Information provision device, information provision method, information provision program, and information provision system for providing personalized alerts related to AI.

Abstract

To analyze user requests to the AI ​​and provide personalized alerts about the AI ​​related to those requests. [Solution] The system classifies requests made by users of an AI-based application to the AI ​​to identify the request type, identifies alert data associated with the request type, creates personalized alerts about the AI ​​based on the request type, the identified alert data, and user preference data that includes at least the user's knowledge of the identified alert data and their preference for alert reminders, and provides these alerts individually to users of the AI-based application.

Description

Technical Field

[0001] The present invention relates to the technical fields of information processing, content analysis, and personalized alerts, and particularly to an information providing apparatus, an information providing method, an information providing program, and an information providing system that analyze information on AI (papers, reports, etc.) and individually provide personalized alerts regarding AI to users who issue requests to the AI.

Background Art

[0002] AI sometimes returns incorrect or ambiguous answers to user requests. However, users do not always recognize the characteristics, limitations, and problems of AI.

[0003] Non-Patent Document 1 proposes a system that analyzes the text responses of AI (specifically, LLM (Large Language Model)) and provides alerts regarding the response content to users. The alerts of the final prototype system of Non-Patent Document 1 are the overall accuracy of the AI's response ("confidence score"), whether the response contains political bias ("political spectrum"), the probability that the response is paid content ("financial interest"), the identification of the false part of the response ("hallucination"), and the self-evaluation score of reliability ("self-evaluation score"). Also proposed are alerts regarding the information source of the response, the number and quality of information sources, ethical and legal considerations, and restrictions regarding medical expertise (when the interaction with AI relates to medical problems). In the prototype of Non-Patent Document 1, the LLM itself is used to identify the existence of content for which alerts are necessary.

[0004] Patent Document 1 proposes an apparatus, method, and program for providing a personalized report to a user when a document related to an AI model is updated. The personalized report includes notifications regarding improvements in AI related to problems the user has experienced in the past. The personalized report is created by analyzing the user's past problems related to AI and the update information about AI, and checking whether the update information about AI is relevant to the user's problems. Then, the personalized report is sent to the user.

Prior Art Documents

Non-Patent Documents

[0005]

Non-Patent Document 1

[0006]

Non-Patent Document 2

[0007] [Patent Document 1] Japanese Patent Application No. 2023-049216 (Filing Date: March 27, 2023; Title of Invention: Apparatus, Method, and Program for Providing Personalized Update Information Regarding AI Models; Inventor: Vanessa Bracamonte) [Overview of the Initiative] [Problems that the invention aims to solve]

[0008] Non-patent document 1 describes a system that is limited to alerts for a predetermined set of AI problems. However, AI may have other problems that are not limited to that set, and new AI problems are continuously being identified. In addition, because AI is continuously updated, some problems are solved, but new problems may arise, and the system does not have the flexibility to take these into account.

[0009] Patent Document 1 describes an information provision device, information provision method, and information provision program that provide users with personalized reports and notify them of AI updates related to past problems with the user's AI. However, users do not receive information about limitations of other AIs or unresolved issues. Furthermore, Patent Document 1 only sends reports to users after there has been an update to the AI ​​information. In other words, users do not receive information at the time a problem occurs.

[0010] The present invention aims to provide an information provision device, information provision method, information provision program, and information provision system that analyze user requests to AI and provide personalized alerts about AI related to those requests. [Means for solving the problem]

[0011] This invention provides a system for receiving and classifying user requests to an AI during AI-based user interaction (two-way communication; dialogue) in an AI-based application. It also identifies alerts related to user requests based on alert data. Furthermore, it checks user preference data and returns personalized alerts.

[0012] User preference data consists of information about the AI ​​user, including but not limited to the user's level of knowledge about AI, knowledge about alerts, and preferences (user preferences for choices) regarding when to be reminded about alerts. Alert data consists of information about alerts, such as the type and priority of the alert. Alert data is created based on information about AI, such as problems and limitations of AI, and external information, such as the impact of AI on society.

[0013] The alert information providing device according to the present invention analyzes a request made by a user to the AI ​​and provides the user with a personalized alert regarding limitations of the AI ​​related to that request.

[0014] The alert information providing device according to the present invention comprises: a request type identification unit that classifies requests made to the AI ​​by a user of an AI-based application and identifies the request type; an alert data identification unit that identifies alert data related to the request type; and an alert creation unit that creates a personalized alert regarding the AI ​​based on the request type, the identified alert data, and user preference data that includes at least the user's knowledge of the identified alert data and their preference for alert reminders, and provides the alert individually to a user of the AI-based application.

[0015] Furthermore, the alert information providing device may further include an alert data creation and update unit that creates and updates the alert data based on information about the AI ​​and information about the impact of the AI, and a user preference data creation and update unit that creates and updates the user preference data based on information about the user and interaction information with the user's alerts.

[0016] Alternatively, an information provision system may be configured consisting of a device that runs the AI-based application and the alert information provision device.

[0017] Furthermore, the alert information provision method according to the present invention includes the steps of: classifying requests made to the AI ​​by a user of an AI-based application and identifying the request type; identifying alert data related to the request type; and creating a personalized alert regarding the AI ​​based on the request type, the identified alert data, and user preference data that includes at least the user's knowledge of the identified alert data and their preference for alert reminders, and providing the alert individually to the user of the AI-based application.

[0018] Furthermore, the alert information providing program according to the present invention includes a request type specifying unit that classifies requests made by a user who uses an AI-based application to the AI and specifies the request type, an alert data specifying unit that specifies alert data related to the request type, and an alert creation unit that creates a personalized alert regarding the AI based on the request type, the specified alert data, and user preference data including at least the presence or absence of knowledge about the specified alert data of the user and preferences regarding alert reminders, and functions as an information providing device that individually provides the alert to a user who uses the AI-based application.

Advantages of the Invention

[0019] According to the present invention, it becomes possible to provide a user with a personalized alert regarding an AI during an interaction with the AI. By separating the task of providing the personalized alert, the present invention can be implemented for different AI-based applications.

Brief Description of the Drawings

[0020] [Figure 1] It is a diagram showing a configuration example of a personalized alert providing device according to an embodiment of the present invention. [Figure 2] It is a diagram showing a flow of a process of providing a personalized alert in the personalized alert providing device according to the embodiment. [Figure 3(A)] It is a diagram for explaining details of a process when the personalized alert providing device according to the embodiment provides a personalized alert to an AI-based application. [Figure 3(B)] It is a diagram for explaining details of a process (continuation of FIG. 3(A)) when the personalized alert providing device according to the embodiment provides a personalized alert to an AI-based application. [Figure 4(A)] It is a diagram showing the creation and maintenance (update) of alert data in the above embodiment. [Figure 4(B)] It is a diagram showing the creation and maintenance (update) of user preference data in the above embodiment. [Figure 5(A)] It is a diagram showing an operation example of a personalized alert providing device when AI is an LLM (Large Language Model) and the AI-based application in the above embodiment is a chatbot. [Figure 5(B)] It is a diagram showing an operation example (continuation of FIG. 5(A)) of a personalized alert providing device when AI is an LLM and the AI-based application in the above embodiment is a chatbot. [Figure 6(A)] It is a diagram showing an operation example of a personalized alert providing device when the AI-based application in the above embodiment is an image generation application (image creator). [Figure 6(B)] It is a diagram showing an operation example (continuation of FIG. 6(A)) of a personalized alert providing device when the AI-based application in the above embodiment is an image generation application (image creator). [Figure 7] It is a diagram showing a specific example of alert data in the above embodiment. [Figure 8] It is a diagram showing a specific example of user preference data in the above embodiment.

Embodiments for Carrying Out the Invention

[0021] First, some definitions of terms used to explain the embodiments of the present invention are shown. "Interaction" refers to the mutual influence (communication) between two or more entities (subjects, elements, etc.) through actions such as responding to the actions of another entity (rather than a one-sided relationship), or to such two-way exchanges or interactions. In the field of IT, it mainly refers to the interaction between humans and machines or systems, and the "interaction" between humans and AI can sometimes appear as if two humans are having a conversation or dialogue.

[0022] "User" refers to a user who interacts with "AI" by using an "AI-based application". "AI" refers to AI that "interacts" with "users" (for example, the aforementioned LLM (Large Language Model)). An "AI-based application" refers to an application that allows "users" and "AI" to "interact," such as the chatbot application and image generation application (image creator) described later.

[0023] The "Personalized Alert Provider (PA Provider)" is an alert provider according to the present invention. The "PA Provider" can be implemented as an independent service or as part of an "AI-based application". "User preference" refers to the preferences (likes) that a "user" has for multiple choices (options), and "user preference data" refers to data related to "user preference."

[0024] Hereinafter, an example of an embodiment of the present invention will be described with reference to the figures. Figure 1 shows an example configuration of an alert provider according to one embodiment of the present invention. Note that this alert provider is shown as an independent service, not as part of an AI-based application.

[0025] As can be seen from Figure 1, the alert providing device 10 is an information processing device (computer) equipped with a control unit 20, an input unit 30, an output unit 40, a storage unit 50, and a data storage unit 60. The control unit 20 controls the entire alert provisioning device 10, and performs the processing described later by appropriately reading and executing the operating system (OS) and various programs stored in the memory unit 50. The control unit 20 is not particularly limited, but may be a CPU. The memory unit 50 is a storage area for the OS, various programs for enabling the hardware group to function as an alert provider 10, and temporary storage data, and is not particularly limited, but may be ROM, RAM, flash memory, etc.

[0026] The alert provider 10 receives information about an interaction (described later) between a user's request to the AI ​​and a user's alert from an AI-based application via the input unit 30, and outputs (sends) a personalized alert regarding the AI ​​to the AI-based application (or the user using it) via the output unit 40. Furthermore, in order to provide the alert provider 10 with user requests to the AI ​​and user interaction information with alerts, the AI-based application may include an application programming interface (API) for providing the alert provider 10 with the above information.

[0027] Furthermore, the alert provider 10 of this embodiment is not limited to handling one type of AI, but can handle multiple AIs. Therefore, Figure 1 shows that it receives requests and interactions from multiple AI-based apps A, B, ... and sends personalized alerts to multiple apps A, B, ...

[0028] Furthermore, the alert provider 10 receives, via the input unit 30, interaction information between the user's request to the AI ​​and the user's alert, as well as information about the user, information about the AI, and information about the impact of the AI. From this information, it creates and updates alert data and user preference data, which will be described later, and stores them in the data storage unit 60. The process of creating and updating alert data and user preference data will be described later.

[0029] The data storage unit 60 is a storage device that stores various data necessary to provide personalized alerts, and the stored data includes at least alert data and user preference data. The data storage unit 60 is not particularly limited, but may be a hard disk drive (HDD), a solid state drive (SSD), or the like.

[0030] Figure 2 is a diagram showing the flow of the process for providing personalized alerts in the personalized alert provider device 10 (PA provider device) of the embodiment. Referring to Figure 2, when a user submits a request to the AI ​​through an AI-based application, the PA provider receives the user request (step S101) and classifies it (step S102). Here, a single request can be classified into one or more types. The classification of user requests can be carried out in a variety of ways, including but not limited to machine learning-based classification. Furthermore, the process of classifying user requests and identifying request types in step S102 can be performed by a request type identification unit (not shown), which is a functional block provided by the PA provider.

[0031] The PA provider checks the alert data and identifies the alerts related to the type of request (step S103). The alert data includes information such as the alert description, priority, alert hierarchy, and the type of request associated with the alert. The type of request can be associated with one or more alerts. Furthermore, the process of identifying alert data related to the request type in step S103 can be performed by an alert data identification unit (not shown), which is a functional block provided by the PA provider. The PA provider checks user preference data related to the user who made the request (step S104). User preference data includes information about the user's characteristics, such as the user's knowledge level regarding AI and preferences regarding alerts.

[0032] The PA provider creates a personalized alert based on the request type, alert data, and user preference data (step S105), and returns a response to the AI-based application containing the created alert (step S106). The PA provider's response may, but is not limited to, include information about each alert, such as the alert priority, the alert message text, and reminder options after the alert is displayed. Furthermore, the process of creating a personalized alert regarding the AI ​​based on the request type, the identified alert data, and user preference data that includes at least the user's knowledge of the identified alert data and their preferences regarding alert reminders, can be performed by an alert creation unit (not shown), which is a functional block of the PA provider device.

[0033] When there are multiple alerts, the logic for determining which alerts to include in the PA provider's response can be implemented in different ways. For example, when there are multiple alerts, the response can be restricted to include only the highest-priority alerts, restricted based on the user's level of knowledge about AI, or a more complex logic with multiple variables can be used. Alert and reminder options and logic can also be personalized. For example, a user might prefer a learning-based approach to reminders. In that case, the decision to remind about an alert could be based on a learning strategy such as the spaced iteration described in Non-Patent Document 2.

[0034] If the user interacts with a personalized alert (Yes in step S107), the PA provider receives information about the interaction (Step S108) and updates the user preference data (Step S109). If the user does not interact with a personalized alert for any reason (No in step S107), the PA provider asynchronously records the event in the user preference data (Step S110).

[0035] PA (Public Access) devices are not limited to handling a single type of AI. When handling multiple AIs, all logic and data, such as alert data and user preference data, are structured to handle information from multiple AIs.

[0036] Figures 3(A) and 3(B) illustrate the details of the process by which the personalized alert provider 10 (PA provider) of the embodiment provides personalized alerts to an AI-based application. Referring to Figure 3(A), when a user sends a request to an AI-based application, the user request is also sent to the PA provider, which receives it (corresponding to step S101 in Figure 2).

[0037] In the user request classification process (corresponding to step S102 in Figure 2), user requests are classified into, for example, type A and type B. In the alert data checking process (identification of alerts related to the type of request; corresponding to step S103 in Figure 2), for example, alert X related to type A and alerts Y and Z related to type B are identified.

[0038] In the process of checking user preference data related to the user who made the request to the AI ​​(corresponding to step S104 in Figure 2), for example, it is identified that the user who made the request does not have prior knowledge about alert X, but does have knowledge about alerts Y and Z, and the options (choices) that the user has selected regarding reminders are also identified.

[0039] Next, referring to Figure 3(B), the process of creating a personalized alert based on the request type, alert data, and user preference data (corresponding to step S105 in Figure 2) prepares the content of the alert that the PA provider will respond to, for example, a message and reminder options about alert X. In the process of sending a response that includes the created alert (corresponding to step S106 in Figure 2), a message about alert X is sent to the AI-based application.

[0040] After notifying the user of a message about alert X, the user can interact with the alert X that was notified. Figure 3(B) shows two user interactions displayed to the user with checkboxes: "I have learned about alert X, so please do not remind me again" and "Please remind me about alert X again next time," and the user checks the latter checkbox. In this case, the user has interacted with alert X (corresponding to Yes in step S107 of Figure 2), so the PA provider performs the reception processing of the interaction (corresponding to step S108 of Figure 2).

[0041] The PA provider that has received and processed the user interaction performs the user preference data update process (corresponding to step 109 in Figure 2). In this case, the reminder option (user reminder preference) in the user preference data is updated to "Remind user of Alert X".

[0042] Next, we will discuss the creation and maintenance (updating) of support information. The PA provider device also includes the creation and maintenance of support information necessary to provide personalized alerts. Here, the support information necessary to provide personalized alerts includes at least the alert data shown in Figure 4(A) and the user preference data shown in Figure 4(B).

[0043] The alert data shown in Figure 4(A) is generated by analyzing multiple sources of information about AI (such as AI-related documents, research papers, articles, AI models, cards, and AI performance reports) and information about the impact AI has on users (such as documents, research papers, and articles on AI bias and reports on the impact AI errors have had on users in the past). The analysis can be performed manually, automatically, or semi-automatically. The generated alert data may include information such as the alert description, priority, the type of request it is associated with, and the alert hierarchy. Alert data is updated as needed, for example, as a result of updates to AI knowledge, updates to AI, or external events that affect the priority or content of alerts.

[0044] The user preference data shown in Figure 4(B) is created based on user characteristics and user preferences regarding AI and alerts. For example, it includes the user's general AI knowledge level, specific knowledge about alerts, whether the user likes to learn about AI, and records of interactions with personalized alerts. User preference data is updated as needed, such as when requested by the user or after personalized alerts have been sent.

[0045] The following describes examples of the operation of the PA (Public Address) provisioning device of the present invention in the case of an AI-based application, specifically a chatbot (Figures 5(A) and 5(B)) and an image creator (Figures 6(A) and 6(B)). In addition, we will explain specific examples of alert data (Figure 7) and user preference data (Figure 8) when providing personalized alerts to different AI-based applications such as chatbots and image creators.

[0046] <Example of operation in a chatbot> Figures 5(A) and (B) show examples of how a PA (Public Address) provider operates when the AI ​​is the aforementioned LLM (Large Language Model) and the AI-based application is a chatbot. Referring to Figure 5(A), when a user sends a request to the AI ​​via the chatbot, "Which number is bigger, 9.9 or 9.11? Translate the answer into Japanese," the PA provider receives the user request from the chatbot and classifies it into two types: "math-related" and "translation into Japanese."

[0047] The PA provider checks the alert data and identifies "math limitations" alerts related to the "math-related" type, and "limited translation support" alerts and "low performance in Japanese translation" alerts related to the "translation to Japanese" type. The PA provider checks user preference data and identifies that the user has no prior knowledge of the "math limitations" alert, but does have knowledge of the "limited translation support" alert and the "poor performance in Japanese translation" alert. It also identifies that the user's preference for reminders is interval repetition.

[0048] Next, referring to Figure 5(B), the PA provider references the request type, alert data, and user preference data to create the message "This AI may give incorrect answers to math questions!" as the content of the alert to respond to the chatbot, and sends it to the chatbot. The alert message received by the chatbot is then notified to the user.

[0049] In this example, the AI, upon receiving a user request, gives the user an incorrect answer. Specifically, the AI ​​responds, "While 9.9 may appear larger than 9.11, 9.11 is actually larger. When comparing decimal numbers, we first compare the digit before the decimal point, and then compare the digits after the decimal point from left to right. Therefore, 9.11 is larger than 9.9."

[0050] For example, in a 100-meter race in track and field, 9.11 seconds is faster than 9.9 seconds. When comparing the numbers 9.9 and 9.11, 9.9 is larger (note that the "." (period) in this case represents the radix point). According to the mathematical rules of positional notation, when comparing the size of the fractional part to the right of the decimal point, just as when comparing the size of the integer part to the left of the decimal point, it is correct to compare them in order from the highest digit.

[0051] On the other hand, for example, in the United States Patent Rules (Code of Federal Regulations Title 37; Patents, Trademarks and Copyrights), §1.9 is a lower number than §1.11. In this case, §1.9 refers to Chapter 1, Section 9, and §1.11 refers to Chapter 1, Section 11, and the "." (period) is not a decimal point but simply a separator. Therefore, the numbers to the right of the period are not ranked from left to right but are compared as a whole (9 < 11), resulting in §1.9 being a lower number than §1.11. In other words, the rule for determining the magnitude of the numbers to the right of the period in this case differs from mathematical rules (hereinafter, this rule will be referred to as the "non-mathematical rule" for convenience).

[0052] This AI gives an explanation similar to the (mathematical) rule of comparing decimal numbers according to positional numerals, stating, "It compares the decimal parts from left to right." However, it then derives the conclusion, "Therefore, 9.11 is greater than 9.9," which follows a non-mathematical rule. The explanation and conclusion of the answer contradict each other. Thus, it is evident that this AI has mathematical limitations.

[0053] After notifying the user of a "math limit" alert, the user can interact with the alert. Figure 5(B) shows two user interactions displayed to the user with checkboxes: "I have learned about the 'math limit' alert, so please do not remind me again" and "Please remind me about the 'math limit' alert again next time." The user checks the latter checkbox. In this case, since the user has interacted with the "math limit" alert, the PA provider processes the interaction, updates the user preference data, and reminds the user the next time a "math limit" alert is created.

[0054] <Example of operation in Image Creator> Figures 6(A) and (B) show examples of how a PA (Public Address) provider operates when the AI ​​is an image generation AI and the AI-based application is an image generation application (image creator). Referring to Figure 6(A), when a user sends a request to the AI ​​through the image creator saying, "Create a dinosaur-themed birthday card with the message 'Happy Birthday'!", the PA provider receives the user request from the image creator and classifies it as an "image with text" type.

[0055] The PA provider checks the alert data and identifies the "Restrictions on generating images with text" alert related to the "Images with text" type. The PA provider checks user preference data and identifies that the user has no prior knowledge of the "Restrictions on generating images with text" alert. It also identifies that the user's preference for reminders is for interval repetition.

[0056] Next, referring to Figure 6(B), the PA provider, referencing the request type, alert data, and user preference data, creates the alert message "This AI may generate incorrect text in images!" as the content of the alert to respond to the image creator, and sends it to the image creator. The alert message received by the image creator is then notified to the user.

[0057] In this example, the AI, upon receiving a user request, generates an image of a dinosaur-themed birthday card. However, despite the user requesting a birthday card with the message "Happy Birthday," the generated image contains incorrectly spelled or extraneous text such as "HAppppy," "HanpDday," and "Birt·hday." This demonstrates that the AI ​​generated incorrect text within the image.

[0058] After notifying the user of the "Restrictions on generating images with text" alert, the user can interact with the alert. Figure 6(B) shows two user interactions displayed to the user with checkboxes: "I have learned about the 'Restrictions on generating images with text' alert, so please do not remind me again" and "Please remind me again about the 'Restrictions on generating images with text' alert next time." The user checks the latter checkbox. In this case, since the user has interacted with the "Restrictions on generating images with text" alert, the PA provider processes the interaction, updates the user preference data, and reminds the user the next time the "Restrictions on generating images with text" alert is created.

[0059] <Specific examples of alert data> Figure 7 shows a specific example of alert data. Alert data consists of basic information and other information. Basic data includes, for each alert data, at least the alert ID, information identifying the AI ​​that is the subject of the alert (e.g., AI name), the type of user request associated with it, the alert message, and information indicating the priority. Other information includes alert hierarchy information. For example, an alert named "Poor Performance in Japanese Translation" may include information indicating that it is a sub-level of the alert named "Limited Translation Support".

[0060] <Specific examples of user preference data> Figure 8 shows a specific example of user preference data. User preference data consists of basic information, a record of the user's interactions with alerts, and other user information. Basic information may include, at a minimum, information about the user's general AI knowledge level, knowledge of specific alerts related to AI, and reminder-related preferences. The record of user interactions with alerts can include, for each alert, the number of times it was viewed, the date and time it was last viewed, and the user's selection.

[0061] The provision of personalized alerts by the PA provider device 10 in Figure 1 is achieved by an alert provision method that follows the procedure of the processing flow in Figure 2. Furthermore, this process is executed by a command from an alert provision program installed in the storage unit 50 of the PA provision device 10. This program may be distributed by recording it on removable media such as a CD-ROM or DVD-ROM, or by downloading it to the PA provision device 10 via a network. In addition, this program may be executed from the PA provision device 10 as a web service via the network without being downloaded.

[0062] Furthermore, this invention makes it possible, for example, for users to receive personalized alerts about AI during their interaction with AI, and by separating the task of providing personalized alerts, it becomes possible to provide personalized alerts to different AI-based applications. This contributes to Goal 9 of the United Nations-led Sustainable Development Goals (SDGs), "Build resilient infrastructure, promote sustainable industrialization and foster innovation."

[0063] Although embodiments of the present invention have been described above, the present invention is not limited to the embodiments described above. The embodiments described above are examples in which the PA provisioning device is implemented as an independent service rather than as part of an AI-based application, but the PA provisioning program can also be implemented as part of an AI-based application. Furthermore, the system can also include a device that runs an AI-based application and a PA provisioning device.

[0064] Furthermore, the effects described in the embodiments above are merely a list of the most typical effects resulting from the present invention, and the effects of the present invention are not limited to those described in the embodiments. [Explanation of Symbols]

[0065] 10. Alert provision device 20 Control Unit 30 Input section 40 Output section 50 Storage section 60 Data storage unit

Claims

1. A request type identification unit that classifies requests made to the AI ​​by users of an AI-based application and identifies the request type, An alert data identification unit identifies alert data related to the aforementioned request type, The system comprises an alert creation unit that creates a personalized alert regarding the AI ​​based on the request type, the identified alert data, and user preference data that includes at least the user's knowledge of the identified alert data and their preferences regarding alert reminders, An information provider that individually provides the alerts to users of the AI-based application.

2. An alert data creation and update unit creates and updates the alert data based on information about the AI ​​and information about the impact of the AI, A user preference data creation and update unit creates and updates the user preference data based on information about the user and interaction information with user alerts, The information providing device according to claim 1, further comprising:

3. An information provision system comprising a device for executing the aforementioned AI-based application and an information provision device according to claim 1 or claim 2.

4. The steps include classifying requests made by users of an AI-based application to the AI ​​and identifying the request type, The steps include identifying alert data related to the aforementioned request type, The process includes the steps of creating a personalized alert for the AI ​​based on the request type, the identified alert data, and user preference data that includes at least the user's knowledge of the identified alert data and their preferences regarding alert reminders, A method for providing information to individually provide the aforementioned alerts to users of the aforementioned AI-based application.

5. Computers, A request type identification unit that classifies requests made to the AI ​​by users of an AI-based application and identifies the request type, An alert data identification unit identifies alert data related to the aforementioned request type, The system comprises an alert creation unit that creates a personalized alert regarding the AI ​​based on the request type, the identified alert data, and user preference data that includes at least the user's knowledge of the identified alert data and their preferences regarding alert reminders, An information provision program that functions as an information provision device to individually provide the alerts to users of the AI-based application.

Images