Information processing method and task support system

Abstract

To enable a task worker to easily recognize defects in task processing such as expense settlement and quickly deal with them. [Solution] When a task is executed, the information is sent to a supervisor agent, which verifies any deficiencies in the task, and if any deficiencies are found, instructions for communication with the user, etc. are sent. The first sub-agent that receives the instructions obtains user information from a user data source and notifies the user of the deficiencies through an interpersonal communication tool.

Description

[Technical Field]

[0001] The present disclosure relates to an information processing method and a task assistance system. [Background technology]

[0002] In an information processing program, information processing device, and information processing method that can make necessary corrections to detail data while preventing the settlement amount from being changed against the intentions of the person settling expenses, an information processing server acquires the expense detail data and correction data indicating the correction, determines the difference between the settlement amount in the detail data and the settlement amount in the detail data after correction using the acquired correction data, and if it is determined that there is a difference, displays a warning message on the approver terminal indicating that corrections that would change the amount cannot be made (Patent Document 1).

[0003] Also known is a system and method in which an orchestrator manages multiple agents and generates responses to inputs, where the orchestrator utilizes one or more multimodal models, such as large-scale language models, to process or decompose prompts into a set of instructions for each agent, and each agent uses one or more machine learning models to process disparate inputs or different portions of inputs related to the prompts, and the system summarizes structured and unstructured data records in natural language using the orchestrator, forms an output, and transmits the summary (Patent Document 2). [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent No. 6603426 [Patent Document 2] U.S. Patent No. 12,111,859 Summary of the Invention [Problem to be solved by the invention]

[0005] Incidentally, external business support systems such as external SaaS are used for internal tasks such as expense settlement. In the past, when an employee used an external business support system to perform tasks such as expense settlement, if a procedural error occurred in the external business support system, the system would issue a warning message to the employee.

[0006] However, the procedures for task workers, such as employees who are not familiar with the task processing requirements, to resolve the deficiencies were complicated, and employees from specialized departments such as accounting department staff had to spend time explaining the procedures, resulting in a significant loss of human resources.

[0007] Therefore, the purpose of this disclosure is to enable task workers themselves to easily recognize and quickly address deficiencies in task processing, such as expense settlement, in an external business support system, thereby reducing the burden on personnel, including task workers, and improving work efficiency.

[0008] In addition, problems that are obvious to a person skilled in the art and can be read from the characteristic embodiments of the present disclosure and their explanations, as described in the specification, drawings, etc. of the present disclosure, may also become problems that the divided invention must solve if a divisional application based on the present disclosure is filed. [Means for solving the problem]

[0009] An information processing method according to the present disclosure is an information processing method for supporting a predetermined task on an external work support system performed by a user using a task support system including a plurality of AI agents, the method being implemented using at least one memory for storing instructions and at least one processor configured to execute instructions for performing operations, the task support system including: a supervisor agent that is at least a language model agent and autonomously supports the task; a first subagent that is at least a language model agent that communicates with the supervisor agent and supports a part of the task by autonomously operating an interpersonal communication tool to communicate with the user; a connector that communicates with the external work support system through an interface; a system specification data source related to the task performed by the user using the external work support system; and a user data source that includes user identification information related to the user and the interpersonal communication tool, which is referable to at least the first subagent. a step of the connector transmitting a first instruction to a supervisor agent, the first instruction including information about the user and the task that has occurred on the external business support system, the first instruction including information about the user and the task that has occurred on the external business support system; a step of the supervisor agent using the first instruction received from the connector to refer to the system specification data source and verify any deficiencies in the procedures or file submissions required for the completion of the task; a step of the supervisor agent, when detecting a deficiency, transmitting a second instruction to a first subagent, the second instruction including information about the user and requirements for resolving the deficiency, for communicating with the user to correct the deficiency; and a step of the first subagent using the user data source and an interpersonal communication tool to communicate with the user to correct the deficiency.

[0010] The information processing method according to the present disclosure is an information processing method for supporting a predetermined task on an expense reimbursement system related to invoices from suppliers using a task support system including a plurality of AI agents, the method being implemented using at least one memory for storing instructions and at least one processor configured to execute instructions for performing operations, wherein the task support system includes a supervisor agent that is at least a language model agent and autonomously supports the task, a first sub-agent that is at least a language model agent that communicates with the supervisor agent and supports part of the task by autonomously operating an interpersonal communication tool to communicate with the supplier, and a second sub-agent that is an agent that handles files related to invoices submitted by the supplier. and a supplier data source containing supplier identification information related to suppliers and an interpersonal communication tool that is accessible to at least a first subagent, the method including the steps of a supervisor agent using the data source to directly or indirectly operate the expense reimbursement system to perform reconciliation work, identifying suppliers to which files related to unbilled receivables and receivables should be sent when performing the reconciliation work, the supervisor agent sending to the first subagent, for the identified supplier, a billing instruction containing information for having the supplier submit a billing file, and the first subagent using the supplier data source and the interpersonal communication tool, requesting the supplier to submit a billing file.

[0011] The present disclosure also provides a task assistance system using a plurality of AI agents, which is implemented using at least one memory for storing instructions and at least one processor configured to execute instructions for performing operations, and the task assistance system includes a supervisor agent that is at least a language model agent and autonomously assists with a task, a first subagent that is at least a language model agent that communicates with the supervisor agent and assists with a part of the task by autonomously operating an interpersonal communication tool to communicate with the supervisor agent, a connector that communicates with an external task assistance system through an interface, a system specification data source related to a task to be performed by a user using the external task assistance system, and a system specification data source that can be referenced by at least the first subagent. and a user data source containing user identification information related to the user and an interpersonal communication tool, wherein the connector sends a first instruction containing user and task information to a supervisor agent regarding a user and a task of the user that has occurred on the external business support system, the supervisor agent uses the first instruction received from the connector to refer to the system specification data source and verify any deficiencies in the procedures or file submission required for completing the task, and if the supervisor agent detects a deficiency, sends a second instruction to a first subagent containing information about the user and requirements for resolving the deficiency, for communicating with the user to correct the deficiency, and the first subagent uses the user data source and the interpersonal communication tool to communicate the requirements for resolving the deficiency to the user.

[0012] The present disclosure also provides a task assistance system that uses a plurality of AI agents to assist in a predetermined task on an expense reimbursement system related to invoices from suppliers, the task assistance system being implemented using at least one memory that stores instructions and at least one processor configured to execute the instructions to perform operations, the task assistance system including a supervisor agent that is at least a language model agent and that autonomously assists in the task, a first sub-agent that is at least a language model agent that communicates with the supervisor agent and assists in part of the task by autonomously operating an interpersonal communication tool to communicate with the supplier, and a second sub-agent that is an agent that handles files related to invoices submitted by the supplier. and a supplier data source containing supplier identification information related to suppliers and interpersonal communication tools that can be referenced by at least the first subagent, wherein a supervisor agent uses the data source to directly or indirectly operate the expense reimbursement system to perform reconciliation work, and identifies suppliers to which files related to unbilled receivables and receivables should be sent when performing the reconciliation work, and the supervisor agent sends billing instructions to the first subagent for the identified suppliers, including information for having the suppliers submit files related to billing, and the first subagent uses the supplier data source and the interpersonal communication tool to communicate a request to the supplier to submit files related to billing. [Effects of the Invention]

[0013] The information processing method or task support system disclosed herein allows task workers themselves to easily recognize and quickly address deficiencies in task processing, such as expense settlement, in an external business support system, thereby reducing the burden on personnel, including task workers, and improving work efficiency. [Brief explanation of the drawings]

[0014] [Figure 1] 1 is an overview of a task assistance system used in an information processing method of the present disclosure. [Figure 2] 1 is a block diagram showing the configuration of a computing device 100. FIG. [Figure 3] FIG. 5 illustrates an example of a data source 500. [Figure 4] FIG. 3 is a sequence diagram showing a workflow according to the first embodiment. [Figure 5] FIG. 3 is a sequence diagram showing a workflow according to the first embodiment. [Figure 6] FIG. 10 is a sequence diagram showing a workflow according to the second embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0015] In the present disclosure, detailed descriptions of well-known technologies may be omitted to ensure conciseness. In the description, numerical labels such as "first," "second," etc. are used to identify each element and do not specify the number of elements. In addition, in the terms used, the singular form may generally be understood to include the plural form, and vice versa.

[0016] 1 is a schematic diagram of a task assistance system used in the information processing method of the present disclosure. As will be described later, the task assistance system is implemented using at least one memory that stores instructions and at least one processor configured to execute the instructions to perform operations, and includes multiple AI agents.

[0017] The external business support system 200 is used by users who are, for example, company employees to perform business tasks, and includes external SaaS and the like. For example, the external business support system 200 may be an ERP system 220 or an expense settlement system 210. Other external business support systems include accounting systems, tax management systems, attendance management systems, payroll management systems, entry / exit management systems, production management systems, ordering systems, purchasing systems, inventory management systems, sales support systems, CRM (customer relationship management) systems, approval systems, contract management systems, patent information management systems, integrated mission-critical systems (IMBS), mission-critical systems, and business management systems. While the present disclosure primarily uses the expense settlement system 210 as an example, the present disclosure is not limited to this example.

[0018] A user operates, for example, a general-purpose computing device 100 to perform a task using an external task assistance system. In this disclosure, the computing device 100 used by the user may be referred to simply as a terminal or a user terminal. Since the computing device also has a common configuration with a computing device such as a server computer that implements the task assistance system 1000, the following description will also be given.

[0019] 2 is a block diagram showing the configuration of a computing device 100. The computing device 100 includes a communication interface 111, an input interface 112, an output interface 113, a processor 114, a memory 115, and a storage 116. The computing device 100 is provided as a computational resource for realizing the functions described herein. This may be provided as a physical device or a virtual device.

[0020] The communication interface 111 is a component that allows the computing device 100 to send and receive data to and from other external devices. Communication methods may include both wired and wireless communication. The input interface 112 is an interface that receives instructions and data from a user or an external device. Examples of devices that can be connected include a keyboard, mouse, and touch screen. The output interface 113 is a component that allows the computing device 100 to provide processing results and information to a user or an external device. Examples of such interfaces include a display, speaker, and printer.

[0021] The processor 114 is a processing device that performs overall control of the operation of the computing device 100, and can be configured with a single or multiple processor units. The processor 114 is realized by a central processing unit (CPU) that performs general-purpose arithmetic processing, a graphics processing unit (GPU) that specializes in image processing and parallel computing, a neural processing unit (NPU) or tensor processing unit (TPU) that is optimized for artificial intelligence processing, or a combination of these.

[0022] The processing by the processor 114 may be performed as parallel processing by multiple processor units, or each process may be assigned to a different processor unit for distributed execution. Each processor can operate independently, and one processor does not need to execute multiple tasks sequentially. This improves processing efficiency and load distribution. The type and configuration of the processor 114 are selected appropriately depending on the use and purpose of the computing device 100.

[0023] The processor 114 executes programs temporarily stored in the memory 115 to perform various types of arithmetic processing and control. Specifically, programs stored in the storage 116 are loaded into the memory 115, and the processor 114 executes them to control the operation of the entire computing device 100. The memory 115 is a volatile memory that can be accessed at high speed and temporarily stores programs and data required for processing by the processor 114.

[0024] Storage 116 is a storage device for long-term storage of data and programs. Specific examples include non-volatile memory and a hard disk drive. Storage 116 includes a data area and a program area, as shown below. The data area stores application data and operating system data. The application data includes information related to applications used by users. Meanwhile, the operating system data includes information supporting the basic operation of the entire system. The program area stores application programs and the operating system. Application programs are software for performing specific tasks, and the operating system is basic software that manages the resources of computing device 100. These components work together to realize the functions expected of computing device 100.

[0025] Each computing device can operate as an independent device, or can operate while connected to other devices or a network. Computing devices can be realized in various forms, such as a configuration in which they operate independently in a local environment, a configuration in which they function as part of a distributed system in which multiple devices operate in cooperation with each other, or a configuration in which they function as a computing resource that can be remotely accessed via a network. Appropriate data is sent and received according to these different operating forms via a communication interface.

[0026] Returning to Fig. 1, a user terminal 100, an external business support system 200, and a task support system 1000 are connected via a network. The task support system 1000 includes a system orchestrator 300 for managing the external business support system and a hierarchical group of AI agents.

[0027] The hierarchical AI agent group includes a supervisor agent 400, which is a higher-level agent, and at least a communication agent 610, which is a sub-agent for communicating with the user.

[0028] Here, we will explain the AI ​​agent of the present disclosure. The concept and possibilities of an AI agent itself extend to being installed in a robot, automobile, machine, etc., and equipped with sensors and actuators that can physically perform tasks in the real world, but in this disclosure, it is an AI agent that only performs information processing, and refers to an autonomous software module that runs on a computer, updates its internal state for a given task using data acquired from the digital environment as input, and performs non-physical actions such as outputting data to external systems or other agents, calling APIs, and sending messages through information processing such as inference, learning, and planning.

[0029] In this way, AI agents achieve overall goals through a series of learning and inference processes, either individually or as components of a multi-agent system, working together. Their implementation is limited to program code or a virtualized execution environment, and does not include physical actuators. Furthermore, they are distinguished from robotic agents in that they only perform actions that are completed within computer resources, such as log analysis, user interface operations, and database updates, and do not perform external mechanical actions.

[0030] In this disclosure, a language model agent refers to an autonomous software unit equipped with a probabilistic natural language generation model such as a large-scale language model (LLM) as its core inference mechanism, which analyzes received text in a semantic space and executes linguistic actions such as dialogue, summarization, plan generation, and knowledge retrieval in a chain. The language model is not limited to a large-scale language model; it may also be a language model such as a small language model (SLM) or a diffusion language model (Diffusion LM), or it may be a model constructed using data and its machine learning. For example, a language model that utilizes the Transformer mechanism is well known. When text or a sentence is input, it can output text or a sentence according to the content of the input sentence. While a code-only agent directly manipulates abstract syntax trees, bytecode, etc. and specializes in program editing and execution, a language model agent uses natural language as a medium to perform goal decomposition, external API calls, and collaboration with other agents, and can flexibly adapt to unstructured requests.

[0031] Multimodal models can also be used as foundational models for core inference mechanisms other than language models. One example of a multimodal model is the Vision-Language Model (VL). The VL model (Vision-Language Model) incorporates the natural language understanding and generation capabilities of language models, while mapping visual information such as images and videos into the same representation space. It performs inference and generation through cross-referencing of text and visual signals. While it naturally handles purely linguistic tasks such as dialogue, summarization, translation, and planning, it also provides additional functionality not possible with language models alone, such as caption generation with visual input, OCR, image question answering, image editing suggestions based on visual instructions, and object location identification. For example, a multimodal model that can input images and text and output text can accept text information and uploaded images of supporting documents and generate responses and judgments based on them.

[0032] The underlying model data, such as the language model that forms the basis of the AI ​​agent, may be installed locally or may be a model available externally via the cloud, on-premise, or via API. In other words, local models can be used as models become lighter and their inference accuracy improves. Furthermore, a model that is rational in terms of token economics may be selected as long as it can ensure the necessary inference accuracy for the task assigned to each agent. Available models and APIs are not limited to those currently known; those released after the filing of this disclosure may also be used.

[0033] 1, the supervisor agent 400 is at least a language model agent, and may further be a multimodal model agent based on a model such as a VL model, and can autonomously assist the user in performing tasks. The communication agent 610 is also at least a language model agent, and can assist with part of the user's tasks by autonomously operating an interpersonal communication tool 611 to communicate with the user.

[0034] In this way, the task assistance system 1000 is a system that includes multiple AI agents, and is also called a multi-agent system or agentic AI system. The components that make up the task assistance system 1000 will be described in detail below.

[0035] The system orchestrator 300 is a configuration that can be called a system management system. The system orchestrator 300 has a reception and conversion layer that connects external services such as external business support systems with its own agent platform, and in this disclosure, this part is tentatively referred to as a connector 310. In addition to a connector, it may also be called an adapter, an injector, or the like.

[0036] The connector 310 receives external triggers, such as webhooks sent by an external task transmission system, via a network, converts them into instructions that the supervisor agent 400 can easily understand as tasks, and sends them to the supervisor agent 400. Because the information sent by the external task transmission system is standardized and structured, making conversion easy, the connector 310 may be implemented in code and does not require a language model agent. The reception, conversion, and transfer process performed by the connector 310 is not limited to push-type information communication from the external task transmission system, but may also be pull-type information communication in which the external task support system responds to a request from the connector 310. In this way, communication between the external task support system 200 and the system orchestrator 300 or the connector 310 may be performed via an API (application programming interface) in a broad sense. Note that a similar function may be performed by the data / file loader agent 620, described later.

[0037] The supervisor agent 400 receives a first instruction regarding a task from the system orchestrator 300 or the connector 310 and autonomously analyzes the task, refers to context such as a data source as necessary, creates a plan, identifies a lower-level subagent to which the instruction should be issued, and performs output such as sending an instruction to assign the task. Therefore, the processing performed by the supervisor agent 400 cannot be expressed by a simple flowchart or the like, but is complex, including a combination of request, reference, analysis, inference, and information generation, and repeating these steps.

[0038] The task assistance system 1000 includes data sources 500 that the AI ​​agents can access according to their assigned permissions. Although these data sources are not explicitly shown in the figure, each agent may be authorized to access them as needed to accomplish a task. FIG. 3 shows an example of the data source 500. The data source 500 may include multiple data sources, such as a user data source 510, a system specification data source 520, an internal rules data source 530, a domain knowledge data source, and an internal shared data source 550.

[0039] These data sources may be vector databases or search expansion and generation (RAG) so that each AI agent can refer to them with improved token utilization efficiency. Therefore, information to be stored in the data sources may be converted into vectors using an embedded model. The hardware configuration of the storage device used in the data source 500 may include, for example, a main memory device (hard disk, SSD, optical disk, etc.), a secondary memory device (ROM, RAM, etc.), or a register.

[0040] Referring back to FIG. 1, the supervisor agent 400 can send instructions to other agents, assign tasks, and receive responses to requests. Communication between agents can also be achieved via APIs or A2A (Agent-to-Agent) protocols. In this way, the supervisor agent 400 can autonomously divide tasks and assign them to other agents, thereby efficiently utilizing its own context length, which is more efficient than if a single agent were to carry out the task. For example, even if a long context is required to execute a complex task, the agent does not need to consume its own context length.

[0041] The communication agent 610 (first subagent) communicates with the user using an interpersonal communication tool 611 to carry out the instruction (second instruction) received from the supervisor agent 400. The instruction includes the user to be communicated with and the content to be contacted, and can use the interpersonal communication tool 611 that is available to the communication agent.

[0042] The interpersonal communication tool 611 is, for example, a communication tool used by humans. One example is an integrated email and calendar application for businesses. This email application allows each employee to have their own account and email address as a user, and can use specific accounts and addresses for internal and external communication. Another example is a team messaging service, a chat app that organizes topics into virtual rooms called channels and tracks the flow of conversations using a thread function. Once an account is identified, a mention function allows messages to be sent only to that person. These interpersonal communication tools 611 have plug-in functions that the communication agent 610 can use as tools.

[0043] The data source 500 may include a user data source 510 required to identify a user account to which a message should be sent when using these interpersonal communication tools 611. In other words, the supervisor agent 400 and the communication agent 610 can directly or indirectly obtain information about the user who operated the external business support system, but the data source 510 is used in cases where the user's information alone is not enough to access the account on the interpersonal communication tool 611. For example, the user data source 510 includes identification information (e.g., employee ID), email address, chat account, etc. of an employee A, and the communication agent 610 can compare these and contact the specific user instructed by the supervisor agent 400.

[0044] Furthermore, by using organizational information such as the hierarchical relationships between users contained in the user data source 510, it is possible to identify a user who is the superior of a specific user as a related user. This makes it possible to contact not only a specific user but also related users.

[0045] Furthermore, since the communication agent 610 is at least a language model agent, it can flexibly generate sentences to be written in e-mails and chats in an undefined form using its own language model function. This makes it possible to create sentences that are easier for human users to understand. For example, if the supervisor agent 400 specifies a deadline or a level of importance, it can generate sentences with a corresponding level of urgency.

[0046] The data / file loader agent 620 (second subagent) is an AI agent that can operate files (upload, download, etc.) using a tool that it can operate itself, such as a web browser or RPA. The data / file loader agent 620 can input or acquire information from an external business support system. Therefore, it is possible for the data / file loader agent 620 to input or acquire information from an external business support system. Therefore, the data / file loader agent 620 may be responsible for collecting information from the external business support system instead of the connector 310.

[0047] For example, when expense claim information is shared in the form of a CSV file or the like with the cloud data source 700 through the collaboration function of an external business support system, the contents of the CSV file may be analyzed and a first instruction may be sent to the supervisor agent 400.

[0048] The document analysis agent 630 can analyze, for example, document files, spreadsheet files, presentation files, or image files that can be OCR-processed, all uploaded to the internal shared data source 550, and has a function of analyzing the text contained in files received from users, etc. Therefore, the document analysis agent 630 can create a summary of the text, check the consistency between multiple sentences, and further check compliance with various laws and regulations (for example, whether there are any legal issues in the contract or inconsistencies with internal company rules), whether the image meets the requirements for evidentiary documents, etc.

[0049] The task planner agent 640 is an agent that responds with a plan for accomplishing a task in response to an instruction from the supervisor agent 400. For example, based on an instruction from the supervisor agent 400, it can read the objectives, constraints, priorities, deadlines, etc., inquire about missing information if necessary, break down the target into sub-goals such as goals and milestones, determine the execution order and resource allocation, create a plan, and send it to the supervisor agent 400.

[0050] The specialist agent 650 is a specialist in specific domain knowledge, and can respond as a specialist in a specific area by referring to the domain knowledge data source 540. For example, if domain knowledge related to accounting is stored in the domain knowledge data source 540, the specialist agent 650 can act as an accounting specialist.

[0051] As described above, the multi-agent community of the present disclosure has a hierarchical structure. In some multi-agent communities, agents discuss and vote with each other to make decisions democratically. However, the tasks that users perform using external business support systems are often routine, and a top-down structure with a supervisor agent 400 allows tasks to be completed more efficiently.

[0052] An information processing method using the above task assistance system 1000 will be described below. FIG. 4 is a sequence diagram showing a workflow of essentially minimal processing for a first example of the information processing method according to the first embodiment of the present disclosure. This workflow is realized by the operation of the processor and the operation of each AI agent. As mentioned above, a computing device including a processor may be installed in a local environment or may be virtually realized on a cloud platform. In particular, a cloud environment allows for efficient use of computing resources.

[0053] A process for expense settlement, which is a first example of the first embodiment of the present disclosure, will be described with reference to Fig. 4. In this case, the external business support system 200 is an expense settlement system 210.

[0054] In step S101, the user operates the user terminal 100 to perform user task 1, which is part of a task, using the expense settlement system 210. This corresponds to, for example, the user logging in to the expense settlement system and entering application information into the system. In other words, the flow starts from the system, and the task operator does not need to enter a prompt to interact with the generation AI.

[0055] Other examples of tasks include applying for meeting expenses, entertainment expenses, and accommodation expenses.

[0056] Next, in step S102, the expense reimbursement system 210 communicates information about the user and the user task 1 performed by the user with the system orchestrator 300, the connector 310, or the data / file loader agent 620 (hereinafter also referred to as the connector 310, etc.).

[0057] Next, in step S103, the connector 310 or the like transmits a first instruction containing information about the user and user task 1 to the supervisor agent 400 regarding the user and the user task 1 that occurred on the expense settlement system 210. This corresponds to, for example, the connector 310 or the like automatically transmitting to the supervisor agent 400 a standard output instruction such as "Perform processing on the expense claim in accordance with the requirements of the expense claim system," or an instruction to that effect.

[0058] Next, in step S104, the supervisor agent 400 uses the first instruction received from the connector 310 or the like to directly or indirectly reference the system specification data source 520 and verify any deficiencies in the procedures or file submission required to perform user task 1. This corresponds to, for example, the supervisor agent directly or indirectly referencing the system specification for the expense reimbursement system 210 and verifying whether an approval request has been submitted for the expense claim. During the verification, the supervisor agent 400 may perform complex processing that includes a combination of request, reference, analysis, inference, and information generation, and repeats these processes to autonomously analyze the task, reference context such as a data source as needed, create a plan, identify subordinate subagents to which instructions should be issued, and send instructions to assign tasks. The complex processing performed by the supervisor agent 400 will not be described further below.

[0059] Next, in step S105, if the supervisor agent 400 detects a deficiency, it transmits a second instruction to the communication agent 610, which is the first subagent, including information about the user and the requirements for resolving the deficiency, to communicate with the user and correct the deficiency. This corresponds to, for example, the supervisor agent 400 detecting that an approval request is required for expense settlement, identifying a task of sending a message to the applicant prompting the applicant to submit an approval request, determining that a second instruction needs to be sent to the communication agent 610, and issuing an output instruction to the communication agent 610 such as "Notify the applicant of a message prompting the applicant to submit an approval request." Note that, upon receiving the second instruction, the communication agent 610 can not only receive unilateral instructions (single-pass) from the supervisor agent 400, but also make inquiries through exchanges (multi-pass) to obtain information as needed. This allows for more accurate task execution.

[0060] Other examples of procedural deficiencies include when applying for meeting expenses and entertainment expenses, the number of participants in a meeting or social gathering is not entered, and when applying for accommodation expenses, the amount applied for exceeds the amount set at 10,000 yen per night according to internal rules data source 530.

[0061] Next, in step S106, the communication agent 610, which is the first subagent, uses the user data source 510 and the interpersonal communication tool 611 to communicate to the user the requirements for resolving the deficiency. For example, the communication agent 610 analyzes the second instruction from the supervisor agent 400 and determines that an effective way to implement the instruction is to send the applicant a message urging them to submit a request for approval using one of the interpersonal communication tools 611. The communication agent 610 then sends the applicant a message using a team messaging service, such as, "Please submit a request for approval for the expense request with request number XX." Other messages that can be sent depending on the specific deficiency include, "Please enter the number of participants for the meeting or social gathering." and "The accommodation fee is capped at 10,000 yen. This request exceeds 10,000 yen, so only 10,000 yen will be reimbursed." The most effective message transmission method can be selected from available message transmission methods, such as email, a text-based internal messaging application, or displaying the message on the user's screen, by referring to the system specification data source 520 or various data sources.

[0062] Furthermore, in step S107, the communication agent 610 can use the user data source 510 to identify related users other than the user who are necessary for the user to complete the task, and can communicate to the related users the requirements for resolving the deficiencies. In other words, a message similar to the message sent in step S106 can also be sent to related users such as the user's superior. The communication agent 610, which is a language model agent, can also create a message tailored to the relationship, such as "There is a deficiency in Mr. / Ms. X's application. Please urge him / her to correct it," taking into consideration that the related user is the user's superior, for example.

[0063] As described above, in the first example of the first embodiment of the present disclosure, even if there is a defect in the expense claim processing, the task worker can easily recognize it and quickly respond. This reduces the burden on personnel, including the task worker, and improves work efficiency. In this case, the task worker does not need to input prompts to interact with the generation AI.

[0064] Next, a process for verifying evidence such as receipt images, which is a second example of the first embodiment of the present disclosure, will be described with reference to FIG. 5. Continuing on, the external business support system 200 is the expense reimbursement system 210. FIG. 5 is a sequence diagram showing a workflow of essentially minimal processing for the second example of the first embodiment of the information processing method of the present disclosure. Note that the following description will be based on the example, and overlapping explanations already given will be omitted.

[0065] In steps S201 to S203, when the applicant user logs in to the expense reimbursement system 210 and enters application information, the system automatically sends a standard output instruction to the supervisor agent 400, stating, "Perform processing of the expense application in accordance with the requirements of the expense reimbursement system."

[0066] In step S204, the supervisor agent 400 references a data source such as a system specification data source and determines that uploading a receipt image is required to execute the first instruction. The supervisor agent 400 then determines that a task is required to notify the applicant of a message prompting them to upload a receipt image. At this time, the supervisor agent 400 can inquire of an agent responsible for handling information about the file submitted by the user, such as the data / file loader agent 620, about whether or not a receipt image has been uploaded, and confirm this.

[0067] Other examples of deficiencies that require the uploading of images include when the content of a receipt image differs from the expense subject even if the image has been uploaded, or when only dollar-denominated images of supporting documents are uploaded when yen-denominated images of supporting documents are required.

[0068] Therefore, to detect these defects, the agent that analyzes the content of the receipt image may be a multimodal model agent that is capable of recognizing at least images and text, and if the file submitted by the user is an image, it can recognize the image and analyze the content of said image.

[0069] In this way, the supervisor agent 400 can detect any defects in the uploaded evidence image with the cooperation of other agents.

[0070] In step S205, the supervisor agent 400 transmits a second instruction to the communication agent 610, which is to "send a message to the applicant urging him to upload a receipt image." Other example instructions include "send a message to the applicant notifying him that the receipt image is incorrect and urging him to upload a correct receipt image," and "send a message to the applicant urging him to upload a yen-denominated document image instead of a dollar-denominated document image."

[0071] In step S206, the communication agent 610 analyzes the second instruction and determines that an effective way to execute the instruction is to convey a message prompting the user to submit an image of the receipt. The communication agent 610 then conveys the message "Please upload an image of the receipt." Other examples of messages include "The receipt image is incorrect. Please upload the correct image of the receipt," and "Please upload an image of the document denominated in yen, not the document denominated in dollars."

[0072] As described above, in the second example of the first embodiment of the present disclosure, even if an image is submitted during expense claim processing or if there is a defect in the image itself, the task worker can easily recognize this and quickly respond. This reduces the burden on personnel, including the task worker, and improves work efficiency. In this case, the task worker does not need to enter prompts to interact with the generation AI.

[0073] Next, a request process for an unbilled invoice, which is a third example of the second embodiment of the present disclosure, will be described using FIG. 6. Continuing on, the external business support system 200 is the expense settlement system 210. FIG. 6 is a sequence diagram showing a workflow of essentially the minimum processing for the third example of the second embodiment of the information processing method of the present disclosure. Note that the following description will be based on the example, and any overlapping with the explanation already given will be omitted.

[0074] In step S301 of FIG. 6, the data / file loader agent 620 acquires a file related to billing, such as an invoice, from the expense settlement system 210.

[0075] Next, in step S302, at a time such as the end of the month, the supervisor agent 400 uses the various data sources 500 to operate the expense settlement system 210 directly or indirectly via the data / file loader agent 620 to perform settlement work.

[0076] Next, in step S303, the supplier to which the unbilled payables and the file relating to the billing of the payables are to be sent when the settlement work is performed is identified.

[0077] Next, in step S304, the supervisor agent 400 sends billing instructions to the communication agent 610 for the identified supplier, the instructions including information for the supplier to submit a file relating to the bill.

[0078] Next, in step S305, the communication agent 610 uses the supplier data source and the interpersonal communication tool 611 to communicate to the supplier a request to submit a file related to the claim.

[0079] As described above, in the third example of the second embodiment of the present disclosure, even if there is an unclaimed invoice, the task assistance system can easily recognize it and respond quickly. This reduces the burden on personnel and improves work efficiency. In this case, there is no need for employees to interact with the generation AI by entering prompts.

[0080] The implementations described herein are merely exemplary. Modifications and modifications may be made by various means or structures to achieve some or all of the functions and results and advantages. All such variations and modifications are considered to be within the scope of the implementations described herein. Furthermore, the configurations described herein are merely exemplary and may be modified depending on the actual application. Those skilled in the art will recognize and ascertain many equivalents to the specific implementations described herein through routine design and design. Therefore, the implementations described herein are merely examples. Practices, including aspects not expressly described herein, are possible within the scope of the claims and their equivalents. The present disclosure includes each feature, system, object, code, and information described herein, as well as all combinations thereof to the extent not mutually inconsistent.

[0081] Certain embodiments of the present disclosure are specifically described herein below. [1] 1. An information processing method for supporting a predetermined task on an external business support system to be performed by a user using a task support system including a plurality of AI agents, the task support system being implemented using at least one memory for storing instructions and at least one processor configured to execute instructions for performing operations, the method comprising: The task assistance system includes: a supervisor agent, which is at least a language model agent, that autonomously assists in the task; a first sub-agent, which is at least a language model agent, that communicates with the supervisor agent and assists with a part of the task by autonomously operating an interpersonal communication tool to communicate with the user; and a connector that communicates with the external business support system through an interface; a system specification data source relating to tasks performed by a user using the external business support system; a user data source, referenceable by at least the first subagent, containing identification information of the user relating to the user and the interpersonal communication tool; a step in which the user performs a part of a task using the external business support system; the external business support system communicating information about the user and the task performed by the user to the connector; a step of the connector transmitting a first instruction to the supervisor agent regarding the user and the task of the user that has occurred on the external business support system, the first instruction including information about the user and the task; the supervisor agent using the first instruction received from the connector to reference the system specification data source and verify any deficiencies in the submission of procedures or files required to perform the task; If the supervisor agent detects a deficiency, sending a second instruction to the first subagent to communicate with the user and correct the deficiency, the second instruction including information about the user and requirements for resolving the deficiency; the first subagent using the user data source and the interpersonal communication tool to communicate to the user requirements for resolving the deficiency; An information processing method, including: [2] the first subagent uses the user data source to identify participant users different from the user who are necessary for the user to perform the task; The information processing method described in [1] further includes a step of informing the relevant user of the requirements for resolving the deficiency. [3] The information processing method described in [1], wherein the task support system further includes a second sub-agent that is an agent that handles information of files submitted by users. [4] the second sub-agent is a multimodal model agent capable of recognizing at least images and text; The information processing method according to [3], wherein the second subagent analyzes the content of the image if the file submitted by the user is an image. [5] 1. An information processing method for assisting a predetermined task on an expense reimbursement system relating to invoices from suppliers using a task assistance system including a plurality of AI agents, the method being implemented using at least one memory for storing instructions and at least one processor configured to execute the instructions to perform operations, the method comprising: The task assistance system includes: a supervisor agent, which is at least a language model agent, that autonomously assists in the task; a first sub-agent, which is at least a language model agent, that communicates with the supervisor agent and assists with a part of the task by autonomously operating an interpersonal communication tool to communicate with the supplier; and a second subagent, which is the agent handling the files relating to claims submitted by said supplier; a data source relating to tasks performed using the expense reimbursement system; a supplier data source, visible to at least the first subagent, containing supplier identification information relating to the supplier and the interpersonal communication tool; Including, the supervisor agent directly or indirectly operates the expense reimbursement system using the data source to perform reconciliation work; Identifying suppliers to whom files related to unbilled payables and billings related to said payables should be sent when said clearing operation is performed; sending, by the supervisor agent, for the identified supplier, to the first sub-agent, billing instructions including information for having the supplier submit a file relating to the bill; the first sub-agent communicating a request to the supplier to submit a file regarding the claim using the supplier data source and the interpersonal communication tool; An information processing method, including: [5] 1. A task assistance system using multiple AI agents implemented with at least one memory storing instructions and at least one processor configured to execute the instructions to perform operations, comprising: The task assistance system includes: A supervisor agent, which is at least a language model agent and autonomously assists with tasks; a first sub-agent, which is at least a language model agent, that communicates with the supervisor agent and assists with a part of the task by autonomously operating an interpersonal communication tool to communicate with the user; a connector that communicates with an external business support system through an interface; a system specification data source relating to tasks performed by a user using the external business support system; a user data source, referenceable by at least the first subagent, containing identification information of the user relating to the user and the interpersonal communication tool; the connector transmits a first instruction to the supervisor agent regarding the user and the task that has occurred on the external business support system, the first instruction including information about the user and the task; the supervisor agent, using the first instruction received from the connector, consulting the system specification data source to verify any deficiencies in the submission of procedures or files required to perform the task; If the supervisor agent detects a deficiency, sending a second instruction to the first subagent to communicate with the user and correct the deficiency, the second instruction including information about the user and requirements for resolving the deficiency; the first subagent uses the user data source and the interpersonal communication tool to communicate to the user requirements for resolving the deficiency; Task support system. [7] the first subagent uses the user data source to identify participant users different from the user who are necessary for the user to perform the task; The task support system described in [6], which communicates requirements for resolving the deficiencies to the relevant users. [8] The task support system described in [7], further comprising a second subagent that handles information on files submitted by users. [9] the second sub-agent is a multimodal model agent capable of recognizing at least images and text; The task support system described in [8], wherein the second subagent analyzes the content of the image if the file submitted by the user is an image.

[10] 1. A task assistance system for assisting a predetermined task on an expense reimbursement system relating to supplier invoices using a plurality of AI agents, the system being implemented using at least one memory for storing instructions and at least one processor configured to execute the instructions to perform operations, the system comprising: The task assistance system includes: a supervisor agent, which is at least a language model agent, that autonomously assists in the task; a first sub-agent, which is at least a language model agent, that communicates with the supervisor agent and assists with a part of the task by autonomously operating an interpersonal communication tool to communicate with the supplier; and a second subagent, which is the agent handling the files relating to claims submitted by said supplier; a data source relating to tasks performed using the expense reimbursement system; a supplier data source, visible to at least the first subagent, containing supplier identification information relating to the supplier and the interpersonal communication tool; Including, the supervisor agent uses the data source to directly or indirectly operate the expense reimbursement system to perform settlement work; Identifying suppliers to whom files related to unbilled payables and billing related to said payables should be sent when said clearing operation is performed; the supervisor agent, for the identified supplier, sending to the first subagent a claim instruction including information for causing the supplier to submit a file regarding the claim; the first subagent communicates to the supplier a request for a file regarding the claim using the supplier data source and the interpersonal communication tool; Task support system.

[0082] The above embodiments [1] to

[10] can be arbitrarily combined. For example, all or part of the embodiments [1], [5], [6], and

[10] may be combined with at least part of the configuration of at least one of the inventions [2] to [4] and [7] to [9]. Furthermore, any configuration may be extracted from the embodiments [1] to

[10] and combined. The applicant of this application intends to obtain rights to inventions including these configurations. Furthermore, even if the specification contains a phrase such as "in the case of..." or "when...," it is not intended to describe a configuration that is limited to that case or time. These are merely examples of better configurations, and the applicant intends to obtain rights to configurations that are not in these cases or times. Furthermore, any descriptions that specify an order are not limited to this order. Configurations in which some parts are deleted or the order is changed are also disclosed, and the applicant intends to obtain rights to such disclosures. [Explanation of symbols]

[0083] 100 computing devices 111 Communication Interface 112 input interface 113 Output Interface 114 processors 115 memory 116 Storage Devices 200 External Information Support System 210 Expense settlement system 220 ERP system 300 System Orchestrator 310 Connector 400 Supervisor Agent 500 Data Sources 510 User Data Source 520 System Specification Data Source 530 Internal Rules Data Source 540 Domain Knowledge Data Sources 550 internal shared data sources 560 supplier data sources 610 Communication Agent 611 Interpersonal Communication Tools 620 Data / File Loader Agent 621 Data / File Loader Tool 630 Document Analysis Agent 640 Task Planner Agent 650 Specialist Agents 700 cloud data sources 1000 Task Support System

Claims

1. An information processing method for supporting a task of expense reimbursement on an external business support system, which is an expense reimbursement system performed by a user who is an employee of a company, using a task support system including a plurality of AI agents, the task support system being implemented using at least one memory for storing instructions and at least one processor configured to execute instructions to perform operations, the method comprising: The task assistance system includes: a supervisor agent, which is at least a language model agent, that autonomously assists in the task; a first subagent, which is at least a language model agent, that communicates with the supervisor agent and assists with a part of the task by autonomously operating and communicating through an interpersonal communication tool that is a corporate email application or team messaging service in which accounts are held for each employee of the company and that can send messages to and from the user's account; a system orchestrator for managing the external business support system; a connector of the system orchestrator that communicates with the external business support system through an interface; a system specification data source relating to the task performed by the user using the external business support system; a user data source of the company necessary to identify the user, the user data source including identification information of the user and the interpersonal communication tool, the user data source being authorized to be referenced by at least the first subagent; The user logs in to the external business support system and inputs application information; the external business support system communicating information about the user and the task performed by the user to the connector; a step in which the connector converts information about the user and the task performed by the user that has occurred on the external business support system into a first instruction and transmits the first instruction to the supervisor agent; the supervisor agent using the first instruction received from the connector to reference the system specification data source to verify any deficiencies in the submission of procedures or files required to perform the task; If the supervisor agent detects a deficiency, sending a second instruction to the first subagent to communicate with the user and correct the deficiency, the second instruction including information about the user and requirements for resolving the deficiency; the first subagent using the user data source and the interpersonal communication tool to identify the user's account for the interpersonal communication tool, and creating and sending to the user a message informing the user of requirements for resolving the deficiency; An information processing method, including:

2. The user data source further includes organizational information of the company, including hierarchical relationships between the users; the second instruction includes a deadline for response or a level of importance; the first subagent uses the user data source to identify an account of the interpersonal communication tool of a related user different from the user that is necessary for the user to perform the task; The information processing method according to claim 1 , further comprising the step of creating and transmitting to said concerned user a message informing said concerned user of the requirements for resolving said deficiency and said deadline for response or said level of importance.

3. The task support system further includes a second sub-agent that handles information of a file submitted by a user; the second sub-agent is a multimodal model agent capable of recognizing at least images and text; When the supervisor agent refers to the system specification data source and determines that uploading an image as evidence is required to execute the first instruction, inquiring of the second subagent as to whether the image has been uploaded; Detecting defects when the image is uploaded, sending the second instruction to the first subagent to send a message prompting the first subagent to upload the image; 3. The information processing method according to claim 1.

4. The task assistance system further comprises an internal rules data source of the company; The information processing method according to claim 1 , wherein the supervisor agent detects a defect by referring to the in-house rule data source.

5. The task assistance system further comprises a document analysis agent as a third subagent, The third sub-agent checks whether the uploaded image satisfies the requirements for evidentiary documents by referring to the internal company rule data source, and cooperates with the supervisor agent in detecting any defects in the evidentiary documents. The information processing method according to claim 4.

6. A task support system that supports a task of expense reimbursement on an external business support system, which is an expense reimbursement system performed by a user who is an employee of a company, using a plurality of AI agents, the task support system being implemented using at least one memory that stores instructions and at least one processor configured to execute the instructions to perform operations, The task assistance system includes: A supervisor agent, which is at least a language model agent and autonomously assists with tasks; a first subagent, which is at least a language model agent, that communicates with the supervisor agent and assists in part of the task by autonomously operating and communicating through an interpersonal communication tool that is a corporate email application or team messaging service in which accounts are held for each employee of the company and that can send messages to and from the user's account; a system orchestrator for managing the external business support system; a connector that communicates with an external business support system of the system orchestrator through an interface; a system specification data source relating to the task performed by the user using the external business support system; a user data source of the company necessary to identify the user, the user data source including identification information of the user and the interpersonal communication tool, the user data source being authorized to be referenced by at least the first subagent; the connector communicates with the external business support system regarding the user and the task of the user that has occurred on the external business support system as a result of the user logging in to the external business support system and inputting application information, converts information regarding the user and the task performed by the user into a first instruction, and transmits the first instruction to the supervisor agent; the supervisor agent uses the first instruction received from the connector to consult the system specification data source to verify any deficiencies in the submission of procedures or files required to perform the task; If the supervisor agent detects a deficiency, the first sub-agent to communicate with the user to correct the deficiency. Sending a second instruction including information about the user and requirements for resolving the deficiency; The first sub-agent communicates with the user data source and the interpersonal communication using an application tool to identify the user's account of the interpersonal communication tool, and create and send to the user a message informing the user of the requirements for resolving the deficiency; Task support system.

7. The user data source further includes organizational information of the company, including hierarchical relationships between the users; the second instruction includes a deadline for response or a level of importance; The first subagent uses the user data source to Identifying an account of the interpersonal communication tool of a related user different from the user that is necessary for the performance of the task; The task assistance system according to claim 6 , wherein a message is created and transmitted to the user concerned, the message conveying the requirements for resolving the deficiency and the deadline for response or the level of importance.

8. The task assistance system includes a second sub-enterprise that handles information on files submitted by users. further includes an agent, the second sub-agent is a multimodal model agent capable of recognizing at least images and text; When the supervisor agent refers to the system specification data source and determines that uploading an image as evidence is required to execute the first instruction, inquiring of the second subagent as to whether the image has been uploaded; Detecting defects when the image is uploaded, sending the second instruction to the first subagent to send a message prompting the first subagent to upload the image; The task support system according to claim 7 .

9. The task assistance system further comprises an internal rules data source for the company; The supervisor agent refers to the in-house rules data source to detect deficiencies; The task support system according to claim 6 .

10. The task assistance system further comprises a document analysis agent as a third subagent, The third sub-agent checks whether the uploaded image satisfies the requirements for evidentiary documents by referring to the internal company rule data source, and cooperates with the supervisor agent in detecting any defects in the evidentiary documents. The task support system according to claim 9 .

Images