Service provision system

Abstract

To make it possible to provide a comprehensive service of information provision or the like including the entire artificial intelligence of a plurality of specialized learned models even while providing a user a service using artificial intelligence of the respective specialized learned models.SOLUTION: A service provision system includes a plurality of slave AI groups 117 capable of providing users 1 to 5 with a service that specializes respective contents A, B, C on the basis of the contents, and a master AI 116 for exchanging information with the plurality of slave AI. The master AI 116 selects slave artificial intelligence desired by an accessing user among slave artificial intelligence usable by the user to permit its use, and provides the user with a service including all of the plurality of slave AI usable by the accessing user on the basis of information exchanged with the plurality of slave AI groups 117.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a service providing system that uses artificial intelligence to provide comprehensive services to users who use multiple contents. [Background technology]

[0002] A technology has been proposed in which a trained model based on machine learning of artificial intelligence is trained to learn the content of an external site, and the trained model then executes processing based on that content (for example, Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 7446023 Summary of the Invention [Problem to be solved by the invention]

[0004] One idea to improve on this technology is to construct a trained model using a machine learning model that has processing capabilities specialized for the type of content that is trained or loaded into the trained model.

[0005] For example, if a study guide for test takers is trained or loaded into a trained model as content, it is desirable for the trained model (artificial intelligence) to have processing functions specialized for study guide content, such as generating questions based on the study guide and testing them with the user (test taker).

[0006] Furthermore, for example, in the case of content for English conversation lessons, it is desirable for the model to be a trained model (artificial intelligence) with processing functions specialized for English conversation lesson content, such as picking out new words and phrases that appear in each unit and presenting them to the user, or checking and correcting the user's English pronunciation.

[0007] In light of this situation, one possible solution is to create specialized trained models with processing functions optimally designed to match each piece of content, and to build an artificial intelligence system that uses each specialized trained model to provide users with services optimally designed for the target content.

[0008] However, in an artificial intelligence system constructed in this way, the following new problems arise:

[0009] The most efficient way to solidify learned material in memory is to repeat 70% interval review, in which the material is reviewed just before it is about to be forgotten (for example, when 70% of the material is still in memory on the forgetting curve). For example, if a user is studying using three specialized trained models—one for content A, one for content B, and one for content C—the next review times for each piece of content will be misaligned, making it difficult for the user to grasp all of the review times. This problem becomes more serious as the number of content items to be studied increases. For example, in the case of the Common University Entrance Examination, which covers 30 subjects across six subjects, it becomes extremely difficult to accurately grasp the review times for all of the subjects.

[0010] In other words, while artificial intelligence using specialized trained models has the advantage of being able to provide users with services that are optimally designed for the target content, it also has a new disadvantage in that if a user uses multiple specialized trained models, it is unable to provide comprehensive services such as information that encompasses all of those multiple specialized trained models.

[0011] These issues arise not only in the field of learning but also in other fields (such as healthcare). Some possible examples are described below.

[0012] Examples of healthcare content include electronic medical record data and electronic medication history data from Selecon, various measurement data from Apple Watch (registered trademark) (heart rate and irregular heart rhythm, atrial fibrillation history, decreased cardiopulmonary function level, oxygen level taken into the body, etc.), calorie intake data from the HEALBE smart watch, recorded data on calories burned and distance from a running app, and body fat measurement history data from a body fat measurement and management app.

[0013] Even if specialized trained models with processing functions optimized to match each of these content data are generated and services optimized for the target content are provided to users using each specialized trained model, a new drawback arises: comprehensive services, such as information provision across all of the multiple specialized trained models, cannot be provided. For example, atrial fibrillation history data from an Apple Watch reveals heart rate and irregular heart rhythm, but the cause cannot be identified from this content alone. However, by comparing the atrial fibrillation history data with body fat measurement history data from a body fat measurement and management app, it is possible to determine that the heart rate and irregular heart rhythm are caused by obesity. Furthermore, by comparing calorie intake data from the HEALBE smartwatch and calorie expenditure data from a running app, it is possible to determine that obesity is caused by excessive or insufficient calorie intake.

[0014] The present invention was devised in light of this situation, and its purpose is to provide users with services using the artificial intelligence of each specialized trained model, while also enabling the provision of comprehensive services such as information provision that encompasses the entire range of artificial intelligence of multiple specialized trained models. [Means for solving the problem]

[0015] One aspect of the present invention is a service providing system that provides a comprehensive service to a user who uses a plurality of contents using artificial intelligence, the system comprising: a plurality of slave artificial intelligences capable of providing users with services specialized for the content based on the acquired content data; a master artificial intelligence that exchanges information with the plurality of slave artificial intelligences; The plurality of slave artificial intelligences are available for use by a user on the condition that the user has selected them as targets for use, The master artificial intelligence is A first function of selecting a slave AI desired by the accessing user from among the slave AIs available to the user and allowing the user to use the slave AI; and a second function of providing the accessing user with a comprehensive service covering all of the multiple slave artificial intelligences that can be used by the user, based on information exchanged with the multiple slave artificial intelligences.

[0016] With this configuration, each slave artificial intelligence can provide a user with a service that is specialized for the content, while also providing the user with a service that encompasses all of the multiple slave artificial intelligences that the accessing user can use.

[0017] Another aspect of the present invention is a service providing system that includes a server that provides a service using artificial intelligence, and a terminal that can be connected to the server via the Internet, and that provides a comprehensive service to a user who uses a plurality of contents using artificial intelligence, a plurality of slave artificial intelligences capable of providing users with services specialized for the content based on the acquired content data; a master artificial intelligence that exchanges information with the plurality of slave artificial intelligences; The plurality of slave artificial intelligences are available for use by a user on the condition that the user has selected them as targets for use, The master artificial intelligence is A first function of selecting a slave AI desired by the accessing user from among the slave AIs available to the user and allowing the user to use the slave AI; and a second function of providing the accessing user with a comprehensive service covering all of the multiple slave artificial intelligences that can be used by the user, based on information exchanged with the multiple slave artificial intelligences.

[0018] With this configuration, it is possible to provide the user with a service that is specialized for the content using each slave artificial intelligence, while also providing the user with a service that encompasses all of the multiple slave artificial intelligences that the user can use.

[0019] Preferably, the plurality of contents include study content used by the user for studying; The second function includes a review schedule notification function that notifies the user who has studied the plurality of contents of a review schedule indicating when to review each content.

[0020] With this configuration, a user who has studied a plurality of contents does not have to manage review times by himself.

[0021] More preferably, the plurality of slave artificial intelligences provide, as a service specialized in the content, a test generation means for generating a test consisting of questions based on the acquired learning content; an answer marking means capable of providing a marking service to a user for marking the user's answers to the test generated by the test generating means, The review schedule notification function notifies the user of a review schedule indicating the optimal review time personalized for the user, calculated according to the user's unique forgetting curve derived from the results of scoring the user's answers.

[0022] With this configuration, the user can review based on a review schedule that is most suitable for him or her.

[0023] More preferably, the slave artificial intelligence: a weak point determination means for determining weak points in the learning of an individual user; and weak part review timing adjustment means for adjusting the timing for reviewing the parts determined to be weak by the weak part determination means to be earlier.

[0024] With this configuration, weak points tend to be forgotten earlier than usual, but by speeding up the review period, it is possible to deal with early forgetting of weak points.

[0025] More preferably, the weak point determination means determines the weak points of an individual user by comparing the scoring results by the answer scoring means for multiple users studying the same learning content with the scoring results by the answer scoring means for the individual user.

[0026] More preferably, the slave artificial intelligence: a difficulty level determination means for determining the difficulty level of each learning subject based on the results of scoring by the answer scoring means for multiple users studying the same learning content; and a difficulty level-based review timing adjustment means for adjusting the review timing for the learning subject based on the determination result of the difficulty level determination means.

[0027] More preferably, the slave artificial intelligence: an examination question acquisition means for acquiring data on examination questions given to examinees from the learning content; and test generation means for generating a test consisting of questions based on the learning content, with reference to the test questions acquired by the test question acquisition means.

[0028] With this configuration, users can take high-quality tests that are generated with reference to test questions given to examinees.

[0029] Another aspect of the present invention is a learning support service providing system for providing a service to support foreign language learning, comprising: a content acquisition means for acquiring content for foreign language learning; output means for outputting the acquired content to a user terminal; a notification means for notifying the user of the meaning of a selected and specified part in the user's native language when the user selects and specifies a part that the user does not understand in the content output by the output means; a review time notification means for calculating a review time for the user to review the part that the user has studied and notifying the user of the time; a test setting means for generating a test consisting of questions based on the selected and designated portion when the user reviews the material, and setting the test to the user; and a score notification means for scoring the user's answers to the test set by the test setting means and notifying the user of the scores, The review time notification means has a review schedule notification function for notifying the user of a review schedule indicating the review time for each content when the user is studying a plurality of content items.

[0030] With this configuration, when the user later reviews the parts he or she has selected and specified in the content he or she is studying and does not understand, he or she can take a test consisting of questions based on the parts he or she does not understand, and the user can review only the parts he or she does not understand.

[0031] Preferably, the review time notification means calculates and notifies the user of an optimal review time personalized for the user, calculated in accordance with the user's unique forgetting curve derived from the scoring results of the user's answers to the test. [Brief explanation of the drawings]

[0032] [Figure 1] 1 is a conceptual diagram of a service providing system that uses artificial intelligence to provide comprehensive services to users who use multiple contents. [Figure 2] 1 is a diagram illustrating the overall configuration of the present service providing system. [Figure 3] (A) is a diagram showing personalized data for each user, and (B) is a diagram showing data stored in a database of GPTs that can be used by each user. [Figure 4] FIG. 2 is a diagram illustrating a hardware configuration of the present service providing system. [Figure 5] FIG. 1 is a graph illustrating the optimal review interval according to the human forgetting curve. [Figure 6] FIG. 1 is a diagram illustrating a method for obtaining a general model of an optimal review interval by regression in machine learning. [Figure 7] 1 is a flowchart showing the main routine between the developer terminal and the ChatGPT cloud server. [Figure 8] 10 is a flowchart showing a subroutine program of a slave GPTs creation process and a slave GPTs creation support process. [Figure 9] 10 is a flowchart showing a subroutine program of a master GPTs creation process and a master GPTs creation support process. [Figure 10] 10 is a flowchart showing a subroutine program of the upgrade process to the GPT Store and the upgrade support process to the GPT Store. [Figure 11] 10 is a flowchart showing a subroutine program of master GPTs access processing and master GPTs access response processing. [Figure 12] 10 is a flowchart showing a subroutine program of a slave GPTs usage process and a slave GPTs usage adaptation process. [Figure 13] 10 is a flowchart showing the continuation of the flowchart showing the subroutine program of the slave GPTs usage support processing and a flowchart showing the ChatGPT operation processing program. [Figure 14] FIG. 10 is a diagram showing a display screen of a user terminal in a modified example. [Figure 15] 10 is a flowchart showing a subroutine program of a slave GPTs usage process and a slave GPTs usage adaptation process in a modified example. [Figure 16] 10 is a flowchart showing a subroutine program of learning processing by a user terminal and learning response processing by slave GPTs in a modified example. [Figure 17] FIG. 10 is a schematic diagram illustrating the overall configuration of the service providing system according to a modified example. DETAILED DESCRIPTION OF THE INVENTION

[0033] 1 is a conceptual diagram of a service providing system that provides comprehensive services to users who use multiple contents using artificial intelligence (AI). In this embodiment, artificial intelligence is expressed as AI (Artificial Intelligence).

[0034] In this service providing system, slave AIs A, B, and C of slave AI group 117 acquire various pieces of content A, B, and C, respectively, and provide users with services specialized for the acquired content based on the acquired content.

[0035] For example, if content A is teaching materials for English reading comprehension (such as English news), it may have a specialized function that provides the user with services specialized for English reading comprehension teaching material content, such as displaying the Japanese meaning of unknown words or phrases in the displayed English text by selecting and specifying them. It may also have a specialized function that provides the user with services specialized for English learning, such as checking and correcting the user's English pronunciation. This English news and other content exists on the Web via the Internet, and the slave AI and the content are linked via an API (Application Programming Interface) so that the slave AI can obtain the content.

[0036] Slave AI A has acquired content A, slave AI B has acquired content B, and slave AI C has acquired content C. Users 1, 3, and 5 have selected and designated slave AI A and can use slave AI A. Users 2, 3, and 5 have selected and designated slave AI B and can use slave AI B. Users 1, 2, and 4 have selected and designated slave AI C and can use slave AI C.

[0037] The master AI 116 stores the slave AIs available to each user. When a user accesses the master AI 116 and specifies a slave AI available to the user, the master AI 116 determines whether the specified slave AI is available to the user, and if it is available, the master AI 116 permits access to the slave AI.

[0038] The master AI 116 has a function to provide a service that encompasses multiple slave AIs that can be used by a user. For example, if contents A, B, and C are study contents, it is desirable for the user to review them at an appropriate time after the initial study, and a service is provided that notifies the user of the review period for each content. For example, since the available slave AIs for user 1 are slave AI A and slave AI C, a review schedule (e.g., a one-week review schedule) created based on the review period for content A and the review period for content C is notified.

[0039] Figure 2 shows a service provision system that uses GPTs, an example of AI, and books, an example of content. Master GPTs 16 and a group of slave GPTs 17 may be linked via an API (Application Programming Interface) or the like to exchange information between them, but in Figure 2, information exchange between the master GPTs 16 and a group of slave GPTs 187 is performed via ChatGPT (registered trademark) 15.

[0040] First, a developer operates the developer terminal 54K to create the master GPTs 16 and various slave GPTs A23, B24, Z24, and other GPTs 22. The creation of these will be described later with reference to Figures 8 and 9. Note that the slave GPTs A23, B24, and Z24 may be created by each publisher operating the developer terminal 54K. Data for books A, B, and Z are input into the slave GPTs A23, B24, and Z24, respectively. Each slave GPT may not only acquire and store content, but may also perform machine learning based on that content. While GPTs are pre-trained AI models, transfer learning and fine-tuning based on the acquired content may also be performed.

[0041] The master GPTs 16 builds a database 19 that stores GPTs that can be used by each user, and has the function of presenting each user's review schedule and other functions 20.

[0042] Slave GPTsZ25 stores data common to all users (past questions, difficulty level for each unit, etc.) 27, Book Z data 25, and personalized data 26 for each user. Slave GPTsZ25 acquires past questions related to Book Z from test questions given to examinees. Slave GPTsZ25 is configured to be able to perform a function 28 for setting questions from Book Z and a personalization function 29 for each user based on this data. Other slave GPTsA23, slave GPTsB24, etc. also have similar data and functions.

[0043] A user can access ChatGPT15 by operating a user terminal 54, view the review schedules of each slave GPT that the user can use, access a slave GPT that is approaching the time for review, take a test using the book question function, and then review using the book that the slave GPT is responsible for.

[0044] The personalized data 26 for each user will be explained with reference to Figure 3(A). This data is composed of a two-dimensional table of each user's user ID and personalized data. The personalized data includes learning history, Ti = T0·i, and strengths and weaknesses. Ti = T0·i is a general model showing the relationship between the number of reviews and the review interval, as described below with reference to Figure 6. T0 in Ti = T0·i refers to the period from the user's first learning to just before forgetting (specifically, the time from the first learning until the memory retention rate reaches 70%), as described below with reference to Figure 6. T0 in Ti = T0·i stored in the personalized data 26 for each user is different for each user and is a value unique to each user, stored corresponding to each user ID. Furthermore, the general model Ti = T0·i may not be a linear function like this for some users, and for such users, a unique curve function matching the user is stored.

[0045] Referring to FIG. 3(B), a database 19 storing GPTs available to each user will be described. The stored data of GPTs available to each user is configured as a two-dimensional table consisting of each user's user ID, available GPTs, next review time, and weekly review schedule. For example, for a user with user ID u3gf95, the available GPTs are slave GPTsA, GPTsZ, and GPTsY, and the next review times for each are March 18, 2024, April 1, 2024, and March 22, 2024. The weekly review schedule is sentences 20-25 of book A on March 18, 2024, and sentences 28-35 of book Y on March 22, 2024.

[0046] The hardware configuration of the service providing system will be described with reference to Figure 4. Each terminal 54S of publishers 18A to 18Z is connected via the Internet 50 to a cloud server 51 in a data center 44 on the cloud 43. This cloud server 51 is connected to a GPS database 60 in a GPT Store 52. The GPT Store 52 is a platform that allows users to browse, search, and use GPTs specialized for specific purposes and tasks, and to create and share original GPTs. Each user terminal 54 is also connected via the Internet 50 to the cloud server 51 in the data center 44.

[0047] The cloud server 51 is provided with a GPU (Graphics Processing Unit) 10g as a control center. The cloud server 51 is configured with a storage unit such as a RAM (Random Access Memory) 9 that functions as a work area for the GPU 10g, a ROM (Read Only Memory) 11 that stores data and programs, an SDD (Solid State Drive) 12, an input operation unit 7 such as a display and a keyboard, a communication unit 5, a display unit 6, an interface 8, a bus 13, and various other hardware. Note that a HDD (Hard Disk Drive) may be used as the storage unit in addition to or instead of the SDD.

[0048] The hardware configurations of the developer terminal 54K, the user terminal 54, and the terminal 54S are almost the same as that of the cloud server 51, and furthermore, a speaker and a camera are connected to the interface 8. Also, a CPU (Central Processing Unit) is provided instead of the GPU 10g.

[0049] Referring to Figure 5, we will explain the most efficient review method for solidifying memories in the human brain. The degree to which a memory is solidified in the brain is not proportional to the number of times it is reviewed, but rather to the total period of time it is temporarily stored in the brain. Therefore, the most efficient review method is interval review, in which you wait until just before forgetting the content of the temporary memory and then review it. Specifically, as shown in Figure 5, it is effective to repeat 70% interval review, which reviews the content at a point where the memory retention rate is 70% on Ebbinghaus's forgetting curve.

[0050] Figure 6 is a graph in which the vertical axis indicates the time Ti until a 70% memory retention rate is reached, and the horizontal axis indicates i (the number of reviews indicating how many times the review has been). The experimental results for three users with user IDs u3gf95, pi7n52, and 37gpt6 are plotted as circles, crosses, and triangles. Using a "regression" algorithm, the data from these multiple users is used as input data, and this input data is considered to output a target based on a certain function, and this function is found. The function found is: Ti=T0·i The coefficient T0 is a factor that varies from user to user (memorization ability, etc.). Specifically, it is the period from the first learning to just before forgetting for each user (specifically, the time from the first learning until the memory retention rate reaches 70%). In this way, the factor that varies from user to user (memorization ability, etc.) is expressed as a constant T0 in a general model as follows: Ti=T0·i The time T0 from the initial learning to when the memory retention rate reaches 70% is stored by each slave GPT as personalized data for each user (Figures 2 and 3(A)). This T0 is used to calculate the timing of each user's next review (i-th review). More specifically, the timing of the next review (i-th review) is calculated taking into account the user's strengths and weaknesses (Figure 3(A)). For units or content in which the user is weak, the timing of the next review (i-th review) can be adjusted to be earlier. Conversely, for units or content in which the user is strong, the timing of the next review (i-th review) can be adjusted to be later. Furthermore, since the general model Ti = T0·i described above may not be a linear function but a curve depending on the user, the test results from each slave GPT are continuously monitored and adjusted.

[0051] 7, a flowchart of the main routine program between the developer terminal 54K and the cloud server 51 of ChatGPT 15 will be described. The developer terminal 54K performs a slave GPTs creation process in step S (hereinafter simply referred to as "S") 1. In response to this, the cloud server 51 performs a slave GPTs creation corresponding process in step S10.

[0052] The developer terminal 54K performs a process for creating a master GPTs in S2. In response, the cloud server 51 performs a process for handling the creation of a master GPTs in S11. The developer terminal 54K performs a process for uploading to the GPT Store 52 in S3. In response, the cloud server 51 performs a process for handling uploading to the GPT Store 52 in S12. Through this process, the created master GPTs 16 and slave GPTs 22, 23, 24, ... 25 are stored in the GPTs database 60. The developer terminal 54K performs other processes in S4. In response, the cloud server 51 performs other handling processes in S13.

[0053] A flowchart of the subroutine program for the slave GPTs creation process in S1 and the slave GPTs creation support process in S10 will be described with reference to Figure 8. The developer terminal 54K determines whether to create a slave GPT in S20. If it is determined not to create a slave GPT, the process returns and control moves to S2. If the developer performs an operation to create a slave GPT on the developer terminal 54K, the determination in S20 is YES, and ChatGPT 15 is accessed in S21.

[0054] The cloud server 51 determines in S29 whether or not there has been access, and if it determines that there has been no access, it returns and the control proceeds to S11. If there has been access in S21, the determination in S29 is YES and the control proceeds to S30.

[0055] On the developer terminal 54K, the following functions are requested while communicating with ChatGPT via S22. - Function to remember data such as entered books - Question function based on data from books, etc. · Ability to generate and store personalized data for each user - Function to calculate each user's review schedule -Storage function for data common to all users, etc. In response to this request, the cloud server 51 creates a slave GPT with the following functions requested by S30: - Function to remember data such as entered books - Question function based on data from books, etc. · Ability to generate and store personalized data for each user - Function to calculate each user's review schedule -Storage function for data common to all users, etc. A flowchart of the subroutine program for the master GPTs creation process in S2 and the master GPTs creation support process in S11 will be described with reference to Figure 9. The developer terminal 54K determines whether to create a master GPTs in S40. If it is determined that a GPT will not be created, the process returns and control moves to S3. If the developer performs an operation to create a master GPTs on the developer terminal 54K, the determination in S40 is YES, and ChatGPT 15 is accessed in S41.

[0056] The cloud server 51 determines in S49 whether or not there has been access, and if it determines that there has been no access, it returns and the control proceeds to S12. If there has been access in S41, the determination in S49 is YES and the control proceeds to S50.

[0057] On the developer terminal 54K, the following functions are requested while communicating with ChatGPT via S42. - Ability to exchange information with each slave GPT via ChatGPT - Building a database that stores the GPTs available to each user - Presentation of each user's review schedule Other GPTs reading functions, etc. In response to this request, the cloud server 51 creates a master GPT with the following functions requested by S30: - Ability to exchange information with each slave GPT via ChatGPT - Building a database that stores the GPTs available to each user - Presentation of each user's review schedule Other GPTs reading functions, etc. The slave GPTs and master GPTs described above can be created in two ways: using the "GPT Builder" provided in ChatGPT15 or using the "Manual Configuration Method (Configure tab)." For the slave GPTs to acquire content such as book data and provide services based on that content (such as tests with questions), the content must be uploaded. Specifically, this is done from "Knowledge" under "Configure" in ChatGPT15. To connect to an external API, use the "Actions" function on the "Configure" screen.

[0058] 10A, a flowchart of the subroutine program for the upload process to the GPT Store 52 in S3 and the process for uploading to the GPT Store 52 in S12 will be described. The developer terminal 54K determines in S60 whether to upload to the GPT Store 52, and if not, returns and control proceeds to S4.

[0059] When the developer performs an upload operation to the GPT Store 52 using the developer terminal 54K, a YES determination is made in S60, and ChatGPT 15 is accessed in S61. The cloud server 51 of ChatGPT 15 receives this in S69, determines a YES determination in S69, and uploads the specified GPTs to the GPT Store 52 in S70. If slave GPTs are specified, the slave GPTs are uploaded, and if master GPTs are specified, the master GPTs are uploaded.

[0060] 10B, a flowchart of the main routine program between the user terminal 54 and the cloud server 51 will be described. The user terminal 54 performs master GPTs access processing in S75, and in response, the cloud server 51 performs master GPTs access handling processing in S80.

[0061] The user terminal 54 performs slave GPTs use processing in S76, and in response, the cloud server 51 performs slave GPTs use readiness processing in S81. The user terminal 54 performs other GPTs use processing in S77, and in response, the cloud server 51 performs other GPTs use readiness processing in S82.

[0062] A flowchart of the subroutine program for the master GPTs access processing in S75 and the master GPTs access response processing in S80 will be described with reference to Figure 11. The user terminal 54 determines whether to access the master GPTs 16 in S90, and if not, returns and control proceeds to S76. If the user performs an operation to access the master GPTs 16 using the user terminal 54, the determination in S90 is YES, and the user ID is transmitted in S91 to access ChatGPT 15. The cloud server 51 receives this in S99, determines YES in S99, and control proceeds to S100. The user terminal 54 requests ChatGPT 15 to call the master GPTs in S92, and upon receiving this, ChatGPT 15 calls the requested master GPTs 16 in S100.

[0063] In the cloud server 51, the called master GPTs 16 identifies each slave GPT used by the user, receives the review time via ChatGPT, and generates and transmits the user's review schedule (S101). Specifically, based on the user ID (see S91) transmitted from the user terminal 54, a database 19 of GPTs available to each user (see FIG. 3(B)) is searched to identify available GPTs. For example, if the user ID is "u3gf95," GPTsA, GPTsZ, and GPTsY are available GPTs. The next review schedule for GPTsA is March 18, 2024, the next review schedule for GPTsZ is April 1, 2024, and the next review schedule for GPTsY is March 22, 2024 (see FIG. 3(B)). A one-week review schedule for the user is created from these schedules.

[0064] For example, for user ID u3gf95, the sentences to be reviewed would be "pages 20-25 of book A on March 18, 2024, and pages 28-35 of book Y on March 22, 2024" (see Figure 3(B)). The sentences to be reviewed are determined from the "learning history" (see Figure 3(A)) in "each user's personalized data 26." The user terminal 54 communicates with the master GPTs 16 in S93 and views the generated review schedule. The user reviews the sections of the book that are due for review based on the review schedule. However, before reviewing, it is desirable for the user to take a test from the corresponding slave GPTs using questions from the sections to be reviewed. This is because taking the test on the sections to be reviewed before reviewing them allows the slave GPTs to consider the need to revise the user's personalized data (general model Ti = T0·i) based on the results of the test and, if necessary, make the necessary revisions. Specifically, if the user takes a test from the relevant slave GPTs with questions from the section to be reviewed before reviewing, the answers to the test are scored and the user's personalized data (including the general model Ti = T0·i) is modified according to the scoring results (see S124 to S126).

[0065] The flowchart of the subroutine program for the slave GPTs usage process in S76 and the slave GPTs usage adaptation process in S81 will be described with reference to Figure 12. In S110, the user terminal 54 determines whether to access the slave GPT of a book that has been newly studied or reviewed. If access is desired, the process proceeds to S111, where the user specifies the slave GPT to be accessed and notifies the user of the portion that has been studied (or reviewed).

[0066] The cloud server 51 receives the request in S120 and searches the database 19 (see FIG. 3(B)) in S121. If the specified slave GPTs are available to the user (see FIG. 3(B)), the cloud server 51 calls the specified slave GPT from among the available slave GPTs and notifies the user of the study (or review) points. The slave GPT that receives the notification updates the user's learning history (see FIG. 3(A)) in S122.

[0067] Next, the user terminal 54 determines in S112 whether to specify and access the slave GPT whose review time is approaching. If the user views the review schedule (see S93) and performs an operation to specify and access the slave GPT whose review time is approaching, the determination in S112 is YES, and processing to specify and access the slave GPT whose review time is approaching is performed in S113.

[0068] The cloud server 51 receives the request and calls the specified slave GPTs in S123. The called slave GPTs generate questions for the review portion in S124 and transmit the test to the user terminal 54. These questions are generated with reference to past exam data stored as data 27 common to all users. Questions that are the same as or similar to past exam questions may be generated.

[0069] The user terminal 54 receives the test from the user terminal 54, outputs the test for the review portion, and the user takes the test. The user's answers are sent to the slave GPTs. The slave GPTs grade the user's answers and return them in S125, and the user terminal 54 receives the graded results and displays them to the user (S115).

[0070] In step S126, if the scoring results are inconsistent with the user's personalized data, such as T0 or the general model Ti = T0·i, the slave GPTs corrects the user's personalized data (including the general model Ti = T0·i) according to the scoring results.

[0071] 13, the slave GPTs then calculates the next review time for the user based on the personalized data (including strengths and weaknesses) and the difficulty level of the unit in S140. In principle, the next review time is calculated based on the user's general model Ti = T0·i (the latest one if updated), but if the next unit to be reviewed is a difficult one or the unit or content is one that the user is not good at, the next review time is calculated slightly earlier.

[0072] In S131, ChatGPT 15 determines whether a request for a review schedule specifying slave GPTs has been received from master GPTs 16. When a user requests ChatGPT 15 on cloud server 51 to call master GPTs 16 (see S92), master GPTs 16 is called (see S100), and master GPTs 16 identifies each slave GPT used by the user, receives the review schedule via ChatGPT, and generates the user's review schedule (see S101). In response to the processing of S101, ChatGPT 15 determines in S131 that a request for a review schedule specifying slave GPTs has been received from master GPTs, and in S132, sends the user ID to the specified slave GPTs to request the next review time. In S141, the slave GPTs determine whether a request for the next review time has been received. If not, the process proceeds to S143. If received, the slave GPTs return the next review time for the received user ID in S142.

[0073] Upon receiving this, ChatGPT 15 transmits the next review time for the returned user ID to the master GPTs 16 in S133. This next review time is transmitted via ChatGPT 15 from all slave GPTs corresponding to the user ID. For example, if the user ID is u3gf95, the next review times from slave GPTs A, Z, and Y will be transmitted via ChatGPT 15. Upon receiving these next review times, the master GPTs 16 generates a one-week review schedule for the user ID (see FIG. 3(B)) and makes it viewable on the user terminal 54 (see S101 and S93).

[0074] Next, the slave GPTs updates the difficulty data for each unit based on the test scores of multiple users in step S143. For example, the slave GPTsZ is used by users with user IDs u3gf95 and pi7n52 (see Figure 3(B)), and the slave GPTs can update the difficulty data for each unit by comprehensively assessing the test scores of these multiple users.

[0075] Next, in S144, the slave GPTs compare the difficulty level of each unit with the scoring data of each user, calculate the strengths and weaknesses of each user, and update the personalized data.

[0076] Next, a modified example will be described with reference to Figures 14 to 17. This modified example follows the main embodiment described above, and the following mainly describes the differences from the main embodiment.

[0077] Figure 14 shows an English learning screen displayed on the user terminal 54. This service provision system handles foreign language learning content (English learning content) such as English news data provided by English newspaper sites or English news video and audio sites. These sites are linked to the slave GPTs via API, and the slave GPTs acquire the English news data and send it to the user terminal 54. Figure 14 shows the screen displayed on the user terminal 54 with the transmitted English news data.

[0078] A flowchart of the subroutine programs for the slave GPTs usage process and the slave GPTs usage response process in this modified example will be described with reference to Figure 15. In S150, the user terminal 54 determines whether to access the slave GPT for English news to be newly studied or reviewed. If the user performs an operation to access the slave GPT for English news, the determination in S150 is YES, and in S151, the slave GPT is designated and accessed, and the process of notifying the subject of study (or review) is performed.

[0079] Upon receiving this notification, the cloud server 51 determines YES in S160 and calls the specified slave GPTs in S161 to notify them of the study (or review) target. The user can use the called slave GPTs to either study or review using the English news data, or to take a review test. The user terminal 54 determines in S152 whether to take a review test. If the user performs an operation to study or review using the English news data on the user terminal 54, a learning process is performed by the user terminal in S153, and in response, the slave GPTs perform a learning process by themselves in S162.

[0080] A flowchart of the subroutine program for the learning process by the user terminal and the learning response process by the slave GPTs will be explained with reference to Figure 16. In S169, the user terminal 54 transmits a message to the slave GPTs of the cloud server 51 indicating that learning will be performed using English news data. In S175, the cloud server 51 determines whether the message to perform learning has been received. If the determination in S152 above is YES, step S153 is not executed, and therefore no transmission is performed in S169. Instead, a NO determination is made in S175, and control is returned, transferring control to S163.

[0081] On the other hand, if a message is received in S169 indicating that learning will be done using English news data, a YES determination is made in S175, and the notified data to be studied (or reviewed) is sent to the user terminal 54 in S176. Upon receiving this, the user terminal 54 outputs the received data to be studied (or reviewed) in S170. If the target content is data from an English newspaper site, the content (in English) is displayed on the screen of the display unit of the user terminal 54 (see FIG. 14). If the target content is data from an English news video or audio site, the video and audio are first output on the display unit and speaker of the user terminal 54, and then the audio is translated into English text and displayed on the screen (see FIG. 14).

[0082] Next, in S171, the user terminal 54 selects and specifies a word that the user does not understand when viewing the screen displayed in English, and displays the selected and specified English word in a distinguished manner (shown in bold in FIG. 14). The selected and specified English word is sent to the slave GPTs of the cloud server 51. The slave GPTs that receive it extract the Japanese meaning of the specified English word, etc., and return it in S177. Upon receiving the return, the user terminal 54 displays the Japanese meaning of the received English word, etc., in S172 (right side of the screen in FIG. 14), and then returns, and control transitions to S154.

[0083] In S178, the slave GPTs on the cloud server 51 determines whether the current study subject is a new study subject rather than a review. If it is not a new study subject, the process returns and control proceeds to S163. On the other hand, if it is a new study subject, the data on the new study subject is stored in S179, along with the English words and phrases specified in S171 and S177. These English words and phrases are English words and phrases that the user did not understand, and are stored to be presented to the user during the test described below.

[0084] Returning to FIG. 15 , if the user performs an operation to take a review test on the user terminal 54, a YES determination is made in S152, and control proceeds to S154, where communication with the slave GPTs is performed to take the test on the reviewed portion. The slave GPTs receive the communication indicating that the user will take the test on the reviewed portion, and in S163 generate questions for the reviewed portion and transmit the test to the user terminal 54. These questions are generated from the “English words and phrases the user did not understand” stored in S179 described above. Upon receiving the questions, the user terminal 54 outputs a test on the reviewed portion and has the user take the test in S154. This test also includes tests necessary for English conversation, such as checking and correcting the user's English pronunciation. The user's test answers are transmitted to the slave GPTs, which score the answers and return them in S164. The user terminal 54 receives the scoring results in S155 and displays them to the user.

[0085] In S165, the slave GPTs update the user's personalized data (T0, Ti = T0·i, strengths and weaknesses, etc.) based on the scoring results. Next, in S166, the slave GPTs add this learning result to the history and update the user's learning history.

[0086] Figure 17 shows a system in which master GPTs and slave GPTs are configured with their own AIs. When configured with a unique master AI 116 and a unique slave AI group 117, the two are linked via API 130, enabling direct information exchange between them without going through ChatGPT 15.

[0087] Further variations are listed below.

[0088] (1) One slave AI (slave GPTs) may acquire multiple pieces of content and provide users with services specialized for each piece of content. For example, referring to FIG. 2, if a publisher publishes multiple books M, L, and N, one slave AI (slave GPTs) may acquire the publisher's multiple books M, L, and N and provide users with services specialized for each of the books M, L, and N. In this case, the slave AI (slave GPTs) may independently have the aforementioned "review schedule notification function that notifies a user who has studied multiple pieces of content (books M, L, and N) of a review schedule indicating when to review each piece of content."

[0089] (2) Content in the healthcare field may be included. For example, electronic medical data and electronic medical history data from a smartphone, various measurement data from an Apple Watch (heart rate and irregular heart rhythm, atrial fibrillation history, decreased cardiopulmonary function level, oxygen level taken into the body, etc.), calorie intake data from a HEALBE smart watch, calorie burn and distance record data from a running app, and body fat percentage measurement history data from a body fat percentage measurement and management app.

[0090] Even if specialized trained models with processing functions optimized to match each of these content data are generated and services optimized for the target content are provided to users using each specialized trained model, a new drawback arises: comprehensive services, such as information provision across all of the multiple specialized trained models, cannot be provided. For example, atrial fibrillation history data from an Apple Watch reveals heart rate and irregular heart rhythm, but the cause cannot be identified from this content alone. However, by comparing the atrial fibrillation history data with body fat measurement history data from a body fat measurement and management app, it is possible to determine that the heart rate and irregular heart rhythm are caused by obesity. Furthermore, by comparing calorie intake data from the HEALBE smartwatch and calorie expenditure data from a running app, it is possible to determine that obesity is caused by excessive or insufficient calorie intake.

[0091] (3) GPTs uploaded to the GPT Store may be charged a fee, allowing users to monetize their usage. It is also possible to charge a fee for only certain features of the GPTs, rather than the entire GPTs. This monetization may be achieved through an advertising model that displays advertisements related to the content acquired by the GPTs. For example, in the case of study content for test takers (such as textbooks), advertisements for preparatory schools and cram schools, online seminars for test takers, etc. may be displayed, and advertising fees may be paid by those advertisers.

[0092] (4) The vendors that handle the content acquired by each slave AI (slave GPTs) may create corresponding slave AIs (slave GPTs) that can exchange information with (link to) the master AI (master GPTs). For example, a slave AI (slave GPTs) that acquires book A may be created by the publisher of book A (or the author of book A).

[0093] By doing so, the creator of the slave AI (slave GPTs) has the advantage of promoting sales of the content (book A, etc.) by creating a slave AI (slave GPTs) that is optimal for the content (book A, etc.) to be read. In other words, content providers who handle the content acquired by each slave AI (slave GPTs) have an incentive to create optimal slave AI (slave GPTs). By having such providers create excellent slave AI (slave GPTs) one after another and linking them with the master AI (master GPTs), the service provision system can be turned into a platform. By turning the system into a platform, fees can be collected from the content providers.

[0094] (5) The aforementioned “review schedule notification function that notifies users who have studied multiple contents of a review schedule indicating when to review each content” may be linked with general schedule management software, so that both schedules can be displayed together. The above-described embodiments disclose the following inventions.

[0095] [Technical field] The present invention relates to a service providing system that uses artificial intelligence to provide comprehensive services to users who use multiple contents.

[0096] [Background technology] A technology has been proposed in which a trained model based on machine learning of artificial intelligence is trained to learn the content of an external site, and the trained model then executes processing based on that content (for example, Patent Document 1).

[0097] [Prior art document]

[0098] [Patent Documents]

[0099] [Patent Document 1] Patent No. 7446023

[0100] [Summary of the Invention]

[0101] [Problem to be solved by the invention] One idea to improve on this technology is to construct a trained model using a machine learning model that has processing capabilities specialized for the type of content that is trained or loaded into the trained model.

[0102] For example, if a study guide for test takers is trained or loaded into a trained model as content, it is desirable for the trained model (artificial intelligence) to have processing functions specialized for study guide content, such as generating questions based on the study guide and testing them with the user (test taker).

[0103] Furthermore, for example, in the case of content for English conversation lessons, it is desirable for the model to be a trained model (artificial intelligence) with processing functions specialized for English conversation lesson content, such as picking out new words and phrases that appear in each unit and presenting them to the user, or checking and correcting the user's English pronunciation.

[0104] In light of this situation, one possible solution is to create specialized trained models with processing functions optimally designed to match each piece of content, and to build an artificial intelligence system that uses each specialized trained model to provide users with services optimally designed for the target content.

[0105] However, in an artificial intelligence system constructed in this way, the following new problems arise:

[0106] The most efficient way to solidify learned material in memory is to repeat 70% interval review, in which the material is reviewed just before it is about to be forgotten (for example, when 70% of the material is still in memory on the forgetting curve). For example, if a user is studying using three specialized trained models—one for content A, one for content B, and one for content C—the next review times for each piece of content will be misaligned, making it difficult for the user to grasp all of the review times. This problem becomes more serious as the number of content items to be studied increases. For example, in the case of the Common University Entrance Examination, which covers 30 subjects across six subjects, it becomes extremely difficult to accurately grasp the review times for all of the subjects.

[0107] In other words, while artificial intelligence using specialized trained models has the advantage of being able to provide users with services that are optimally designed for the target content, it also has a new disadvantage in that if a user uses multiple specialized trained models, it is unable to provide comprehensive services such as information that encompasses all of those multiple specialized trained models.

[0108] These issues arise not only in the field of learning but also in other fields (such as healthcare). Some possible examples are described below.

[0109] Examples of healthcare content include electronic medical data and electronic medical history data from Selecon, various measurement data from Apple Watch (heart rate and irregular heart rhythm, atrial fibrillation history, decreased cardiopulmonary function level, oxygen level taken into the body, etc.), calorie intake data from the HEALBE smartwatch, recorded data on calories burned and distance from a running app, and body fat measurement history data from a body fat measurement and management app.

[0110] Even if specialized trained models with processing functions optimized to match each of these content data are generated and services optimized for the target content are provided to users using each specialized trained model, a new drawback arises: comprehensive services, such as information provision across all of the multiple specialized trained models, cannot be provided. For example, atrial fibrillation history data from an Apple Watch reveals heart rate and irregular heart rhythm, but the cause cannot be identified from this content alone. However, by comparing the atrial fibrillation history data with body fat measurement history data from a body fat measurement and management app, it is possible to determine that the heart rate and irregular heart rhythm are caused by obesity. Furthermore, by comparing calorie intake data from the HEALBE smartwatch and calorie expenditure data from a running app, it is possible to determine that obesity is caused by excessive or insufficient calorie intake.

[0111] The present invention was devised in light of this situation, and its purpose is to provide users with services using the artificial intelligence of each specialized trained model, while also enabling the provision of comprehensive services such as information provision that encompasses the entire range of artificial intelligence of multiple specialized trained models.

[0112] [Means for solving the problem] Below, in the description of the means for solving the problems, relevant parts of the embodiments and drawings are inserted in parentheses.

[0113] One aspect of the present invention is a service providing system that provides a comprehensive service to a user who uses a plurality of contents using artificial intelligence, the system comprising: A plurality of slave artificial intelligences (e.g., a group of slave GPTs 17) that can provide users with services specialized for the content (e.g., generating questions from the review section of a book and conducting a test) based on the acquired content data (e.g., books A, B, Z, etc.), and A master AI (e.g., master AI 116, master GPTs 16, etc.) that exchanges information with the plurality of slave AIs, The plurality of slave AIs are available for use by a user on the condition that the user has selected them as targets for use (for example, available GPTs stored in association with the user ID shown in FIG. 3(B)), The master artificial intelligence is A first function (e.g., S121) that selects a slave AI desired by the accessing user from among the slave AIs available to the user and allows the user to use the slave AI; and a second function (e.g., S93, S101, etc.) that provides the accessing user with a comprehensive service covering all of the slave artificial intelligences that can be used by the user based on information exchanged with the plurality of slave artificial intelligences (e.g., a review schedule consisting of review periods for each of the books imported by the plurality of slave GPTs).

[0114] With this configuration, each slave artificial intelligence can provide a user with a service that is specialized for the content, while also providing the user with a service that encompasses all of the multiple slave artificial intelligences that the accessing user can use.

[0115] Another aspect of the present invention is a service providing system that includes a server (e.g., cloud server 51) that provides a service using artificial intelligence, and a terminal (e.g., user terminal 54) that can connect to the server via the Internet, and provides a comprehensive service to a user who uses a plurality of contents using artificial intelligence, A plurality of slave artificial intelligences (e.g., a group of slave GPTs 17) that can provide users with services specialized for the content (e.g., generating questions from the review section of a book and conducting a test) based on the acquired content data (e.g., books A, B, Z, etc.), and A master AI (e.g., master AI 116, master GPTs 16, etc.) that exchanges information with the plurality of slave AIs, The plurality of slave AIs are available for use by a user on the condition that the user has selected them as targets for use (for example, available GPTs stored in association with the user ID shown in FIG. 3(B)), The master artificial intelligence is A first function (e.g., S121) that selects a slave AI desired by the accessing user from among the slave AIs available to the user and allows the user to use the slave AI; and a second function (e.g., S93, S101, etc.) that provides the accessing user with a comprehensive service covering all of the slave artificial intelligences that can be used by the user based on information exchanged with the plurality of slave artificial intelligences (e.g., a review schedule consisting of review periods for each of the books imported by the plurality of slave GPTs).

[0116] With this configuration, it is possible to provide the user with a service that is specialized for the content using each slave artificial intelligence, while also providing the user with a service that encompasses all of the multiple slave artificial intelligences that the user can use.

[0117] Preferably, the plurality of contents include study contents used by the user for studying (e.g., books A, B, Z, foreign language study contents, etc.), The second function includes a review schedule notification function (for example, S101, etc.) that notifies the user who has studied the plurality of contents of a review schedule indicating when to review each content.

[0118] With this configuration, a user who has studied a plurality of contents does not have to manage review times by himself.

[0119] More preferably, the plurality of slave artificial intelligences provide, as a service specialized in the content, A test generation means (e.g., S124, S) for generating a test consisting of questions based on the acquired learning content; and an answer marking means (e.g., S125) capable of providing a marking service to a user for marking the user's answers to the test generated by the test generating means, The review schedule notification function notifies the user of a review schedule indicating the optimal review time personalized for the user, calculated according to the user's unique forgetting curve derived from the grading results of the user's answers (for example, Ti=T0·i in Figure 3(A), Figures 5 and 6, S126, S131 to S133, S140 to S142, etc.).

[0120] With this configuration, the user can review based on a review schedule that is most suitable for him or her.

[0121] More preferably, the slave artificial intelligence: A weak point determination means (e.g., S143, S144, etc.) for determining weak points in the learning of individual users; and weak portion review timing adjustment means (for example, S140, etc.) for adjusting the timing for reviewing the portions determined to be weak by the weak portion determination means to be earlier.

[0122] With this configuration, weak points tend to be forgotten earlier than usual, but by speeding up the review period, it is possible to deal with early forgetting of weak points.

[0123] More preferably, the weak point determination means determines the weak points of an individual user by comparing the scoring results by the answer scoring means for multiple users studying the same learning content with the scoring results by the answer scoring means for the individual user (e.g., S143, S144, etc.).

[0124] More preferably, the slave artificial intelligence: A difficulty level determination means (e.g., S143) that determines the difficulty level of each learning object based on the scoring results of the answer scoring means for multiple users studying the same learning content; and a difficulty level-based review timing adjustment means (for example, S143, etc.) that adjusts the review timing for the study subject based on the determination result of the difficulty level determination means.

[0125] More preferably, the slave artificial intelligence: A test question acquisition means capable of acquiring data on test questions given to test takers from the learning content (for example, "data common to all users (past questions, difficulty level for each unit, etc.) 27" in FIG. 2, etc.); and test generation means (for example, S124) that generates a test consisting of questions based on the learning content, using the test questions acquired by the test question acquisition means as a reference.

[0126] With this configuration, users can take high-quality tests that are generated with reference to test questions given to examinees.

[0127] Another aspect of the present invention is a learning support service providing system for providing a service to support foreign language learning, comprising: A content acquisition means for acquiring content for foreign language learning (e.g., English news data provided by English newspaper sites or English news video and audio sites) (e.g., by linking these sites with slave GPTs via API, the slave GPTs can acquire English news data); An output means (e.g., S176) for outputting the acquired content to a user terminal; a notification means (e.g., S177) capable of notifying the user of the meaning of a portion that the user does not understand in the content output by the output means when the user selects and specifies the portion (e.g., S171) in the content that the user does not understand in the content output by the output means in the user's native language; A review time notification means (for example) for calculating a review time for the user to review the part that the user has studied and notifying the user of the time; a test setting means (for example) for generating a test consisting of questions based on the selected and designated portion when the user reviews the material, and giving the test to the user; and a scoring notification means (e.g., S93, S101, etc.) for scoring the user's answers to the test set by the test setting means and notifying the user of the scores, The review time notification means has a review schedule notification function (for example, S93, S101, etc.) that notifies the user of a review schedule indicating the review time for each content when the user is studying a plurality of content items.

[0128] With this configuration, when the user later reviews the parts he or she has selected and specified in the content he or she is studying and does not understand, he or she can take a test consisting of questions based on the parts he or she does not understand, and the user can review only the parts he or she does not understand.

[0129] Preferably, the review timing notification means calculates and notifies the user of an optimal review timing personalized for the user, calculated according to the user's unique forgetting curve derived from the scoring results of the user's answers to the test (for example, Ti=T0·i in FIG. 3(A), FIGS. 5 and 6, S126, S131 to S133, S140 to S142, etc.).

[0130] The embodiments disclosed herein should be considered to be illustrative in all respects and not restrictive. The scope of the present invention is defined by the claims, not by the above description, and is intended to include all modifications within the meaning and scope of the claims. [Explanation of symbols]

[0131] 15 ChatGPT, 16 Master GPTs, 17 Slave GPTs, 18 Publishers, 43 Cloud, 44 Data Center, 51 Cloud Computer, 52 GPT Store, 54 User Terminal, 54K Developer Terminal, 60 GPTsDB, 130 API.

Claims

1. A service providing system that provides comprehensive services to users who use multiple contents using artificial intelligence, a plurality of slave artificial intelligences capable of providing users with services specialized for the content based on the acquired content data; a master artificial intelligence that exchanges information with the plurality of slave artificial intelligences; The plurality of slave artificial intelligences are available for use by a user on the condition that the user has selected them as targets for use, The master artificial intelligence is A first function of selecting a slave AI desired by the accessing user from among the slave AIs available to the user and permitting its use; A service providing system having a second function of providing a service to an accessing user that encompasses all of the multiple slave artificial intelligences that the accessing user can use, based on information exchanged with the multiple slave artificial intelligences.

2. A service providing system that provides comprehensive services to users who use a plurality of contents using artificial intelligence, the system comprising: a server that provides services using artificial intelligence; and a terminal that can be connected to the server via the Internet. a plurality of slave artificial intelligences capable of providing users with services specialized for the content based on the acquired content data; a master artificial intelligence that exchanges information with the plurality of slave artificial intelligences; The plurality of slave artificial intelligences are available for use by a user on the condition that the user has selected them as targets for use, The master artificial intelligence is A first function of selecting a slave AI desired by the accessing user from among the slave AIs available to the user and permitting its use; A service providing system having a second function of providing a service to an accessing user that encompasses all of the multiple slave artificial intelligences that the accessing user can use, based on information exchanged with the multiple slave artificial intelligences.

Images