Locally executable generative AI processing system, locally executable generative AI processing method, and computer program

Abstract

This system provides a locally executed generation AI processing system that ensures the security of confidential information while enabling stable operation and flexible use of generation AI. [Solution] In a locally executed generation AI processing system consisting of a user terminal 2 and an on-premise server 1, where AI processing is performed within the user terminal 2, the user terminal 2 includes an input / output unit 22 that receives input questions and outputs responses, and an AI processing unit 23 that generates responses to questions. The on-premise server 1 includes an information storage unit 14 that stores information usable only in the on-premise environment, an acquisition unit 13 that acquires questions from the user terminal 2 and acquires response generation information usable for generating responses to those questions from the information storage unit 14, and a transmission unit 10 that transmits the acquired response generation information to the user terminal 2. The AI ​​processing unit 23 generates responses based on the response generation information and the questions.

Description

Technical Field

[0001] The present invention relates to a locally executable generative AI processing system, a locally executable generative AI processing method, and a computer program suitable for processing by generative AI in a local environment.

Background Art

[0002] In recent years, the popularity of generative AI has been increasing. For example, in Patent Document 1, while managing confidential information in a local environment within the company, only abstract data obtained by vectorizing the confidential information is transmitted to an external generative AI, and it is registered with an external cloud service in a form that cannot be restored to the original information, thereby ensuring security.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, in the system of Patent Document 1 above, since it depends on processing by generative AI in an external cloud service, it is not possible to independently manage and execute generative AI, and while ensuring the security of non-public information such as confidential information, stable operation and flexible utilization of generative AI cannot be achieved. On the other hand, if generative AI is operated on an on-premises server instead of a cloud service, the problem of the security of non-public information can be solved, but due to the problem of server load, multiple operational problems such as difficulty in handling a large number of simultaneous connections, response delay, and processing failure occur. By the way, in recent years, PCs equipped with NPUs specialized for AI processing have been released, and generative AI can be operated relatively smoothly on user PCs, making it possible to run generative AI on user PCs.

[0005] Therefore, the object of the present invention is to provide a locally executable AI generation processing system, a locally executable AI generation processing method, and a computer program that can ensure the security of non-public information while enabling stable operation and flexible use of the generated AI. [Means for solving the problem]

[0006] The present invention A locally executable generative AI processing system consisting of a user terminal and an on-premises server, which performs AI processing within the user terminal, The aforementioned user terminal is An input / output unit that receives input questions and outputs responses, A selection unit that, if the aforementioned question contains confidential information, selects to generate the answer to the aforementioned question itself, and if the aforementioned question does not contain confidential information, selects to have an external cloud server generate the answer to the aforementioned question. If you choose to generate the aforementioned question yourself, An AI processing unit that generates a response to the aforementioned question, Equipped with, The aforementioned on-premises server is An information storage unit that stores information available only in the on-premises environment, An acquisition unit that acquires the question from the user terminal and acquires response generation information from the information storage unit that can be used to generate a response to the question, A transmission unit that transmits the acquired response generation information to the user terminal, Equipped with, The AI ​​processing unit provides a locally executable generative AI processing system that generates the response based on the response generation information and the question.

[0007] According to the present invention, for example, in an on-premises local environment consisting of a user terminal and an on-premises server, the AI ​​processing unit of the user terminal performs processing by a generation AI to generate a response to a question, obtains response generation information that can be used to generate a response to a question from information available only in the on-premises environment, and generates a response based on this response generation information and the question. This allows the system to obtain response generation information from private information in the local environment, and use that response generation information to generate a response to a question on the user's terminal. This enhances the security of private information while ensuring stable operation and flexible use of the generation AI.

[0008] Although this invention is fundamentally in the category of a system, similar effects can be achieved with methods executed by computers constituting a system, or with computer programs that can be read by computers. [Effects of the Invention]

[0009] According to the present invention, it is possible to ensure the security of confidential information while enabling stable operation and flexible use of the generated AI. [Brief explanation of the drawing]

[0010] [Figure 1] This is a schematic diagram illustrating the outline of a locally executed generative AI processing system according to one embodiment of the present invention. [Figure 2] This is a block diagram showing the functional configuration of a locally executed generative AI processing system according to one embodiment of the present invention. [Figure 3] This figure shows the processing flow according to a specific model of a locally executed generative AI processing system related to one embodiment of the present invention. [Figure 4] This is a flowchart showing the flow of local AI processing by a locally executed generative AI processing system according to one embodiment of the present invention. [Figure 5] This flowchart shows the flow of the learning data update process by a locally executable generative AI processing system according to one embodiment of the present invention. [Modes for carrying out the invention]

[0011] Hereinafter, embodiments for carrying out the present invention will be described in detail with reference to the attached drawings. In the following drawings, the same elements are denoted by the same numbers or reference numerals throughout the description of the embodiments.

[0012] Note that the "generative AI" in this specification generally refers to generative artificial intelligence technology that generates new data or content (including not only text but also data such as images, videos, programs, etc.) as an answer based on given input data (questions), and may be implemented by a chatbot utilizing a large language model. In particular, the "locally executed generative AI" referred to in this embodiment is mainly a user terminal that uses computing resources (such as a CPU (Central Processing Unit), NPU (Neural network Processing Unit), or GPU (Graphics Processing Unit)) within the user terminal to generate an answer to a question through inference calculation and inference processing. Also, the "local environment" referred to in this embodiment is synonymous with an on-premises environment that means a management and operation form in which a so-called on-premises network system is constructed.

[0013] [Overview of Locally Executed Generative AI Processing System] FIG. 1 is a schematic diagram showing an overview of a locally executed generative AI processing system. Based on FIG. 1, the components of the locally executed generative AI processing system will be described. The locally executed generative AI processing system is a system in which a generative AI that has server functions and generates an answer to a question is implemented on a user terminal 2. The locally executed generative AI processing system of this embodiment (hereinafter referred to as "this system") includes, in addition to an on-premises server (a server operated in a closed network environment) 1, a user terminal 2 that can be used by a user who asks a question to the generative AI and obtains a response, and an external cloud server (a server available in a public network environment) 3.

[0014] The on-premises server 1 exists in a closed network environment such as an in-house network of a company or the like, and is a computer having server functions. In the on-premises server 1, vector data of non-public information such as confidential information of enterprises, organizations, and individuals, such as important information (customer information, personnel information, financial information, business plans, product technical information, system information, etc.), confidential information (minutes of meetings, customer lists, project plans, contracts, personnel information, accounting information, etc.), and personal information (name, gender, age, address, phone number, job title, salary, family composition, etc.), and a summary of the file entity are stored (memorized). The on-premises server 1 is installed only one in this system.

[0015] The user terminal 2 is a terminal device used by in-house users of a company or the like, for example, terminal devices such as mobile phones, smartphones, tablet terminals, personal computers, laptop computers, etc., and wearable terminals such as smartwatches, smart glasses, and HMDs (Head Mounted Displays). In the user terminal 2 of this system, functions related to generative AI are implemented. When the user terminal 2 uses the generative AI function it has, it once sends a predetermined question to the on-premises server 1, and if there is response generation information available for generating a response to the question, it obtains the response generation information from the on-premises server 1, inputs these questions and the response generation information into the generative AI, and obtains the response output by the generative AI. Also, if the response generation information does not exist in the on-premises server 1, the user terminal 2 can obtain the response generation information from the external cloud server 3 through the on-premises server 1, and by inputting the response generation information and the question obtained from the external cloud server 3 into the generative AI, it can also obtain the response generated by the generative AI. Such user terminals 2 exist in plural in this system.

[0016] The external cloud server 3 exists in a public network environment and is a computer having server functions. The external cloud server 3 has functions of cloud-based generative AI such as SaaS-type generative AI.

[0017] This system accepts general questions as well as questions containing confidential information that should only be handled within the company. For questions containing such confidential information, the system can generate responses using the AI ​​generated by user terminal 2 while ensuring security.

[0018] Furthermore, the details concerning the generative AI itself, such as the data it learns from, its learning method, and the method for constructing the learning model, are not particularly limited. The generative AI may be general-purpose or tuned according to its intended use and method, and is not particularly limited. In addition to the above reasons, since the present invention primarily utilizes the generative AI through the user terminal 2, a detailed explanation of the generative AI is omitted to simplify the description in this specification.

[0019] [Functional Configuration of a Locally Executable Generative AI Processing System] As shown in Figure 2, the on-premise server 1 comprises a transmission unit 10, a reception unit 11, an acquisition unit 13, an information storage unit 14, and a learning data storage unit 15. The transmission unit 10 and the reception unit 11 are mainly implemented by communication devices. The acquisition unit 13 is implemented by a processor executing a program. The information storage unit 14 and the learning data storage unit 15 are implemented by storage devices. The processing by the transmission unit 10, the reception unit 11, the acquisition unit 13, the information storage unit 14, and the learning data storage unit 15 is implemented by a processor executing a program.

[0020] The user terminal 2 comprises a transmission unit 20, a reception unit 21, an input / output unit 22, an AI processing unit 23, a selection unit 24, and a verification unit 25. The transmission unit 20 and the reception unit 21 are implemented by a communication device. The input / output unit 22 is implemented by an input / output device such as a mouse, keyboard, display device, or touch panel. The AI ​​processing unit 23 and the selection unit 24 are implemented by a processor executing a program. The verification unit 25 is implemented by a program that verifies the terminal's own specifications during the installation of the AI ​​processing unit 23. The processing by the transmission unit 20, the reception unit 21, the input / output unit 22, the AI ​​processing unit 23, the selection unit 24, and the verification unit 25 is implemented by a processor executing a program.

[0021] The external cloud server 3 comprises a communication unit 30, an external AI processing unit 31, and a cloud information storage unit 32. The communication unit 30 is implemented by a communication device. The external AI processing unit 31 is implemented by a processor executing a program. The cloud information storage unit 32 is implemented by a storage device.

[0022] As specifically shown in Figure 3, the on-premise server 1 includes a local AI management module 100 and an internal learning data DB 110. The local AI management module 100 is a program executed on the on-premise server 1 to manage the overall processing of the system, and provides functions such as a transmission unit 10, a reception unit 11, and an acquisition unit 13. The internal learning data DB 110 is a database of learning data that associates questions input to the on-premise AI 210 (AI processing unit 23) of the user terminal 2 (described later) with the responses generated in response. The internal learning data DB 110 is used as adjustment material for the on-premise AI 210 to derive appropriate processing results, and as decision material for the local AI management module 100 to extract response generation information corresponding to the questions when it functions as a reception unit 11. The internal learning data DB 110 is constructed in the learning data storage unit 15.

[0023] User terminal 2 has a web browser 200 and an on-premise AI 210. The web browser 200 displays an AI prompt screen for inputting questions to the on-premise AI 210 and the corresponding generated responses, and has the function of exchanging questions and responses with the on-premise server 1, and provides functions such as a transmission unit 20, a reception unit 21, and an input / output unit 22. The on-premise AI 210 is a program that is executed on the user terminal 2 itself as a generating AI, and provides the functions of the AI ​​processing unit 23.

[0024] The external cloud server 3 contains a cloud-based AI 300 and an external training data DB 310. The cloud-based AI 300 is a program that runs as a generative AI that performs general large-scale language modeling and image generation, and provides the functions of the external AI processing unit 31. The external training data DB 310 is used as adjustment material for the cloud-based AI 300 (external AI processing unit 31) to derive appropriate processing results. The external training data DB 310 is built on the cloud information storage unit 32.

[0025] [Functional Configuration of User Terminal 2] The transmission unit 20 launches and displays the AI ​​prompt screen in the web browser 200 and sends the question entered on the AI ​​prompt screen to the on-premise server 1 (see (1) to (3) in Figure 3). The transmission unit 20 also sends the response generated by the AI ​​processing unit 23 to the on-premise server 1 (see (7) in Figure 3). The question may be an interrogative sentence, an imperative sentence, or a request for the creation or analysis of text or images. The response may be a textual answer or may include an image.

[0026] The receiving unit 21 receives the response generation information, described below, transmitted from the on-premises server 1, and inputs the response generation information, along with the question entered into the input / output unit 22, to the AI ​​processing unit 23 (see (5) in Figure 3). The receiving unit 21 also receives the response generated by the external cloud server 3 and transmitted from the on-premises server 1, and transmits this response to the input / output unit 22 (see (8) in Figure 3).

[0027] The input / output unit 22 receives input questions via the web browser 200 and outputs the responses generated by the AI ​​processing unit 23 in response to those questions by displaying, printing, or other means.

[0028] The AI ​​processing unit 23 receives the question input to the input / output unit 22 and the response generation information obtained from the on-premise server 1, etc., and generates a response as a result of processing by the on-premise AI 210's generation AI (see (6) in Figure 3). The AI ​​processing unit 23 executes the processing by the on-premise AI 210's generation AI, and for example, based on the internal training data stored in the training data storage unit 15 of the on-premise server 1, it can train a neural network using a predetermined algorithm and output an appropriate response to a question that includes confidential company information.

[0029] The selection unit 24 selects the destination for the AI ​​processing request, depending on whether the question contains confidential information, whether to perform the AI ​​processing by the generative AI that generates a response to the question in the AI ​​processing unit 23 of the user terminal 2 itself, or in the external AI processing unit 31 of the external cloud server 3. Specifically, if the received question contains confidential information, the selection unit 24 decides to request the AI ​​processing from the AI ​​processing unit 23 of the user terminal 2 itself. In this case, the question containing confidential information is input to the AI ​​processing unit 23. The question containing confidential information is also transmitted to the on-premises server 1. On the other hand, if the received question does not contain confidential information, the selection unit 24 decides to request the AI ​​processing from the external AI processing unit 31 of the external cloud server 3. In this case, the question without confidential information is transmitted to the external cloud server 3 via the on-premises server 1. For example, as shown in (3) of Figure 3, if the question entered into user terminal 2 is something like "How should I process expense reimbursement?", the selection unit 24 determines that the question contains "expenses" which are confidential information and therefore violate the approval regulations. As a result, the question is sent to on-premise server 1 but not to external cloud server 3.

[0030] The verification unit 25 checks the specifications of the user terminal 2 itself when installing the processing program related to the AI ​​processing unit 23. If it is confirmed that the user terminal 2 has the specifications to meet the execution conditions of the AI ​​processing unit 23 and that installation is possible, the verification unit 25 executes the installation so that the processing program of the AI ​​processing unit 23 can be started by the user terminal 2 itself.

[0031] [Functional Configuration of On-Premise Server 1] The transmitting unit 10, under the control of the local AI management module 100, sends a request command to execute AI processing to the AI ​​processing unit 23 of the user terminal 2 or the external AI processing unit 31 of the external cloud server 3, which has been selected by the selection unit 25 of the user terminal 2 as the AI ​​processing request destination, depending on whether or not there is confidential information. When sending a request command to the user terminal 2, the transmitting unit 10 also sends response generation information (see (5) in Figure 3). On the other hand, when sending a request command to the external cloud server 3, the transmitting unit 10 also sends a question that does not contain the confidential information conveyed by the user terminal 2. This prevents the leakage of confidential information outside the local environment. The transmitting unit 10 sends the response generated by the external AI processing unit 31 of the external cloud server 3 to the user terminal 2 (see (8) in Figure 3).

[0032] The receiving unit 11, under the control of the local AI management module 100, receives a question along with a request command sent from the user terminal 2 (see (3) in Figure 3). The receiving unit 11 receives a response generated by the AI ​​processing unit 23 of the user terminal 2 or the external AI processing unit 31 of the external cloud server 3 in response to the request command (see (7) in Figure 3). The response generated by the AI ​​processing unit 23 is transmitted to the learning data storage unit 15 in association with the question. For responses generated by the external AI processing unit 31, the transmitting unit 10 is instructed to send them back to the user terminal 2 (see (8) in Figure 3). As a result, the response generated by the AI ​​processing unit 23 or the external AI processing unit 31 in response to the question is displayed on the input / output unit 22 of the user terminal 2. In particular, questions containing confidential information and the responses generated by the AI ​​processing unit 23 in response are transmitted to the learning data storage unit 15 as internally linked learning data.

[0033] When the acquisition unit 13 receives a request command from the local AI management module 100 that designates the user terminal 2 as the AI ​​processing destination along with a question containing confidential information, it acquires information that can serve as a hint for the answer to the question from the information storage unit 14 or the learning data storage unit 15 as answer generation information (hint information) based on that question. Specifically, the acquisition unit 13 extracts answer generation information related to the confidential information from the information stored in the information storage unit 14 or from the internal learning data DB of the learning data storage unit 15 by performing morphological analysis or vector search of keywords included in the question (see (4) in Figure 3). For example, as shown in (3) in Figure 3, if the question is something like "How should I handle expense reimbursement?", the acquisition unit 13 acquires answer generation information related to "expenses" as confidential information by extracting it from the information in the information storage unit 14 or from the internal learning data of the learning data storage unit 15, as shown in (4) in Figure 3. On the other hand, if the question does not contain confidential information, the acquisition unit 13 can also acquire response generation information related to common keywords in the question by extracting it from the cloud information storage unit 32 of the external cloud server 3. The acquired response generation information is transmitted to the user terminal 2 via the transmission unit 10.

[0034] The information storage unit 14 stores confidential company information and related response generation information that can serve as hints for responses. The learning data storage unit 15 stores questions entered into the user terminal 2 and responses generated by the AI ​​processing unit 23 as internal learning data. The internal learning data is registered as an internal learning data DB, associating questions containing confidential information with their responses. The learning data storage unit 15 also periodically acquires external learning data held by the external cloud server 3 in the cloud information storage unit 32 and updates the internal learning data based on the acquired external learning data. The timing for periodically acquiring external learning data may be set to acquire it at regular intervals using a timer, or at predetermined times using a real-time clock. Furthermore, the content of the acquired external learning data should be limited to content related to business or internal learning data, and it is desirable to acquire only the differences if there are changes, rather than acquiring the same content as previously acquired.

[0035] [Functional Configuration of External Cloud Server 3] The communication unit 30 sends and receives questions, answers, and answer generation information to and from the on-premises server 1. Specifically, the communication unit 30 receives questions that do not contain confidential information sent from the on-premises server 1 and sends answers generated by the external AI processing unit 31 in response to those questions to the on-premises server 1. The communication unit 30 also sends answer generation information related to questions that do not contain confidential information.

[0036] The external AI processing unit 31 generates a response based on a question sent from the on-premises server 1 that does not contain confidential information, as a result of processing by the cloud-based AI 300's generating AI. The external AI processing unit 31 performs processing by the cloud-based AI 300's generating AI, which has been optimized for general questions. For example, it can learn a neural network using a predetermined algorithm based on external training data stored in the cloud information storage unit 32, and output an appropriate response to a general question input.

[0037] The cloud information storage unit 32 stores (saves) the questions input to the external AI processing unit 31 and the responses generated by the external AI processing unit 31 in response to those questions as external training data. The external training data is configured to be registered as an external training data database, associating general questions unrelated to confidential information with their responses.

[0038] [Local AI processing] Next, the local AI processing performed in this system will be explained with reference to Figure 4. Figure 4 is a flowchart showing the flow of local AI processing by this system. Note that the user terminal 2 has already had the AI ​​processing unit 23 installed because it has the specifications to meet the execution conditions for the generated AI after the installation process of the verification unit 25.

[0039] As shown in Figure 4, in local AI processing, first, the input / output unit 22 of the user terminal 2 activates the AI ​​prompt in response to user input (S1). This displays an AI prompt screen on the input / output unit 22 of the user terminal 2 for inputting a question.

[0040] Next, a question is entered into the input / output unit 22 by an internal user (S2). The entered question is sent to the on-premises server 1 via the transmission unit 20.

[0041] Next, the receiving unit 11 of the on-premises server 1 receives the question sent from the user terminal 2 (S3). At this time, it also receives a request command to execute AI processing, which is selected by the selection unit 24 of the user terminal 2 depending on whether or not there is confidential information in the question, to either the AI ​​processing unit 23 of the user terminal 2 or the external AI processing unit 31 of the external cloud server 3.

[0042] Next, the acquisition unit 13 acquires response generation information depending on whether or not there is confidential information in the question (S4). In this case, if the question contains confidential information, the acquisition unit 13 acquires response generation information from the information storage unit 14 or the learning data storage unit 15 of the on-premises server 1. On the other hand, if the question contains confidential information but there is no response generation information in the information storage unit 14 and the learning data storage unit 15, the acquisition unit 13 acquires response generation information from the cloud information storage unit 32 of the external cloud server 3. Note that even if the request command is to request the execution of AI processing to the external AI processing unit 31 of the external cloud server 3, the response generation information may be acquired from the cloud information storage unit 32. However, if the question contains confidential information, the external cloud server 3 will not receive a request command to execute AI processing, nor will it receive a question.

[0043] Next, the transmission unit 10 determines whether or not to send the AI ​​processing request to an internal server depending on whether or not there is confidential information (S5). If the AI ​​processing request is to be sent to an internal server (S5:YES), the transmission unit 10 proceeds to S6, sending the acquired response generation information to the user terminal 2. On the other hand, if the AI ​​processing request is not to be sent to an internal server (S5:NO), the transmission unit 10 proceeds to S9, sending the question along with the request command to the external cloud server 3.

[0044] If the answer in S5 is YES, the transmission unit 10 sends the acquired response generation information to the user terminal 2, with the AI ​​processing unit 23 as the AI ​​processing request destination (S6).

[0045] Next, when the receiving unit 21 of the user terminal 2 receives the response generation information, the AI ​​processing unit 23 of the user terminal 2 generates a response based on the response generation information and the question input to the input / output unit 22 (S7). The generated response is displayed on the input / output unit 22 of the user terminal 2. The generated response is also sent to the on-premise server 1 via the transmission unit 20. As a result, the user terminal 2 displays a response generated by its own AI in response to a question, without the question containing confidential information being leaked outside the local environment. The response generation information may also be displayed along with the response.

[0046] Next, when the on-premises server 1 receives a response from the user terminal 2, the learning data storage unit 15 creates internal learning data that associates the response with the corresponding question and stores it in the internal learning data DB (S8).

[0047] If the answer in S5 is NO, the transmission unit 10 sends a question to the external cloud server 3, designating the external AI processing unit 31 as the AI ​​processing request destination (S9).

[0048] Next, when the communication unit 30 of the external cloud server 3 receives a question, the external AI processing unit 31 of the external cloud server 3 generates a response based on that question (S10). The generated response is sent to the on-premises server 1 via the communication unit 30. The cloud information storage unit 32 creates external training data by associating the generated response with the corresponding question and stores it in the external training data DB (S11).

[0049] Next, when the on-premises server 1 receives a response from the external cloud server 3, the response is forwarded to the user terminal 2 and displayed on the input / output unit 22 of the user terminal 2 (S12). As a result, the user terminal 2 receives a question that does not contain confidential information from the external cloud server 3, and displays a response generated by the AI ​​of the external cloud server 3 in response to that question.

[0050] [Training data update process] Next, the learning data update process performed in this system will be explained with reference to Figure 5.

[0051] As shown in Figure 5, in the training data update process, first, the acquisition unit 13 of the on-premises server 1 determines whether or not it is time to acquire external training data (S20). If it is time to acquire external training data in S20 (S20: YES), the acquisition unit 13 proceeds to the process in S21. If it is not time to acquire external training data (S20: NO), the acquisition unit 13 repeatedly performs the process in S20 until it is time.

[0052] Next, the acquisition unit 13 performs the process of acquiring external training data from the external cloud server 3 (S21).

[0053] Next, the acquisition unit 13 updates the internal learning data stored in the learning data storage unit 15 based on the acquired external learning data (S22). Through this learning data update process, the internal learning data is periodically updated and enriched with external data.

[0054] According to this system, in an on-premises local environment consisting of a user terminal 2 and an on-premises server 1, the AI ​​generation process is performed by the AI ​​processing unit 23 of the user terminal 2 to generate a response to a question input by the user terminal 2. Response generation information usable for generating a response to a question is obtained from information including internal training data stored on the on-premises server 1, which is only available in the on-premises environment. A response is then generated based on this response generation information and the question. As a result, response generation information is obtained from non-public information in the local environment, and the response to the question is generated by the user terminal 2 itself using this response generation information. This enhances the security of non-public information while ensuring stable operation and flexible use of the generation AI.

[0055] Furthermore, the system can choose whether to execute the AI-generated processing on the user terminal 2's own AI processing unit 23 or on the external cloud server 3, depending on whether the question contains confidential information. This allows for the appropriate distribution of AI processing requests.

[0056] Furthermore, response generation information can also be obtained from the external cloud server 3, and the AI ​​processing unit 23 of the user terminal 2 can be made to execute AI processing based on the acquired response generation information and the question. This minimizes the amount of information required for AI processing and reduces resource costs.

[0057] Furthermore, since response generation information can be obtained from either the information storage unit 14 or the learning data storage unit 15 of the on-premises server 1, or the cloud information storage unit 32 of the external cloud server 3, the AI ​​processing unit 23 of the user terminal 2 can also generate an appropriate response based on the response generation information corresponding to the question.

[0058] Furthermore, when the AI ​​processing unit 23 is installed on the user terminal 2, its own specifications are checked, and if it is confirmed that installation is possible, the AI ​​processing unit 23 is installed. As a result, the program related to generating AI that exceeds the specifications is not executed, and the AI ​​processing can be performed on the user terminal 2 itself with a relatively small program size while keeping the computer specifications down.

[0059] Furthermore, by having each user terminal 2 execute AI processing rather than concentrating it on a single server, the load related to AI processing can be distributed and reduced. This allows for a reduction in implementation costs by limiting the computer performance of the on-premise server 1 and user terminal 2.

[0060] Furthermore, by periodically updating the internal training data based on external training data, the training data can be optimized, which in turn allows the AI ​​generated by the AI ​​processing unit 23 to perform training properly.

[0061] It should be noted that the present invention is not limited to the embodiments described above, and can be modified as appropriate within the scope of the present invention.

[0062] For example, internal training data may be stored in association with each user terminal 2, and the generating AI of the AI ​​processing unit 23 may be trained based on the individual internal training data for each user terminal 2. In this case, the AI ​​processing unit 23 of each user terminal 2 can generate appropriate responses to questions specific to each user, thereby personalizing the AI ​​processing. [Explanation of Symbols]

[0063] 1. On-premises server 2 User terminals 3. External cloud server 10 Transmitter 11 Receiving unit 13 Acquisition Department 14 Information storage section 15. Learning Data Storage Unit 20 Transmitter 21 Receiving unit 22 Input / output section 23 AI Processing Unit 24 Selection Section 25 Verification Section 30 Communications Department 31 External AI Processing Unit 32 Cloud Information Storage Unit

Claims

1. A locally executable generative AI processing system consisting of a user terminal and an on-premises server, which performs AI processing within the user terminal, The aforementioned user terminal is An input / output unit that receives input questions and outputs responses, A selection unit that, if the aforementioned question contains confidential information, selects to generate the answer to the aforementioned question itself, and if the aforementioned question does not contain confidential information, selects to have an external cloud server generate the answer to the aforementioned question. If the user selects to generate the aforementioned question, the AI ​​processing unit generates the response to the aforementioned question, Equipped with, The aforementioned on-premises server is An information storage unit that stores information available only in the on-premises environment, An acquisition unit that acquires the question from the user terminal and acquires response generation information from the information storage unit that can be used to generate a response to the question, A transmission unit that transmits the acquired response generation information to the user terminal, Equipped with, The AI ​​processing unit is a locally executed generation AI processing system that generates the response based on the response generation information and the question.

2. In the aforementioned on-premises server, The acquisition unit is capable of acquiring the response generation information from the external cloud server. The local execution type generation AI processing system according to claim 1, wherein the transmitting unit transmits the response generation information obtained from the information storage unit and / or the external cloud server to the user terminal.

3. The aforementioned on-premises server is The system further includes a learning data storage unit that stores the aforementioned question and the response generated by the AI ​​processing unit in response to the question as internal learning data. The acquisition unit can acquire the response generation information from the learning data storage unit. The local execution type generation AI processing system according to claim 1, wherein the transmission unit transmits the response generation information obtained from at least one of the information storage unit, the external cloud server, and the learning data storage unit to the user terminal.

4. The aforementioned user terminal is The locally executable generation AI processing system according to claim 1, further comprising a verification unit that checks its own specifications when the AI ​​processing unit is installed, and installing the AI ​​processing unit when it is confirmed that it can be installed.

5. A locally executable generation AI processing method performed by a locally executable generation AI processing system consisting of a user terminal and an on-premises server, The aforementioned user terminal is The steps include receiving the input question and outputting the response, If the aforementioned question contains confidential information, the user chooses to generate the answer to the aforementioned question themselves; if the aforementioned question does not contain confidential information, the user chooses to have an external cloud server generate the answer to the aforementioned question. If the user chooses to generate the question themselves, the steps include generating a response to the question, It is possible to do this, The aforementioned on-premises server is Steps to store information that is only available in the on-premises environment, The steps include obtaining the question from the user terminal and obtaining response generation information usable for generating a response to the question from an information storage unit that stores information usable only in an on-premises environment, The steps include: transmitting the acquired response generation information to the user terminal; It is possible to do this, The step of generating a response to the aforementioned question is a locally executable generation AI processing method that generates the response based on the response generation information and the question.

6. In a locally executable generative AI processing system consisting of a user terminal and an on-premises server, On the computer of the aforementioned user terminal, A step that receives the entered question and outputs the response. If the aforementioned question contains confidential information, the user chooses to generate the answer to the aforementioned question themselves; if the aforementioned question does not contain confidential information, the user chooses to have an external cloud server generate the answer to the aforementioned question. If the user chooses to generate the question, the step of generating a response to the question, Make it run, On the aforementioned on-premises server computer, Steps to store information that is only available in the on-premises environment, The steps include obtaining the question from the user terminal and obtaining response generation information usable for generating a response to the question from an information storage unit that stores information usable only in an on-premises environment, A step of transmitting the acquired response generation information to the user terminal, Make it run, A computer-readable computer program for generating a response to the aforementioned question, based on the response generation information and the question.

Images