System for providing generation AI services, and method for providing generation AI services.

Abstract

To provide a generative AI service delivery system that can flexibly determine available content in response to queries. [Solution] This is a system for providing an AI service, comprising: a query acquisition unit that acquires queries which are input to an AI (Artificial Intelligence); a location information acquisition unit that acquires location information of additional data related to the query; an additional data acquisition unit that acquires additional data based on the location information acquired by the location information acquisition unit; a user authentication unit that authenticates the user according to the location information acquired by the location information acquisition unit, and if authentication is successful, provides the additional data to the additional data acquisition unit; and a generation unit that uses the additional data acquired by the additional data acquisition unit to acquire a response from the AI ​​to the query.

Description

Technical Field

[0001] The present invention relates to a generative AI service providing system and a generative AI service providing method.

Background Art

[0002] In recent years, services such as ChatGPT (registered trademark) and GEMINI (registered trademark) have been provided, which provide responses generated using large language models for queries such as questions entered by users. In addition, services such as Stable Diffusion and Midjourney (registered trademark), which generate images instead of text as responses, and services that generate audio are also provided.

[0003] Also, when using a large language model, as a technology for obtaining responses based on documents for a specific organization or content such as images and audio that require permission from rights holders such as copyright holders for use, technologies such as fine-tuning, transfer learning, and RAG (Retrieval-Augmented Generation) using those documents and content are also known. Patent Document 1 discloses an information processing system that outputs an answer sentence using a large language model for a question sentence input from a user. This information processing system searches for documents with a high degree of relevance to the question from among a plurality of documents each attached with a content classification item, and inputs a prompt including the retrieved document and the question sentence into the large language model.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] However, the information processing system described in Patent Document 1 has the problem that it cannot flexibly determine the available content in response to queries such as question sentences.

[0006] This invention has been made in view of these circumstances, and provides a generation AI service provision system and a generation AI service provision method that can flexibly determine the available content in response to queries. [Means for solving the problem]

[0007] This invention was made to solve the above-mentioned problems, and one aspect of the present invention is a generation AI service provision system comprising: a query acquisition unit that acquires a query which is an input to a generation AI (Artificial Intelligence); a location information acquisition unit that acquires location information of additional data related to the query; an additional data acquisition unit that acquires the additional data based on the location information acquired by the location information acquisition unit; a user authentication unit that authenticates a user, which performs authentication according to the location information acquired by the location information acquisition unit, and if authentication is successful, provides the additional data to the additional data acquisition unit; and a generation unit that uses the additional data acquired by the additional data acquisition unit to acquire a response from the generation AI to the query.

[0008] Another aspect of the present invention is the above-described generation AI service provision system, wherein the additional data is external information for RAG (Retrieval-Augmented Generation), and the generation unit obtains the response by inputting the query and the additional data to the generation AI.

[0009] Another aspect of the present invention is the above-described generation AI service provision system, wherein the additional data is the difference in model parameters obtained by fine tuning or transfer learning, and the generation unit applies the additional data to the model parameters of the generation AI to obtain the response.

[0010] Another aspect of the present invention is the above-described generation AI service provision system, comprising an additional data location storage unit that stores a plurality of sets of location information and information related to the additional data, wherein the location information acquisition unit searches for information related to the additional data that is related to the query and acquires the location information associated with the information related to the additional data, including the search results.

[0011] Another aspect of the present invention is a method for providing an artificial intelligence (AI) service, comprising: a first step of obtaining a query which is input to an artificial intelligence (AI); a second step of obtaining location information of additional data related to the query; a third step of requesting the additional data based on the location information obtained in the second step; a fourth step of user authentication, which is performed according to the location information obtained in the second step, and if authentication is successful, provides the additional data as a response to the request in the third step; and a fifth step of obtaining a response from the artificial intelligence to the query using the additional data provided in the fourth step. [Effects of the Invention]

[0012] According to this invention, the generation AI service provision system and generation AI service provision method can flexibly determine the available content in response to queries. [Brief explanation of the drawing]

[0013] [Figure 1] This is a schematic block diagram showing the configuration of a generation AI service provision system 100 according to one embodiment of the present invention. [Figure 2] This table shows an example of the contents stored in the additional data location storage unit 13 in the same embodiment (Part 1). [Figure 3] This table shows an example of the contents stored in the additional data location storage unit 13 in the same embodiment (part 2). [Figure 4]This is a flowchart illustrating the operation of the generation AI service provision system 100 in the same embodiment. [Figure 5] This is a flowchart illustrating an example of the operation of the generation unit 15 in the same embodiment (part 1). [Figure 6] This is a flowchart illustrating an example of the operation of the generation unit 15 in the same embodiment (part 2). [Modes for carrying out the invention]

[0014] Embodiments of the present invention will be described below with reference to the drawings. Figure 1 is a schematic block diagram showing the configuration of a generation AI service provision system 100 according to one embodiment of the present invention. The generation AI service provision system 100 generates and outputs a response R1 in response to a query Q1, which is a request input by the user, using generation AI (Artificial Intelligence). When generating this response R1, the generation AI service provision system 100 uses authentication information C1 input by the user to determine whether or not to use additional data to be used when generating the response R1.

[0015] The generation AI service provider system 100 may generate and output a response R1 to a query Q1, which is a question. Alternatively, the generation AI service provider system 100 may generate and output content as response R1 in response to a query Q1, which is a request for content such as text, images, videos, audio, or music.

[0016] The user may input query Q1 and authentication information C1 directly to the generation AI service provision system 100, or via a personal computer, tablet, smartphone, etc. The response R1 may also be output by the generation AI service provision system 100, or via a personal computer, tablet, smartphone, etc.

[0017] The AI service providing system 100 includes a service providing device 10 and one or more additional data providing devices 20. Each of the service providing device 10 and the additional data providing device 20 may be realized by one or more computers reading and executing a program. Also, each of the service providing device 10 and the additional data providing device 20 may be implemented on the cloud. The service providing device 10 and the additional data providing device 20 are communicably connected by a network such as the Internet.

[0018] The service providing device 10 generates and outputs a response R1 corresponding to the query Q1 input by the user, and uses the additional data provided from the additional data providing device 20 when generating the response R1. The additional data providing device 20 stores the additional data and provides the additional data only when the user authentication is successful when requested for the additional data from the service providing device 10. When the AI service providing system 100 includes a plurality of additional data providing devices 20, the additional data available to a certain user is only the additional data stored in the additional data providing device 20 where the user successfully authenticates. For example, the first and second additional data providing devices 20 store additional data corresponding to the contents of Company AA and Company BB respectively, and assume that the user has only contracted with Company AA, so the user authentication is successful only with the first additional data providing device 20. In this case, for the query of the user, the AI service providing system 100 generates a response using the additional data corresponding to the contents of Company AA.

[0019] The service providing device 10 includes a query acquisition unit 11, a search unit 12, an additional data location storage unit 13, an additional data acquisition unit 14, a generation unit 15, and a learned model storage unit 16. The additional data providing device 20 includes a user authentication unit 21, an additional data storage unit 22, and an authentication information storage unit 23.

[0020] The query acquisition unit 11 acquires a query Q1 which is an input to the generative AI (large language model). The search unit 12 (position information acquisition unit) acquires the position information of additional data related to the query Q1 acquired by the query acquisition unit 11. The additional data may be external information for RAG (Retrieval-Augmented Generation), or may be differential data (so-called ΔW) of model parameters by fine-tuning such as LoRA (Low-Rank Adaptation) or transfer learning. The model parameters are parameters (so-called W) that constitute the large language model, such as weights in the neural network that constitutes the large language model.

[0021] In the form where the additional data is external information for RAG, the generative AI service providing system 100 may input a prompt including the query Q1 and the external information or a part thereof to the large language model to generate a response R1. Also, in the form where the additional data is differential data of model parameters, the generative AI service providing system 100 may generate a response R1 using the large language model with model parameters reflecting the differential data. Note that the model parameters reflecting the differential data are model parameters obtained by adding the values of the differential data to the model parameters stored in the learned model storage unit 16.

[0022] The position information of the additional data is information indicating the position of the additional data, such as, for example, a URL (Universal Resource Locator) for acquiring the additional data, the address of the additional data providing device 20 that provides the additional data, etc. The additional data position storage unit 13 stores a plurality of sets in which the position information of the additional data is associated with information related to the additional data. The search unit 12 may search the information related to the additional data for information related to the query Q1 and acquire the position information associated with the information related to the additional data including the search result.

[0023] The information related to the additional data stored in the additional data location storage unit 13 is in vector format, and the search unit 12 may convert query Q1 into vector format and search for information about additional data in vector format that is similar to the vector formatted query Q1. The information related to the additional data may be, for example, text describing the content that becomes the additional data, fine tuning, and the content used for transfer learning.

[0024] The additional data acquisition unit 14 acquires additional data based on the location information acquired by the search unit 12. For example, the additional data acquisition unit 14 uses the location information acquired by the search unit 12 to send a request for additional data to the additional data providing device 20 and acquires additional data in response to that request. The additional data acquisition unit 14 may also include query Q1 or a vectorized version of query Q1 in the request for additional data and request additional data related to query Q1.

[0025] The generation unit 15 uses the additional data acquired by the additional data acquisition unit 14 to obtain the generation AI's response to query Q1. If the additional data acquisition unit 14 acquires additional data from multiple additional data providing devices 20, the generation unit 15 may use the additional data acquired from the multiple additional data providing devices 20 to obtain the generation AI's response to query Q1. The trained model storage unit 16 stores the model parameters that constitute the large-scale language model (trained model) for the generation AI used by the generation unit 15.

[0026] In the case where the additional data is external information for RAG, the generation unit 15 inputs a prompt containing query Q1 and the external information or a part thereof to a large-scale language model using model parameters read from the trained model storage unit 16, and generates a response R1. The part of the external information included in the prompt may be the portion of the external information that is related to query Q1.

[0027] Furthermore, in the case where the additional data is in the form of difference data of model parameters, the generation unit 15 generates the response R1 using a large-scale language model of model parameters that reflects the difference data in the model parameters read from the trained model storage unit 16.

[0028] The additional data may include external information for RAG and difference data of model parameters. In that case, the generation unit 15 generates a response R1 using a large-scale language model of model parameters that reflects the difference data in the model parameters read from the trained model storage unit 16, based on the query Q1 and a prompt including the external information or a part thereof.

[0029] When the user authentication unit 21 receives a request from the additional data acquisition unit 14, it authenticates the user, and if authentication is successful, it provides the additional data stored in the additional data storage unit 22 to the additional data acquisition unit 14. This user authentication is performed by the user authentication unit 21 according to the location information acquired by the search unit 12, and is therefore authentication according to said location information. For example, the authentication may be performed using the authentication information stored in the authentication information storage unit 23 of the additional data provision device 20, thereby making it authentication according to the location information acquired by the search unit 12.

[0030] If authentication fails, the user authentication unit 21 may provide the additional data acquisition unit 14 with additional data for unauthenticated users stored in the additional data storage unit 22. This may include external information in text format with confidential information redacted, or differential data of model parameters used for image generation with limited quality, such as very low quality. This allows the user to recognize that they lack access rights to the information simply by looking at response R1. Furthermore, since the generation unit 15 does not need to perform exception handling due to the lack of necessary additional data, it can be expected that the generation unit 15 will generate response R1 stably.

[0031] For example, if authentication information C1 is the user's ID and password, the user authentication unit 21 may determine that authentication is successful when it compares the authentication information C1 entered by the user with the authentication information stored in the authentication information storage unit 23 and they match. In other words, the authentication information storage unit 23 may store the authentication information of users who can use the additional data stored in the additional data storage unit 22 of the additional data providing device 20 equipped with the authentication information storage unit 23.

[0032] Furthermore, the user authentication unit 21 may inquire with the user about purchasing the right to use the additional data stored in the additional data storage unit 22 of the additional data provision device 20 equipped with the user authentication unit 21, and determine that authentication has been successful if the right to use the data has been purchased. In this case, the authentication information C1 may be information for purchasing the right to use the data, or information indicating that the right to use the data has been purchased. This allows the user to select the additional data to use each time query Q1 is entered.

[0033] Figure 2 is a table showing an example of the contents of the additional data location storage unit 13 in this embodiment (part 1). Figure 2 is a table showing an example of the contents of the additional data in a form in which the additional data is external information for RAG. The table in Figure 2 includes a set that associates information related to the additional data, such as "This is a design drawing of product A1. Product A1 uses parts A2 and A3...", with the location information of the additional data, such as "https: / / ...", and a set that associates information related to the additional data, such as "This is the business manual for department B1. The customers of department B1 are companies B2 and B3...", with the location information of the additional data, such as "https: / / ...". The information related to the additional data may include information that indicates the contents of the document of the external information, such as "This is a design drawing of product A1," or it may include information about the subject of the description, such as "Product A1 uses parts A2 and A3."

[0034] Figure 3 is a table showing an example of the stored contents (part 2) of the additional data location storage unit 13 in this embodiment. Figure 3 is a table showing an example of stored contents in a form where the additional data is difference data of model parameters. The table in Figure 3 includes a set that associates information related to the additional data, such as "Images of characters X3 and X4 from game X2 by company X1. Character X3 is a humanoid robot. Game X2 is...", with the location information of the additional data, "https: / / ...", and a set that associates information related to the additional data, such as "Images of character Y3 and items belonging to character Y3 from manga Y2 by company Y1. Items belonging to character Y3 are...", with the location information of the additional data, "https: / / ...". The information related to the additional data may include information indicating the data (content) used for training (fine-tuning, transfer learning, etc.) to generate the difference data, such as "These are images of characters X3 and X4 from game X2 by company X1," or it may include information describing the attributes of the subject represented by the data (content) used for training, such as "Character X3 is a humanoid robot. Game X2 is..."

[0035] Figure 4 is a flowchart illustrating the operation of the generation AI service provision system 100 in this embodiment. First, the query acquisition unit 11 acquires the query Q1 entered by the user (step Sa1). Next, the search unit 12 acquires location information of additional data corresponding to the query Q1 (step Sa2). Next, the generation AI service provision system 100 sequentially selects the locations indicated by the location information acquired in step Sa2, and performs the processing between steps Sa3 and Sa9 (steps Sa4 to Sa8) for each location (steps Sa3, Sa9).

[0036] In step Sa4, the additional data acquisition unit 14 sends a request for additional data to the additional data providing device 20 at the selected location. Next, in step Sa5, the user authentication unit 21 of the additional data providing device 20, which received the request in step Sa4, receives the request. Next, in step Sa6, the user authentication unit 21 authenticates the user. If authentication is successful (step Sa6-Yes), the process proceeds to step Sa7. If authentication is unsuccessful (step Sa6-No), the process proceeds to step Sa8. In step Sa7, the user authentication unit 21 reads additional data from the additional data storage unit 22 and provides the additional data to the additional data acquisition unit 14. In step Sa8, the user authentication unit 21 reads additional data for unauthenticated users from the additional data storage unit 22 and provides the additional data for unauthenticated users to the additional data acquisition unit 14. Note that in step Sa8, the user authentication unit 21 may choose not to provide additional data for unauthenticated users to the additional data acquisition unit 14. In other words, in step Sa8, the user authentication unit 21 may choose not to provide anything to the additional data acquisition unit 14.

[0037] After performing the processing between steps Sa3 and Sa9 (steps Sa4 to Sa8) for all locations indicated by the location information obtained in step Sa2, the generation unit 15 generates a response to the query using the additional data obtained by the additional data acquisition unit 14 (step Sa10).

[0038] Figure 5 is a flowchart illustrating an example of the operation of the generation unit 15 in this embodiment (part 1). Figure 5 shows an example of the operation of the generation unit 15 in step Sa9 of Figure 2, and is an example of operation in a form where the additional data is external information for RAG. The generation unit 15 obtains query Q1 from the query acquisition unit 11 (step Sb1). Next, the generation unit 15 obtains additional data from the additional data acquisition unit 14 (step Sb2). Next, the generation unit 15 obtains the model parameters of the trained model from the trained model storage unit 16 (step Sb3).

[0039] Next, the generation unit 15 generates a prompt including query Q1 and additional data (step Sb4). Next, the generation unit 15 inputs the prompt generated in step Sb4 to a large-scale language model using the model parameters obtained in step Sb3 (step Sb5). Next, the generation unit 15 obtains the response generated by the large-scale language model based on the input in step Sb5 (step Sb6).

[0040] Figure 6 is a flowchart illustrating an example of the operation of the generation unit 15 in this embodiment (part 2). Figure 5 is an example of the operation of the generation unit 15 in step Sa9 of Figure 2, and is an example of operation in a form where the additional data is difference data of the model parameters. The generation unit 15 obtains query Q1 from the query acquisition unit 11 (step Sc1). Next, the generation unit 15 obtains additional data from the additional data acquisition unit 14 (step Sc2). Next, the generation unit 15 obtains the model parameters of the trained model from the trained model storage unit 16 (step Sc3).

[0041] Next, the generation unit 15 reflects the additional data obtained in step Sc2 into the model parameters obtained in step Sc3. Next, the generation unit 15 inputs query Q1 as a prompt to the large-scale language model using the model parameters that now reflect the additional data (step Sc5). Next, the generation unit 15 obtains the response generated by the large-scale language model based on the input in step Sc5 (step Sc6).

[0042] As described above, the generation AI service provision system 100 in this embodiment comprises a query acquisition unit 11, a search unit 12, an additional data acquisition unit 14, a user authentication unit 21, and a generation unit 15. The query acquisition unit 11 acquires a query Q1, which is the input to the generation AI. The search unit 12 acquires location information of additional data related to the query Q1. The user authentication unit 21 acquires additional data based on the location information acquired by the search unit 12. The user authentication unit 21 authenticates the user, performing authentication according to the location information acquired by the search unit 12, and if authentication is successful, provides the additional data to the additional data acquisition unit 14. The generation unit 15 uses the additional data acquired by the additional data acquisition unit 14 to obtain the generation AI's response to the query Q1. As a result, the generation AI service provision system 100 can flexibly determine the available content according to the query Q1.

[0043] In the above embodiment, the large-scale language model and the trained model storage unit 16 were described as being provided by the generation unit 15 of the service provision device 10, but they may be provided by other devices. The service provision device 10 and the additional data provision device 20 may each be composed of multiple devices.

[0044] Furthermore, the present invention may also be provided in the following embodiments. (1) One embodiment of the present invention is a system for providing an artificial intelligence (AI) service, comprising: a query acquisition unit that acquires a query which is an input to an artificial intelligence (AI); a location information acquisition unit that acquires location information of additional data related to the query; an additional data acquisition unit that acquires the additional data based on the location information acquired by the location information acquisition unit; a user authentication unit that authenticates a user according to the location information acquired by the location information acquisition unit, and if authentication is successful, provides the additional data to the additional data acquisition unit; and a generation unit that uses the additional data acquired by the additional data acquisition unit to acquire a response from the artificial intelligence to the query.

[0045] (2) Another embodiment of the present invention is a generation AI service provision system as described in (1), wherein the additional data is external information for RAG (Retrieval-Augmented Generation), and the generation unit obtains the response by inputting the query and the additional data to the generation AI.

[0046] (3) Another embodiment of the present invention is a generation AI service provision system as described in (1), wherein the additional data is the difference in model parameters obtained by fine tuning or transfer learning, and the generation unit applies the additional data to the model parameters of the generation AI to obtain the response.

[0047] (4) Another embodiment of the present invention is a generation AI service provision system as described in any of (1) to (3), comprising an additional data location storage unit that stores a plurality of sets of location information and information related to the additional data, wherein the location information acquisition unit searches for information related to the additional data that is related to the query and acquires the location information associated with the information related to the additional data, including the search results.

[0048] (5) Another embodiment of the present invention is a method for providing an artificial intelligence (AI) service, comprising: a first step of obtaining a query which is input to an artificial intelligence (AI); a second step of obtaining location information of additional data related to the query; a third step of requesting the additional data based on the location information obtained in the second step; a fourth step of user authentication, which is performed according to the location information obtained in the second step, and if authentication is successful, provides the additional data as a response to the request in the third step; and a fifth step of obtaining a response from the artificial intelligence to the query using the additional data provided in the fourth step.

[0049] Alternatively, the functions of the query acquisition unit 11, search unit 12, additional data location storage unit 13, additional data acquisition unit 14, generation unit 15, trained model storage unit 16, user authentication unit 21, additional data storage unit 22, and authentication information storage unit 23 in Figure 1 may be realized by recording a program for realizing these functions on a computer-readable recording medium, loading the program recorded on this recording medium into a computer system, and executing it. The term "computer system" here includes hardware such as the operating system and peripheral devices.

[0050] Furthermore, "computer system" shall also include the homepage provisioning environment (or display environment) if a WWW system is being used. Furthermore, "computer-readable recording media" refers to portable media such as flexible disks, magneto-optical disks, ROMs, and CD-ROMs, as well as storage devices such as hard disks built into computer systems. Moreover, "computer-readable recording media" also includes those that dynamically hold programs for a short period of time, such as communication lines used when transmitting programs over networks such as the Internet or communication lines such as telephone lines, and those that hold programs for a certain period of time, such as volatile memory inside computer systems that act as servers or clients in such cases. In addition, the above-mentioned programs may be for the purpose of realizing some of the functions described above, and may also be able to realize the above-mentioned functions in combination with programs already recorded in the computer system.

[0051] While embodiments of this invention have been described in detail above with reference to the drawings, the specific configuration is not limited to these embodiments and may include design modifications and the like that do not depart from the spirit of this invention. [Explanation of Symbols]

[0052] 100 Generation AI Service Provision System 10 Service provision device 11 Query Retrieval Section 12 Search Section 13 Additional data location storage unit 14 Additional data acquisition unit 15 Generation part 16. Pre-trained model memory 20 Additional data provision device 21 User Authentication Section 22 Additional data storage unit 23 Authentication Information Storage Unit

Claims

1. A query acquisition unit that acquires queries, which are input to the generated AI (Artificial Intelligence), A location information acquisition unit that acquires location information of additional data related to the aforementioned query, An additional data acquisition unit acquires the additional data based on the location information acquired by the location information acquisition unit, A user authentication unit that authenticates a user based on the location information acquired by the location information acquisition unit, and if authentication is successful, provides the additional data to the additional data acquisition unit. The generation unit uses the additional data acquired by the additional data acquisition unit to acquire the response of the generation AI to the query. A system for providing generation AI services, equipped with the following features.

2. The aforementioned additional data is external information for RAG (Retrieval-Augmented Generation), The generation unit obtains the response by inputting the query and the additional data to the generation AI. The generation AI service provision system according to claim 1.

3. The aforementioned additional data is the difference in model parameters obtained through fine-tuning or transfer learning. The generation unit applies the additional data to the model parameters of the generated AI to obtain the response. The generation AI service provision system according to claim 1.

4. The system includes an additional data location storage unit that stores multiple sets of location information and information related to the additional data, The generation AI service provision system according to any one of claims 1 to 3, wherein the location information acquisition unit searches for information related to the query and the additional data, and acquires the location information associated with the information related to the additional data, including the search results.

5. The first step is to obtain the query, which is the input to the Generative AI (Artificial Intelligence), A second step is to obtain location information of additional data related to the aforementioned query, A third step involves requesting the additional data based on the location information obtained in the second step, User authentication, wherein authentication is performed according to the location information obtained in the second step, and if authentication is successful, the additional data is provided as a response to the request in the third step, A fifth step is to obtain the response of the generating AI to the query using the additional data provided in the fourth step. A method for providing a generation AI service, comprising the following:

Images