination step determines whether the user is suitable to use the AI ​​model. A processing method for an AI model provisioning control system, characterized by having the following features. (Item 15) A program to cause a computer to function as an AI model providing control system as described in any one of items 1 through 13. [Explanation of Symbols]

[0127] 100 AI Model-Providing Control System 101 User 102 User Information Management Department 103 Use Case Estimation Unit 104 AI Model Management Department 105 AI Model Availability Determination Unit

Claims

1. A use case estimation means that estimates the use cases in which a user will use the AI ​​model, based on information of a user who intends to use the trained AI model, Based on the use cases estimated by the use case estimation means, an AI model usability determination means determines whether the user is suitable to use the AI ​​model. An AI model provisioning control system characterized by having the following.

2. The AI ​​model provision control system according to claim 1, characterized in that the use case estimation means estimates the use case based on the organizational information to which the user belongs.

3. The AI ​​model provision control system according to claim 1, characterized in that the use case estimation means estimates the use case based on images previously taken by the user.

4. The AI ​​model provisioning control system according to claim 3, characterized in that the use case estimation means estimates the use case based on the orientation of the faces, gaze, or number of people in the captured image.

5. The AI ​​model availability determination means determines whether the user is suitable to use the AI ​​model based on the tasks that the AI ​​model can perform and the use cases estimated by the use case estimation means, characterized in that the AI ​​model provision control system according to claim 1.

6. The AI ​​model provision control system according to claim 1, further comprising a means for setting an expiration date for which the user can use the AI ​​model.

7. The AI ​​model provision control system according to claim 6, characterized in that the usage expiration date setting means sets the usage expiration date based on information regarding images previously taken by the user.

8. The means for determining whether the AI ​​model can be used is: If it is determined that the user is suitable to use the AI ​​model, and the expiration date indicates a future date and time, the AI ​​model with the expiration date is provided to the user. If it is determined that the user is suitable to use the AI ​​model, and the expiration date of the usage period indicates a past date, the AI ​​model will not be provided to the user. The AI ​​model provision control system according to claim 6, characterized in that if it is determined that the user is not suitable to use the AI ​​model, the AI ​​model is not provided to the user.

9. The AI ​​model availability determination means is characterized in that, after having previously determined that the user is suitable to use the AI ​​model, it determines that the user is not suitable to use the AI ​​model in accordance with a new use case estimated by the use case estimation means.

10. The AI ​​model availability determination means determines in the past that the user is suitable to use the AI ​​model, provides the AI ​​model to the user, and then, in accordance with a new use case estimated by the use case estimation means, determines that the user is not suitable to use the AI ​​model and disables the use of the AI ​​model, as described in claim 9.

11. The means for determining whether the AI ​​model can be used is: If it is determined that the user is suitable to use the AI ​​model, the AI ​​model will be provided to the user. The AI ​​model provision control system according to claim 1, characterized in that if it is determined that the user is not suitable to use the AI ​​model, the AI ​​model is not provided to the user.

12. The means for determining whether the AI ​​model can be used is: If it is determined that the user is suitable to use the AI ​​model, the AI ​​model will be provided to the user. The AI ​​model provision control system according to claim 1, characterized in that, if it is determined that the user is not suitable to use the AI ​​model, a warning message is displayed to the user and the AI ​​model is provided to the user.

13. The means for determining whether the AI ​​model can be used is: If it is determined that the user is suitable to use the AI ​​model, the AI ​​model will be provided to the user. The AI ​​model provision control system according to claim 1, characterized in that, if it is determined that the user is not suitable to use the AI ​​model, the AI ​​model is provided to the user after restricting the functions of the AI ​​model.

14. A use case estimation step that estimates the use cases in which the user will use the AI ​​model, based on information about the user who intends to use the trained AI model, Based on the use cases estimated in the use case estimation step, an AI model usability determination step is performed to determine whether the user is suitable to use the AI ​​model. A processing method for an AI model provisioning control system, characterized by having the following.

15. A program for causing a computer to function as an AI model providing control system according to any one of claims 1 to 13.

Images