Information processing systems, information processing methods, programs, and storage media

Abstract

We provide technology that enables the acquisition of highly reliable processing results. [Solution] The information processing system has at least one processor perform the following actions: generate at least multiple tasks based on instruction information using a first agent, execute each of the multiple tasks using a second agent, evaluate the execution results of each of the multiple tasks using a third agent, modify one or more tasks included in the multiple tasks based on the evaluation results of each of the multiple tasks using at least one of the second or fourth agents, execute one or more of the modified tasks using at least one of the second or fourth agents, and output the processing results for the instruction information generated based on the execution of the multiple tasks. The second agent is a coding agent, and the second agent and the third agent are different agents.

Description

Technical Field

[0001] The present disclosure relates to an information processing system, an information processing method, a program, and a storage medium.

Background Art

[0002] There is a technology for executing predetermined processing such as simulation using a so-called AI (Artificial Intelligence) agent.

Prior Art Documents

Non-Patent Documents

[0003]

Non-Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] The present disclosure provides a technology capable of obtaining highly reliable processing results.

Means for Solving the Problems

[0005] An information processing system according to one aspect of this disclosure includes at least one memory for storing program code and at least one processor, wherein the at least one processor executes at least a portion of the program code, thereby causing the at least one processor to perform the following actions: generate at least multiple tasks based on instruction information using one or more first agents; execute each of the multiple tasks using one or more second agents; evaluate the execution results of each of the multiple tasks using one or more third agents; modify one or more tasks included in the multiple tasks based on the evaluation results of each of the multiple tasks using at least one or more second agents or one or more fourth agents; execute the modified one or more tasks using at least one or more second agents or one or more fourth agents; and output the processing results for instruction information generated based on the execution of the multiple tasks, wherein one or more second agents include at least a coding agent that generates code for executing tasks and executes the generated code, and the one or more second agents and one or more third agents include different agents, modifying one or more tasks includes modifying at least a portion of the code used to execute one or more tasks, and executing the modified one or more tasks includes executing the modified code. One or more second agents and one or more third agents operate using a work environment that has access to common resources, one or more second agents and one or more third agents operate using independent contexts, one or more third agents can access at least a portion of the execution results of multiple tasks through the work environment, and at least one processor executes at least a portion of the program code, causing at least one processor to repeatedly perform the evaluation of execution results, the modification of one or more tasks, and the execution of the modified one or more tasks until predetermined criteria are met. . [Brief explanation of the drawing]

[0006] [Figure 1] This is a block diagram showing an example of the overall configuration of an information processing system. [Figure 2] This flowchart shows an example of an information processing method. [Figure 3] This figure shows an example of instruction information. [Figure 4] This figure shows an example of a project plan. [Figure 5] This figure shows an example of a task list. [Figure 6] This figure shows an example of the processing result. [Figure 7]A block diagram showing an example of a computer hardware configuration. [Modes for carrying out the invention]

[0007] Hereinafter, embodiments of this disclosure will be described with reference to the accompanying drawings. In this specification and the drawings, components having substantially the same functional configuration are denoted by the same reference numerals, and redundant descriptions will be omitted.

[0008] One embodiment of this disclosure is an example of an information processing system that performs a predetermined process. The predetermined process may be any process. For example, the predetermined process may be a simulation used in the field of materials development. The information processing system acquires instruction information from a user, performs a process based on the instruction information, and outputs the processing result.

[0009] <Overview> The information processing system according to this embodiment includes a coding agent designed to autonomously and reliably perform complex, long-term tasks. The architecture of the information processing system incorporates the concept of self-reflection. This enables the coding agent to systematically handle processes such as planning, execution, and self-evaluation (self-reflection).

[0010] The information processing system according to this embodiment operates as a multi-agent system, employing a "plan-execute" pattern in which a planning agent (which may also be called a planner) formulates a plan, and multiple execution agents (which may also be called workers) execute that plan. First, the planning agent, while interacting with the user, constructs the project plan and the sequence of tasks necessary to execute it. Each task is executed by an execution agent operating within an independent context limited to that task. This design effectively avoids the constraints and pollution problems of the context window that standard coding agents often face.

[0011] For example, executing a second task immediately after completing the first task, while maintaining the same context, is similar to executing multiple tasks within a single session in a large-scale language model. The information processing system according to this embodiment is configured such that each task is assigned to an execution agent, and each agent starts a different session.

[0012] The planning and execution agents are built upon a standard coding agent, inheriting its task execution capabilities. Furthermore, the information processing system employs a self-evaluation-based self-feedback system to ensure robust long-term task execution. This self-feedback covers not only local errors during code execution but also the consistency of the resulting sequence of processes and the validity of the approach adopted to achieve the task. As a result, even if errors occur during the execution of complex long-term tasks, each agent can autonomously correct them and produce highly reliable results.

[0013] During the planning phase, the planning agent constructs the project plan and task sequence through interaction with the user. This represents a list of tasks to be performed. Each task is designed to be completed by the underlying execution agent alone, or at least almost entirely on its own. The resulting project plan and task sequence are then reviewed and approved by the user.

[0014] The information processing system moves to the next task execution phase after user approval. The information processing system is primarily designed for problems where the task set can be reasonably predicted and broken down, especially in the early stages of a project.

[0015] Once the plan is approved, the information processing system enters an autonomous task execution phase. Each execution agent utilizes self-feedback as it progresses through its respective tasks step by step. Self-feedback is also used for managing the overall project progress. Specifically, the information processing system refers to this feedback when deciding whether to proceed to the next task, and moves on to the next task only if it determines that the output of the current task has reached a sufficient quality level. For example, the information processing system has an evaluation agent (which may also be called a reviewer) generate a report showing the evaluation results of the task, and then inputs the evaluation report into a large-scale language model to determine whether it can proceed to the next task.

[0016] The information processing system will suspend a project if it determines that there are problems that cannot be resolved by correcting them at the individual task level. Typical cases include situations where the project cannot proceed because a preceding task fails to produce the expected results in the plan, or situations where the results of a task cannot meet the requirements defined in the plan even after repeated corrections. Through these mechanisms, the information processing system ensures stability and reliability even for long-term, complex tasks. If a project is suspended, users can review the agent's output and, if necessary, revise the plan or provide new instructions to resume the project from where it was suspended.

[0017] For example, a new instruction could specify a task number and instruct the user to re-execute tasks from that task number onward. Alternatively, a new instruction could not specify a task number and instruct the user to re-execute the entire project. Users may revise the plan if there are problems with it. Users may also execute each task based on feedback without revising the plan. When problems arise in the project, users are not limited to these methods and may address them in any way they wish.

[0018] The information processing system performs operations in a structured work environment. Specifically, each agent uses a working directory shared by all tasks. Files generated within a task (such as code, data, configuration files, etc.) are saved in this working directory and can be utilized in subsequent tasks. For example, the code generated in a preceding task is stored in a dedicated or shared directory within the working directory, so that subsequent tasks can easily reference or execute the code.

[0019] In addition to file-based information sharing, the results and findings of individual tasks are aggregated and passed on to related subsequent tasks. Specifically, the storage location of the output file of each task, the achieved results, the usage method of the code deliverables, the provided format of the data deliverables, etc. are recorded. As a result, subsequent tasks can effectively and smoothly utilize the results of preceding tasks. With this design, the execution agent of a subsequent task can efficiently inherit the context required for task execution without directly accessing detailed deliverables such as a large number of log files and numerous intermediate files.

[0020] The information processing system is designed to process a workflow composed of unit operations executable by a standard coding agent or task units with equivalent complexity. In such a workflow, dependencies may occur between tasks. For example, in a computational science research project, a case can be considered where a certain task evaluates parameters that bring about optimal convergence of a simulation, and the subsequent task uses the result for production execution. As another example, in a data science project, a case can be cited where a certain task compares and evaluates multiple models and hyperparameter settings using training data, the subsequent task learns the final model using the selected settings, and makes predictions for another test data set. These are merely representative usage examples, and the information processing system can handle more complex cases.

[0021] [First Embodiment] The overall configuration of the information processing system according to this embodiment will be described with reference to Figure 1. Figure 1 is a block diagram showing an example of the overall configuration of the information processing system.

[0022] As shown in Figure 1, the information processing system 1000 includes an information processing device 10. The information processing system 1000 may or may not include a terminal device operated by a user U. The information processing device 10 and the terminal device may be connected to each other via a communication network to enable data communication. The communication network may be, for example, a LAN (Local Area Network), a WAN (Wide Area Network), a VPN (Virtual Private Network), or the Internet.

[0023] The information processing device 10 is an example of an information processing device that executes a process instructed by the user U. The information processing device 10 may be a computer such as a personal computer, workstation, or server.

[0024] The information processing device 10 may include a machine learning model M. The machine learning model M may be provided by an external device that can be connected to the information processing device 10 via a communication network. The information processing device 10 may use the machine learning model M to generate code for executing a process instructed by the user U, and execute the generated code. The code generated by the machine learning model M may include, for example, commands, scripts, source code, binary code, etc. The code may be written in an interpreted development language or in a compiled development language.

[0025] Machine learning model M is an example of a machine learning model trained to perform a given task. Machine learning model M may also be a machine learning model capable of generating various types of data such as text, audio, images, and videos. Machine learning model M may be, for example, a neural network, a transformer, a large language model (LLM), a small language model (SLM), a visual language model (VLM), a generative model, or a foundational model.

[0026] The machine learning model M may support multimodal operation. That is, the machine learning model M may take at least two inputs, such as text data, audio data, image data, or video data. The machine learning model M may also output at least two outputs, such as text data, audio data, image data, or video data.

[0027] The machine learning model M may be implemented using a single machine learning model. The machine learning model M may be implemented by the collaboration of multiple machine learning models. The machine learning model M may consist of multiple machine learning models depending on the task being performed.

[0028] The information processing device 10 may include a planning agent AG1, an execution agent AG2, an evaluation agent AG3, and a correction agent AG4. However, the information processing device 10 does not have to include a correction agent AG4. The information processing device 10 may also include an execution agent AG2 that also functions as a correction agent AG4. Hereinafter, the multiple agents included in the information processing device 10 may be collectively referred to as "each agent".

[0029] Planning agent AG1 may be a software agent that plans a project to execute processes instructed by user U. Planning agent AG1 may be an AI agent that uses a machine learning model M. Planning agent AG1 may or may not be a coding agent.

[0030] Planning agent AG1 may generate project information. Project information may include a project plan and a task list. The project plan may also be information that shows a plan for executing a predetermined process. The project plan may include information that needs to be known in order to execute the predetermined process. For example, the project plan may include the purpose of the process, the final goal, the background, the expected results, the processing method, etc.

[0031] A task list may include one or more tasks for performing a predetermined process. A task list may include tasks for carrying out a project plan. A task list may include tasks for generating code. A task list may include tasks that do not generate code. A task list may include tasks for executing code. A task list may include tasks for using an external tool T. A task list may include tasks for generating libraries for performing a predetermined process. A task list may include tasks for verifying external tool T. A task list may include tasks for visualizing processing results (in other words, tasks for presenting processing results to user U). A task list may include tasks for which processing results are not visualized (in other words, tasks for which processing results are not presented to user U). A task list may include tasks for which memory space is specified. A task list may include tasks for which memory space is not specified.

[0032] A task list may be information that associates the content of a task with a task number. The task number may include information indicating the execution order of the tasks. The association between the task content and the task number may, for example, be stored as a set of one piece of information and the other piece of information, or as information that allows the other piece of information to be retrieved from the other piece of information, or as a set of identification information for one piece of information and the other piece of information. The information that associates the task content and the task number may, for example, be text data in which the text describing the task and the task number are separated by a predetermined symbol. Another example of the information that associates the task content and the task number may be a table that stores records containing the text describing the task and the task number.

[0033] Execution agent AG2 may be a software agent that executes tasks. Execution agent AG2 may execute one or more tasks included in the task list. Execution agent AG2 may be an AI agent that uses a machine learning model M. Execution agent AG2 may be a coding agent.

[0034] The execution agent AG2 may perform multiple tasks based on the same machine learning model M. Alternatively, the execution agent AG2 may perform each of the multiple tasks based on a different machine learning model M.

[0035] The execution agent AG2 may generate code to execute a task. The execution agent AG2 may generate code to execute a task by inputting project information and information indicating the task to be executed (e.g., task number) into a machine learning model M. The execution agent AG2 may execute the code to execute the task. The execution agent AG2 may generate a report showing the results of the task execution (hereinafter also referred to as the "execution report").

[0036] The evaluation agent AG3 may be a software agent that evaluates tasks. The evaluation agent AG3 may evaluate the execution results of tasks. The evaluation agent AG3 may evaluate the task execution report. The evaluation agent AG3 may be an AI agent that uses a machine learning model M. The evaluation agent AG3 may be a coding agent, or it may not be a coding agent.

[0037] The evaluation agent AG3 may evaluate the execution results of one or more tasks performed by the execution agent AG2. The evaluation agent AG3 may evaluate the execution results of some of the tasks among the one or more tasks performed by the execution agent AG2. In other words, the one or more tasks performed by the execution agent AG2 may include tasks that are not evaluated by the evaluation agent AG3.

[0038] The evaluation agent AG3 may evaluate the results of multiple tasks based on the same machine learning model M. Alternatively, the evaluation agent AG3 may evaluate the results of each of the multiple tasks based on different machine learning models M.

[0039] The evaluation agent AG3 may generate a report showing the evaluation results (hereinafter also referred to as the "evaluation report"). The evaluation agent AG3 may evaluate a task by inputting project information and the task execution results into a machine learning model M. The evaluation agent AG3 may generate an evaluation report for a task by inputting project information and the task execution report into a machine learning model M.

[0040] The correction agent AG4 may be a software agent that corrects a task based on the task's evaluation results. The correction agent AG4 may also correct the code for executing the task. The correction agent AG4 may also be an AI agent that uses a machine learning model M. The correction agent AG4 may also be a coding agent. The correction agent AG4 may generate corrected code by inputting project information and a task evaluation report into the machine learning model M. The correction agent AG4 may execute the corrected code. The correction agent AG4 may also generate an execution report showing the task execution results.

[0041] Correction agent AG4 may correct one or more tasks executed by execution agent AG2. Correction agent AG4 may correct some of the tasks among the one or more tasks executed by execution agent AG2. In other words, the one or more tasks executed by execution agent AG2 may include tasks that are not corrected by correction agent AG4.

[0042] The correction agent AG4 may correct multiple tasks based on the same machine learning model M. Alternatively, the correction agent AG4 may correct each of the multiple tasks based on a different machine learning model M.

[0043] The coding agent may also be a series of information processing functions that generate and execute code based on input instruction information. For example, the coding agent may analyze the instruction information using a machine learning model. The coding agent may generate one or more tasks based on the results of the analysis of the instruction information. The coding agent may generate code to execute the tasks using a machine learning model. The coding agent may control the execution of the code and obtain the results of the code execution. The coding agent may evaluate the results of the code execution. Based on the evaluation of the execution results, the coding agent may control at least one of the following: regenerating the tasks, regenerating the code, or re-executing the code.

[0044] At least two of the planning agent AG1, execution agent AG2, evaluation agent AG3, and correction agent AG4 may operate using the same machine learning model M. Alternatively, all of the planning agent AG1, execution agent AG2, evaluation agent AG3, and correction agent AG4 may operate using the same machine learning model M. Furthermore, the planning agent AG1, execution agent AG2, evaluation agent AG3, and correction agent AG4 may each operate using different machine learning models M.

[0045] The execution agent AG2 and the evaluation agent AG3 may be different agents. Being different agents may include, for example, having different system prompts input to the machine learning model M. The system prompt may also be the initial message input to the machine learning model M for a given agent. For example, the system prompt may include a message to determine the behavior of the machine learning model M.

[0046] Multiple agents being different agents may include, for example, being able to invoke different tools. Multiple agents being different agents may also include operating using independent contexts. Multiple agents being different agents may also include performing different roles. Two agents performing different roles may include, for example, switching roles by entering a single system prompt corresponding to multiple roles for the first and second agents, and then entering an instruction to act as the first agent or an instruction to act as the second agent.

[0047] The execution agent AG2 and the evaluation agent AG3 may operate using independent contexts. A context may be a set of information that the machine learning model M can access within a processing unit (hereinafter referred to as a session) which is composed of a series of input and output exchanges using the machine learning model M. The context may include input data to the machine learning model, output data from the machine learning model M, and data generated based on the output data from the machine learning model M. Accessible information may also include historical information of input data, output data, and generated data. Accessible information may include some or all of the series of input data to the machine learning model M. Accessible information may also include some or all of the series of output data from the machine learning model M. Accessible information may also include internal information such as memory information, cache information, and intermediate generation information used when the agent operates. A context may or may not have a limit on the amount of information it can hold. Note that an independent context may mean that multiple contexts do not have dependencies on each other. Furthermore, an independent context may mean a configuration in which each agent operates in an individual session, and historical information from other agents' sessions is not automatically referenced or inherited. In this case, this does not preclude obtaining information through data stored in the work environment. Furthermore, an independent context may include the fact that each agent's individual session is logically or physically isolated.

[0048] The execution agent AG2 and the evaluation agent AG3 may operate using a common work environment that allows them to access shared resources. The common work environment may include a common directory. The work environment that allows access to shared resources may also be an environment in which the evaluation agent AG3 can access the resources used by the execution agent AG2 when it was running. For example, the execution agent AG2 and the evaluation agent AG3 may operate on different devices, with the resources used by the execution agent AG2 being uploaded to a specific work environment, and the evaluation agent AG3 downloading the resources used by the execution agent AG2 from that specific work environment.

[0049] If the execution agent AG2 and the evaluation agent AG3 are different agents, the execution agent AG2 and the evaluation agent AG3 may operate as follows: For example, the evaluation agent AG3 may operate without using the history information used in the operation of the execution agent AG2. For example, the evaluation agent AG3 may operate using some of the history information used in the operation of the execution agent AG2, but without using other history information. For example, the evaluation agent AG3 may operate using information obtained by processing the history information used in the operation of the execution agent AG2. The information obtained by processing the history information may include, for example, information that summarizes the history information, information generated based on the history information, etc. For example, the execution agent AG2 and the evaluation agent AG3 may operate using different history information. For example, the evaluation agent AG3 may operate without using internal information such as memory information, cache information, and intermediate generation information that was used when the execution agent AG2 was operating.

[0050] Planning agent AG1 and execution agent AG2 may be different agents. Planning agent AG1 and execution agent AG2 may operate using the same context or using independent contexts. Planning agent AG1 and evaluation agent AG3 may be different agents. Planning agent AG1 and evaluation agent AG3 may operate using the same context or using independent contexts.

[0051] An execution agent AG2 may execute one or more tasks using independent contexts. For example, one execution agent AG2 executing one or more tasks may execute each task using an independent context. Alternatively, one or more execution agents AG2, each using an independent context, may execute each task. An execution agent AG2 may execute all tasks included in the task list using independent contexts. An execution agent AG2 may execute one or more tasks included in the task list using independent contexts.

[0052] The evaluation agent AG3 may evaluate the execution results of one or more tasks using independent contexts. For example, one evaluation agent AG3 that evaluates the execution results of one or more tasks may evaluate the execution results of each task using independent contexts. Alternatively, one or more evaluation agents AG3, each using an independent context, may evaluate the execution results of each task. The evaluation agent AG3 may evaluate the execution results of all tasks included in the task list using independent contexts. The evaluation agent AG3 may evaluate the execution results of some of the tasks included in the task list using independent contexts.

[0053] The information processing device 10 may include an external tool T. The planning agent AG1, execution agent AG2, evaluation agent AG3, and correction agent AG4 may operate using the external tool T. The planning agent AG1, execution agent AG2, evaluation agent AG3, and correction agent AG4 may use a machine learning model M to generate code that includes commands for using the external tool T. The planning agent AG1, execution agent AG2, evaluation agent AG3, and correction agent AG4 may generate code that runs the external tool T in the background and obtains its execution results.

[0054] External tool T may include various tools for performing predetermined processes. External tool T may include tools capable of reading and writing files, operating command lines, searching databases, searching the web, or communicating with external devices. For example, when running a simulation, external tool T may include simulation software or a library capable of running simulations. External tool T may include tools that can obtain information on how to use other tools. For example, external tool T may include tools that can investigate the interface of a simulation or library.

[0055] The external tool T may be an external device that can be connected to from the information processing device 10 via a communication network. The information processing device 10 may use the external tool T via a standardized protocol. For example, the information processing device 10 may use the external tool T using an MCP (Model Context Protocol) server. The MCP server may be an external device that can be connected to from the information processing device 10 via a communication network. The external device that provides the external tool T and the external device that provides the MCP server may be the same device or different devices.

[0056] Each agent may consist of software and a machine learning model M. The information processing device 10 may input input information to each agent and obtain a response from each agent. In the information processing device 10, one or more processors may execute the machine learning model M. One or more processors may execute an external tool T specified by a tool call. The tool call may be generated based on the output from the machine learning model M. One or more processors executing an agent may include one or more processors executing processing using the machine learning model M included in the agent, executing the external tool T specified by a tool call, executing the software included in the agent, etc.

[0057] The information processing device 10 may include a global workspace. The global workspace may be a work environment for executing predetermined processes. The global workspace may include a storage area for structuring and storing data used to execute predetermined processes. The storage area may be, for example, a directory or folder created on an auxiliary storage device. The global workspace may include a storage device that can be connected to from the information processing device 10 via a communication network.

[0058] Each agent provided by the information processing device 10 may operate using a common global workspace. The global workspace may include the memory areas used by each agent. The memory areas used by each agent may be different memory areas (directories or folders).

[0059] The global workspace may include memory areas for each task included in the task list. The global workspace may also include memory areas used by each agent. The global workspace may also include memory areas for storing electronic data that is commonly used by each agent. Examples of electronic data commonly used by each agent may include programs, libraries, tools, system files, etc.

[0060] Each agent may be given information indicating the memory areas used by other agents. Each agent may be able to access the memory areas used by other agents. Each agent may obtain the operational results of other agents by referring to the memory areas used by other agents.

[0061] For example, the global workspace may be a storage area with a predetermined directory structure. For example, the global workspace may be a storage area that stores the following information. The global workspace may include directories corresponding to each task. Each task's directory may contain verification scripts or input files, etc. Each task's directory may contain task-specific outputs. The global workspace may include directories for storing deliverables shared between tasks. The global workspace may include directories for storing reports on past tasks. The global workspace may include directories for storing execution reports for each task. The global workspace may include directories for storing evaluation reports for each task. The global workspace may include information that specifies references to external information. External information may be information used for task execution. External information may include, for example, web content or research paper data, etc.

[0062] The information processing device 10 may construct each agent by inputting one or more prompts to the machine learning model M. A prompt is an example of input information to be input to the machine learning model M in order to give instructions to the machine learning model M. A prompt may be text data, electronic data representing text, audio, still images, or video, or multimodal electronic data containing at least two of the following: text, audio, still images, or video.

[0063] The prompts for building an agent may include system prompts. System prompts may also be prompts for giving constant instructions to the machine learning model M. System prompts may also be prompts common to each agent.

[0064] The system prompt may include information about the global workspace. This information may include a description of the directory structure of the global workspace. It may also include a description of each directory. Furthermore, it may include information about the storage area used by each agent. Finally, it may include information about the data to be stored in each directory. By specifying the storage area for data via the system prompt, the global workspace can be maintained in an organized state that makes it easy to understand the stored data.

[0065] The system prompt may include information specifying the development language in which the code is written. The system prompt may include information specifying executable commands. The system prompt may include information specifying required commands. The system prompt may include information indicating where to refer to specific information. The system prompt may include information indicating the role of each agent. The system prompt may include information indicating that project information includes a project plan and a task list. The system prompt may include information indicating how to use the external tool T. The system prompt may include information specifying the behavior when the simulation is run.

[0066] The actions during simulation execution may include running the simulation in the background. The actions during simulation execution may include controlling the MCP server to run the simulation software in the background. The actions during simulation execution may include starting the simulation in the background and then executing subsequent processes before the simulation is completed.

[0067] For example, simulations for materials development often take a long time to run. If the simulation is not progressing as expected, waiting for it to complete results in idle time, which is inefficient. Running the simulation in the background allows you to refer to the intermediate results and detect problems that occur during the simulation early on.

[0068] The prompts for building an agent may include agent-specific prompts. These agent-specific prompts may differ for each agent. The agent-specific prompts may also include information common to all agents.

[0069] Agent-specific prompts may include information indicating the agent's role. An agent's role may include at least one of the following: planning agent, execution agent, evaluation agent, or correction agent. Agent-specific prompts may also include information indicating required output information. Agent-specific prompts may also include information indicating the format of the output information. At least one of the following information may be common to each agent's prompt: information indicating the agent's role, information indicating required output information, or information indicating the format of the output information.

[0070] The prompts from planning agent AG1 may include information indicating how to handle situations where there is missing information in the instruction information. The prompts from planning agent AG1 may also include information indicating perspectives for generating an appropriate project plan. An appropriate project plan may be, for example, a project plan that can reduce processing failures. Processing failures may include, for example, errors occurring during processing, failure to obtain the correct processing result, abnormal termination of processing, or failure to terminate processing.

[0071] The prompts from the execution agent AG2 may include information indicating the flow of task execution. For example, the task execution flow may include reading the project plan, reading the task list, recognizing the tasks it should perform, obtaining the execution results of related tasks, generating code to execute the tasks, and then executing the tasks using the code.

[0072] The prompts of the evaluation agent AG3 may include information indicating the evaluation method. This information may include, for example, information indicating the data to be evaluated, information indicating the data to be referenced during the evaluation, and information indicating the tools to be used for the evaluation. The prompts of the evaluation agent AG3 may also include information indicating the evaluation perspectives. The prompts of the evaluation agent AG3 may also include information indicating the format of the evaluation results.

[0073] The information processing device 10 may output processing results generated based on the execution of one or more tasks. The processing results output by the information processing device 10 may include at least the processing results for instructions from user U. For example, if user U inputs an instruction such as "I want to know the size dependence of the melting point of gold nanoparticles," the information processing device 10 may output the results of analyzing "the size dependence of the melting point of gold nanoparticles." The processing results output by the information processing device 10 may also include the results of additional tasks. Additional tasks may include tasks that generate graphs visualizing the processing results.

[0074] Outputting processing results generated based on the execution of one or more tasks may include presenting the processing results to user U in any way. Outputting processing results generated based on the execution of one or more tasks may include storing the processing results in a predetermined storage location. Outputting processing results generated based on the execution of one or more tasks may include outputting the results to other configurations of the device or to other devices. Outputting processing results generated based on the execution of one or more tasks may include storing the processing results in a predetermined area. Outputting processing results generated based on the execution of one or more tasks may include outputting the processing results to the display of the device or another device, or including processing for outputting the processing results to the display of the device or another device.

[0075] The information processing device 10 may include at least one of each agent. The information processing device 10 may execute multiple agents in parallel. For example, the information processing device 10 may include multiple evaluation agents AG3. The information processing device 10 may include a first evaluation agent AG3 that evaluates a first task, and a second evaluation agent AG3 that evaluates a second task. The first evaluation agent AG3 and the second evaluation agent AG3 may be the same agent or different agents. The information processing device 10 may include multiple evaluation agents AG3 that evaluate a single task. When there are multiple evaluation agents AG3, for example, the reliability of autonomous task execution by the agents can be dramatically improved by self-detection and self-correction of errors, while overcoming the context length limitations of the machine learning model M.

[0076] User U may separately execute the processes performed by each agent. For example, User U may execute programs, code, scripts, etc., stored in the global workspace. User U may also modify at least a portion of the programs, code, scripts, etc., stored in the global workspace and then execute the modified programs, code, scripts, etc.

[0077] Note that the overall configuration of the information processing system 1000 shown in Figure 1 is just one example, and various system configurations are possible depending on the application and purpose. The information processing system 1000 may consist of one or more devices. Each device included in the information processing system 1000 may be part of a system composed of multiple devices. Each function included in the information processing system 1000 may be implemented by any device that constitutes the system. Each component included in the information processing system 1000 may be included in any device that constitutes the system.

[0078] The information processing system 1000 may include multiple information processing devices 10. The information processing device 10 may be one or more information processing devices, or it may be an information processing system including multiple information processing devices. The information processing device 10 may be implemented by one or more computers, or it may be implemented as a cloud computing service. The information processing system 1000 may be implemented by a standalone computer in which the information processing device 10 and a terminal device are integrated. The terminal device may be implemented by other devices included in the information processing system 1000. The terminal device may be implemented by an external information processing device or information processing system of the information processing system 1000. Note that "external" means not included in the information processing system 1000. The classification of devices, such as the information processing device 10 shown in Figure 1, is just one example.

[0079] <Processing Procedures for Information Processing Systems> The information processing method performed by the information processing system 1000 will be explained with reference to Figure 2. Figure 2 is a flowchart showing an example of an information processing method.

[0080] The information processing method shown in Figure 2 may be implemented by having at least one processor in the information processing device 10 execute at least a portion of program code stored in one or more memories. The program code may include, for example, software, code, programs, scripts, binaries interpretable and executable by the processor, data interpretable and executable by the processor, etc. The program code may also include, for example, code generated by each agent. The program code may be distributed and stored in one or more memories. The program code may be stored in one or more storage media, read into one or more memories in the information processing device 10, and executed by one or more processors. Furthermore, at least a portion of the program code may be stored in at least one of the memories in one or more memories in the information processing device 10 and executed by at least one of the processors.

[0081] In step S1, the information processing device 10 acquires instruction information from user U. The information processing device 10 may, for example, receive instruction information from user U using an interactive user interface. The interactive user interface may, as an example, be a chat-style screen. The chat-style screen may have a display area in which input information (e.g., instructions, etc.) and output information (e.g., responses, etc.) are displayed in chronological order. Another example of an interactive user interface is a voice interaction using voice input via a microphone and voice output via a speaker.

[0082] The instruction information may include information indicating the process to be executed by the information processing device 10. The instruction information only needs to indicate a process that the coding agent can execute. The instruction information may also be information indicating a simple instruction. A simple instruction may include instructions with a small amount of information, instructions that indicate only the information to be obtained, instructions that do not indicate specific processing methods, instructions that do not indicate detailed processing conditions, instructions that do not indicate some of the information used for processing, etc. The instruction information may also be information expressed in natural language. For example, the instruction information may be information such as "I want to perform a molecular simulation of xxx" or "Use xxx as training data to create a new machine learning model."

[0083] The instruction information may include information indicating processing specific to a particular field. The instruction information may also be instruction information relating to fields to which the external tool T provided by the information processing device 10 can be applied. The information processing device 10 may be configured to execute processing specific to a particular field with high accuracy by providing one or more external tools T specific to a particular field.

[0084] In step S2, the information processing device 10 generates project information. The information processing device 10 may generate project information based on the instruction information acquired in step S1. The information processing device 10 may generate project information by executing the planning agent AG1. The information processing device 10 may input the instruction information acquired in step S1 to the planning agent AG1. The information processing device 10 may input system prompts and prompts for the planning agent AG1 to the planning agent AG1.

[0085] Planning agent AG1 may generate project information based on instruction information. Planning agent AG1 may generate project information while referring to the global workspace. Planning agent AG1 may search for data stored in the global workspace. For example, planning agent AG1 may search for data to be processed, existing processing results, available external tools T, etc. Planning agent AG1 may generate project information that assumes the use of data stored in the global workspace. Existing processing results may include processing results generated from processes executed in the past.

[0086] Planning agent AG1 may generate checkpoint tasks. Checkpoint tasks may be specified by system prompts. Checkpoint tasks may also be specified by instruction information. For example, planning agent AG1 may generate checkpoint tasks when user U enters an instruction such as "Please check the results of xxx along the way."

[0087] A checkpoint task may be a task for verifying a processing method or processing result. A checkpoint task may also be a task for determining whether to proceed to a subsequent task based on the execution result of a preceding task. A checkpoint task may include a verification task for an external tool T. A verification task for an external tool T may include a lightweight task that uses the external tool T. A verification task for an external tool T may include a task that uses the external tool T with a short processing time. A verification task for an external tool T may include a task that runs the external tool T using verification data. For example, when using an external tool T whose operation method is unknown, the verification task for the external tool T may include a task that runs the external tool T using test data.

[0088] For example, checkpoint tasks may include tasks that test a short simulation before running a long simulation. Short processing time tasks may also include tasks that run a simulation from start to finish with a reduced number of steps. For example, short processing time tasks may include tasks that run a materials simulation to analyze the properties of a specific atom, but with a smaller atomic structure being analyzed. For example, short processing time tasks may include tasks that run an external tool T with inputs or conditions that reduce computational cost.

[0089] Planning agent AG1 may output the generated project information. The project information may include a project plan and a task list. Planning agent AG1 may store the generated project information in the global workspace. Planning agent AG1 may store the project information in a memory area accessible to other agents. Planning agent AG1 may store the project information in a memory area for planning agent AG1. The memory area for storing the project information may be specified by a system prompt.

[0090] In step S3, the information processing device 10 determines whether or not to modify the project information. The information processing device 10 may present the project information generated in step S2 to the user U. The information processing device 10 may output a screen displaying the project information to a display device provided by the information processing device 10. Presenting information to the user may include the processor performing at least a part of the processing necessary to display the information on the display device. The display device may be provided in the same device as the processor, or in a different device from the processor. There may be multiple display devices.

[0091] The information processing device 10 may accept modifications to project information by user U. For example, the information processing device 10 may present the project plan and task list to user U and accept modification instructions from user U to the planning agent AG1. Alternatively, for example, the information processing device 10 may allow user U to directly edit the project information on the screen that displays the project information. The information processing device 10 does not have to accept modifications by user U for one or more tasks included in the task list.

[0092] The information processing device 10 may receive feedback information from user U. The information processing device 10 may receive feedback information from user U, for example, using an interactive interface. The information processing device 10 may receive feedback information from user U, for example, through a screen displaying project information. The feedback information from user U may include any information that user U wishes to have corrected regarding the project information. For example, the feedback information from user U may include pointing out that steps necessary for preparing the simulation are missing, instructing the addition of a task called a checkpoint, or instructing the change of a method used in the project plan to a desired method. The information processing device 10 does not have to receive feedback information from user U for some of the tasks included in the task list.

[0093] The information processing device 10 may decide to modify the project information if it receives a modification request from user U or feedback information from user U. The information processing device 10 may decide not to modify the project information if it does not receive a modification request from user U or feedback information from user U.

[0094] The information processing device 10 may automatically determine whether or not to modify the project information. For example, the project information generated by the planning agent AG1 may be evaluated by other agents to determine whether or not to modify the project information. The information processing device 10 may also instruct the evaluation agent AG3 to evaluate and modify the project information.

[0095] If it is determined that the project information should be modified (YES), the information processing device 10 proceeds to step S4. On the other hand, if it is determined that the project information should not be modified (NO), the information processing device 10 proceeds to step S5.

[0096] In step S4, the information processing device 10 modifies the project information. The information processing device 10 may generate modified project information by reflecting the modifications received in step S3 to the project information generated in step S2. The information processing device 10 may regenerate the project information based on the feedback information received in step S3. The information processing device 10 may modify the project information by executing the planning agent AG1. The information processing device 10 may generate modified project information by inputting the feedback information received in step S3 to the planning agent AG1.

[0097] In step S5, the information processing device 10 obtains a task list from the project information. The information processing device 10 may obtain a task list from the project information generated in step S2. The information processing device 10 may obtain a task list from the project information modified in step S4. The information processing device 10 may obtain one task from the unprocessed tasks included in the task list. The information processing device 10 may obtain the task with the earliest execution order from the unprocessed tasks.

[0098] The information processing device 10 may obtain multiple tasks from the unprocessed tasks included in the task list that can be executed in parallel. Multiple tasks that can be executed in parallel may be, for example, two or more tasks that are not dependent on each other. A dependency is a relationship in which one task requires the execution result of another task. For example, if the first task and the second task are not dependent on each other, but the first task and the third task are dependent on each other, then the first task and the second task can be executed in parallel, but the first task and the third task cannot be executed in parallel.

[0099] The information processing device 10 executes one or more tasks obtained from the task list. The information processing device 10 may also execute one or more tasks by executing the execution agent AG2. The information processing device 10 may execute multiple tasks in parallel by simultaneously executing multiple execution agents AG2. Task assignment to the execution agents AG2 may be performed by a control program pre-installed on the information processing device 10. The control program may be a different program from each agent provided by the information processing device 10. Whether or not to execute multiple tasks in parallel may be determined by the planning agent AG1. Whether or not to execute multiple tasks simultaneously may be indicated in the task list.

[0100] The information processing device 10 may input the project plan, task list, and task number to the execution agent AG2. The information processing device 10 may also obtain the information necessary for task execution from the global workspace and input it to the execution agent AG2. For example, if the information processing device 10 executes a specific task after other tasks have been completed, it may obtain the execution reports of the other tasks from the global workspace and input them to the execution agent AG2. The execution agent AG2 may autonomously obtain the information necessary for task execution from the global workspace.

[0101] Execution agent AG2 may execute tasks identified by task numbers based on the project plan, task list, and task numbers. Execution agent AG2 may also execute tasks using other information obtained from the global workspace. Execution agent AG2 may execute tasks identified by task numbers after recognizing the context of the tasks to be executed based on the project plan and task list.

[0102] The execution agent AG2 may generate code to execute a task. The execution agent AG2 may execute the generated code. The execution agent AG2 may execute a task without generating code. Tasks that do not require code execution may include, for example, a task to prepare a directory.

[0103] Execution agent AG2 may output the task execution results. Execution agent AG2 may generate an execution report showing the task execution results. Execution agent AG2 may store the task execution results or execution report in the global workspace. Execution agent AG2 may store the task execution results or execution report in a storage area accessible by other agents. Execution agent AG2 may store the task execution results or execution report in a storage area for execution agent AG2. The storage area for storing the task execution results or execution report may be specified by a system prompt.

[0104] In step S6, the information processing device 10 evaluates the execution result of the task. The information processing device 10 may evaluate the intermediate results of the task while the task is being executed. That is, step S6 may be executed in parallel with step S5.

[0105] The information processing device 10 may evaluate the execution results of tasks by executing the evaluation agent AG3. The information processing device 10 may evaluate the execution results of multiple tasks in parallel by executing multiple evaluation agents AG3 simultaneously. The assignment of tasks to the evaluation agents AG3 may be performed by a control program. Whether or not to evaluate multiple tasks in parallel may be decided by the planning agent AG1. Whether or not to evaluate multiple tasks simultaneously may be indicated in the task list.

[0106] The information processing device 10 may input the project plan, task list, task number, and instructions to confirm the task execution results to the evaluation agent AG3. The information processing device 10 may also acquire the information necessary for task evaluation from the global workspace and input it to the evaluation agent AG3. The evaluation agent AG3 may autonomously acquire the information necessary for task evaluation from the global workspace.

[0107] The evaluation agent AG3 may identify the memory area where the execution results of a task are stored based on the task number. If the task identified by the task number has dependencies on other tasks, the evaluation agent AG3 may retrieve the execution results of the other tasks. The evaluation agent AG3 may identify the memory area where the execution results of the other tasks are stored based on the task numbers of those other tasks.

[0108] The evaluation agent AG3 may evaluate a task based on its execution results. The evaluation agent AG3 may evaluate a task based on pre-configured evaluation criteria. The evaluation criteria may be indicated in the evaluation agent AG3's prompts. The evaluation criteria may be set by user U. The evaluation criteria may be generated by other agents. For example, the evaluation agent AG3 may evaluate a task according to evaluation criteria generated by planning agent AG1. The evaluation agent AG3 may determine whether a task is completed or not. For example, the evaluation agent AG3 may evaluate "what the status of the task is, whether the task is completed and the next task can be started, and if not completed, what corrections are needed."

[0109] The evaluation agent AG3 may generate code to evaluate the task. The evaluation agent AG3 may generate code to determine whether the execution result is correct or not. The evaluation agent AG3 may execute the generated code. The evaluation agent AG3 may execute the code generated by the execution agent AG2. The evaluation agent AG3 may store the generated code and the execution result of the code in the global workspace.

[0110] The evaluation agent AG3 may output the evaluation results of a task. The evaluation agent AG3 may generate an evaluation report showing the evaluation results of a task. The evaluation agent AG3 does not have to generate an evaluation report for some of the tasks included in the task list. The evaluation agent AG3 may store the evaluation results or evaluation report of a task in the global workspace. The evaluation agent AG3 may store the evaluation results or evaluation report of a task in a storage area accessible to other agents. The evaluation agent AG3 may store the evaluation results or evaluation report of a task in a storage area for the evaluation agent AG3. The storage area for storing the evaluation results or evaluation report of a task may be specified by a system prompt.

[0111] In step S7, the information processing device 10 determines whether or not to modify the task. The information processing device 10 may determine whether or not to modify the task based on the evaluation report generated in step S6. The information processing device 10 may determine whether or not the current task is completed based on the evaluation report. The information processing device 10 may determine whether or not the current task needs to be modified based on the evaluation report. The information processing device 10 may determine whether or not to proceed to the next task based on the evaluation report. For example, the information processing device 10 may have the evaluation agent AG3 determine whether or not the current task is completed. For example, the information processing device 10 may have the user U determine whether or not the current task is completed. For example, the information processing device 10 may present the evaluation report to the user U and determine whether or not the current task is completed based on the information entered by the user U in response to the evaluation report.

[0112] The information processing device 10 may determine whether or not to modify a task based on a machine learning model. The information processing device 10 may also determine whether or not to modify a task by inputting an evaluation report into the machine learning model. The machine learning model may be machine learning model M, or it may be another machine learning model different from machine learning model M (for example, a large-scale language model). The information processing device 10 does not have to determine whether or not to modify a task for one or more tasks included in the task list.

[0113] The information processing device 10 may determine whether or not to stop the execution of a task. For example, the information processing device 10 may determine to stop the execution of a task if the task has not been completed for a predetermined number of consecutive times. The information processing device 10 may determine whether or not to stop the execution of a task based on a machine learning model M. For example, the information processing device 10 may input project information and the execution results of each task into the machine learning model M, causing the machine learning model M to determine whether or not to stop the execution of a task. The information processing device 10 may also cause the user U to determine whether or not to stop the execution of a task. For example, the information processing device 10 may present project information and the execution results of each task to the user U and determine whether or not to stop the execution of a task based on the information entered by the user U. If the information processing device 10 determines to stop the execution of a task, it may notify the user U that the current task cannot be completed. If the information processing device 10 determines to stop the execution of a task, it may terminate processing without executing tasks after the current task.

[0114] In step S8, the information processing device 10 modifies the task. The information processing device 10 may modify the task by executing the modification agent AG4. The information processing device 10 may also modify the task by executing the execution agent AG2. In this embodiment, a configuration in which the task is modified by the modification agent AG4 is described, but the modification agent AG4 may be replaced with the execution agent AG2. The information processing device 10 does not have to modify one or more tasks included in the task list.

[0115] The information processing device 10 may input the project plan, task list, task number, instructions to confirm the execution results of the task, and instructions to confirm the evaluation results of the task to the correction agent AG4. The information processing device 10 may also obtain the information necessary for task correction from the global workspace and input it to the correction agent AG4. The correction agent AG4 may autonomously obtain the information necessary for task correction from the global workspace.

[0116] The correction agent AG4 may identify the memory area where the execution results of a task are stored based on the task number. The correction agent AG4 may also identify the memory area where the evaluation results of a task are stored based on the task number. If the task identified by the task number has dependencies on other tasks, the correction agent AG4 may obtain the execution results and evaluation results of the other tasks. The correction agent AG4 may also identify the memory area where the execution results and evaluation results of the other tasks are stored based on the task numbers of the other tasks.

[0117] Correction agent AG4 may correct the task identified by the task number based on the task execution results and task evaluation results. Correction agent AG4 may also correct any information related to task execution. The information to be corrected may include, for example, the code for executing the task, the data to be processed, the external tool T to be used, how to use the external tool T, the file arrangement, etc. Project plans and task lists may be excluded from correction by correction agent AG4.

[0118] The correction agent AG4 may decide not to correct the task after reviewing the task execution results and the task evaluation results. If the task is not corrected, the correction agent AG4 may instruct the execution agent AG2 to execute the next task.

[0119] The correction agent AG4 may execute the corrected task. The correction agent AG4 may output the execution results of the corrected task. The correction agent AG4 may generate an execution report showing the execution results of the corrected task. The correction agent AG4 may store the execution results or execution report of the corrected task in the global workspace. The correction agent AG4 may store the execution results or execution report of the corrected task in a storage area accessible to other agents. The correction agent AG4 may store the execution results or execution report of the corrected task in a storage area for the correction agent AG4. The storage area for storing the execution results or execution report of the corrected task may be specified by a system prompt.

[0120] The information processing device 10 may repeatedly execute steps S5 to S8 for each task included in the task list. The information processing device 10 may repeatedly execute steps S5 to S8 based on predetermined criteria such as the number of executions, determination of whether task modification is necessary based on the evaluation report, determination by user U, evaluation value, and determination by a machine learning model. If the task list includes multiple tasks that can be executed in parallel, the information processing device 10 may execute steps S5 to S8 multiple times in parallel. In other words, for each of the one or more tasks for executing the processing instructed by user U, the information processing device 10 may repeatedly execute one or more tasks and evaluate one or more tasks. In this case, the information processing device 10 may additionally perform modification of one or more tasks and re-execution of one or more tasks depending on the evaluation result. In this way, the information processing device 10 may sequentially execute one or more tasks for executing the processing instructed by user U in the order shown in the task list.

[0121] A single task may be completed by the execution agent AG2 (step S5) and evaluation agent AG3 (step S6). A single task may be completed by the execution agent AG2 (step S5), evaluation agent AG3 (step S6), correction agent AG4 (step S8), and re-evaluation agent AG3 (step S6). A single task may be completed by repeating the process of correction agent AG4 (step S8) and re-evaluation agent AG3 (step S6) multiple times until the task is completed.

[0122] In step S9, the information processing device 10 outputs the processing result. The processing result may include the execution result of the processing based on the instruction information obtained in step S1. The processing result may include the execution result of at least the last task. The processing result may include the execution results of all tasks included in the task list.

[0123] The processing results may include data generated during processing. The data generated during processing may include data output by each task. The data generated during processing may include reports showing the execution results of each task. The data generated during processing may also include reports showing the evaluation results of each task.

[0124] The processing results may include information used in the processing. The information used in the processing may include data entered into each task. The information used in the processing may include information indicating the external tool T used. The information used in the processing may include information entered into the external tool T. The information used in the processing may include information output from the external tool T.

[0125] By configuring it as described above, the information processing system 1000 can achieve the following functions.

[0126] Each agent can autonomously use tools such as file reading and writing, command line operations, and web searches to perform tasks. For example, each agent can autonomously investigate the working directory, generate code, execute it, check the results, and make further modifications. The information processing system 1000 can execute extremely complex processes by using these coding agents, which are capable of performing such complex operations, as components and controlling them appropriately. For example, the information processing system 1000 can stably and accurately execute highly difficult processes that standard coding agents might fail to complete, such as processes requiring hundreds of steps or long processes that take several days to complete.

[0127] The execution agent AG2 can look up how to use external libraries on the spot and make simple problem fixes on the spot. The evaluation agent AG3 not only simply refers to the execution results of tasks, but can also actually execute tasks to find problems, find specific ways to fix errors and mistakes, and provide feedback.

[0128] The information processing system 1000 divides the context of each agent by task. Information sharing between agents occurs through project information created by the planning agent AG1 and a global workspace that stores data generated by each agent. Each agent can objectively execute its own processing without being affected by the context of other agents. Furthermore, each agent can autonomously acquire the necessary data through the global workspace.

[0129] The global workspace structures and stores data so that each agent can easily retrieve the information they need. This allows for complete isolation of each agent's context. The information processing system 1000 can handle even very long and complex tasks that are too large to be contained within the context window.

[0130] [Application Examples] The information processing system 1000 can realize a simulation agent that automates materials research. If this simulation agent is realized, for example, complex and highly specialized materials simulations can be automated, allowing advanced simulations to be easily performed regardless of the user's skill level.

[0131] Coding agents can perform simple tasks with high accuracy. Furthermore, with instructions and feedback, they can perform more complex tasks. However, coding agents have challenges in handling complex tasks, such as declining performance or going in the wrong direction. Additionally, coding agents may automatically change the requirements and declare the task complete if it fails. Moreover, because coding agents rely on the knowledge of machine learning models, they are prone to failure in tasks in areas where training data is insufficient.

[0132] Information processing system 1000 is a multi-agent system in which multiple agents plan, execute, evaluate, and modify tasks. The overall research status and the results and reports of each task are structured within the work environment and can be utilized by each agent. By controlling multiple agents and performing appropriate context engineering, information processing system 1000 can autonomously execute complex research tasks.

[0133] As an example, the results of a materials simulation performed by the information processing system 1000 will be described. This materials simulation was performed to analyze the size dependence of the melting point of gold nanoparticles.

[0134] Figure 3 shows an example of instruction information. Figure 3 shows instruction information 400, which instructs a simulation to analyze the size dependence of the melting point of gold nanoparticles. Instruction information 400 contains only concise instructions such as, "We want to know the size dependence of the melting point of gold nanoparticles," and "It should be feasible within a realistic computation time and with a scientifically appropriate setup. Use a heating simulation. Otherwise, use the most suitable method." Instruction information 400 only includes the purpose and outline of the material simulation and does not include detailed processing methods or conditions.

[0135] Figure 4 shows an example of a project plan. Figure 4 shows the project plan 410 generated by the planning agent AG1 based on the instruction information 400 shown in Figure 3. As shown in Figure 4, the project plan 410 includes the project overview, scientific background, melting determination indicators, etc.

[0136] Project Plan 410 outlines the creation of five gold nanoparticles with diameters ranging from 1.5 to 3.5 nanometers to quantitatively analyze the size dependence of the melting point of gold nanoparticles, and the performance of heating simulations using molecular dynamics (MD). Furthermore, Project Plan 410 indicates the use of energy-to-temperature plots and the Lindemann index as indicators for determining melting. Finally, Project Plan 410 also presents the scientific background underlying the project plan.

[0137] Figure 5 shows an example of a task list. Figure 5 shows the task list 420 generated by the planning agent AG1 based on the instruction information 400 shown in Figure 3. Note that the task list 420 shown in Figure 5 is only a part of the task list. The number of tasks included in task list 420 can be arbitrary, but for example, the entire task list may contain several dozen tasks. Note that each task may contain one or more subtasks. The number of subtasks included in a single task list can be arbitrary, but for example, the entire task list may contain several hundred subtasks.

[0138] As shown in Figure 5, task list 420 may include tasks that do not generate code (e.g., tasks 1, 11, etc.). Task list 420 may include tasks that generate libraries (e.g., tasks 2, 5, 8, 13, etc.). Task list 420 may include tasks that generate code (e.g., tasks 3, 6, 9, 14, 16, etc.) and tasks that execute the generated code (e.g., tasks 4, 7, 10, 15, 17, etc.). Task list 420 may include tasks that utilize external tools (e.g., task 12, etc.). Task list 420 may include tasks that verify the operation of external tools (e.g., task 10, etc.).

[0139] Figure 6 shows an example of the processing results. Figure 6 shows the graph generated by Task 17 of Task List 420 shown in Figure 5. Task 17 is a task that generates a graph by executing a visualization script. The visualization script is a script that visualizes the analysis results and is generated in Task 16. The analysis results are the results of determining the melting point of each size using the analysis script generated in Task 14, based on the results of the temperature rise simulation executed in Task 12.

[0140] As shown in Figure 6, the processing results include Graph 431, which visualizes the temperature dependence of energy, and Graph 432, which visualizes the temperature dependence of the Lindemann index. For example, in Graph 431, which shows the relationship between energy and temperature, a temperature at which the energy increases discontinuously can be considered the melting point. On the other hand, in Graph 432, which shows the relationship between the Lindemann index and temperature, if the Lindemann index exceeds 0.1, it can be considered that the gold nanoparticles have melted. Both Graph 431 and Graph 432 show that the melting point is around 800K.

[0141] This application demonstrates that the information processing system 1000 can autonomously perform complex, highly specialized simulations, such as analyzing the temperature dependence of gold nanoparticles. Furthermore, it was shown that even with simple instructions like those shown in Figure 3, the system can automatically execute complex simulations.

[0142] <Summary> As is clear from the above description, an information processing system 1000 according to one embodiment of the present disclosure comprises at least one memory for storing program code and at least one processor, wherein the at least one processor executes at least a portion of the program code, thereby causing the at least one processor to perform the following actions: generate at least a plurality of tasks based on instruction information using a first agent; execute each of the plurality of tasks using a second agent; evaluate the execution result of each of the plurality of tasks using a third agent; modify one or more tasks included in the plurality of tasks based on the evaluation result of each of the plurality of tasks using at least one of the second or fourth agents; execute the modified one or more tasks using at least one of the second or fourth agents; and output the processing result for the instruction information generated based on the execution of the plurality of tasks, wherein the second agent is a coding agent, and the second agent and the third agent are different agents.

[0143] Furthermore, the use of each agent by the processor includes the execution of each agent by the processor. Additionally, the use of each agent by the processor includes the execution of program code necessary for each agent to run.

[0144] Modifying a task includes, based on the evaluation results, making the task generated by the first agent executable in a different way than before the modification, or changing, modifying, or generating various information necessary for that purpose. Modifying a task also includes, based on the evaluation results, modifying at least a portion of the information (including code, etc.) used to execute the task before the modification.

[0145] The information processing system 1000 includes a configuration in which a fourth agent does not exist. "Multiple task executions" in "generated based on the execution of multiple tasks" includes the execution of modified tasks.

[0146] The information processing system 1000 may operate without modifying tasks. In other words, the information processing system 1000 only needs to have a modification function that responds to the execution results of tasks. For example, the information processing system 1000 may operate without modifying tasks based on a certain instruction information A, but modify tasks based on other instruction information B. The information processing system 1000 in this case is also included in the scope of the information processing system 1000 according to this embodiment.

[0147] The information processing system 1000 may obtain multiple different instruction pieces from a single user and execute a task. The information processing system 1000 may obtain instruction pieces from multiple users and execute a task.

[0148] The information processing system 1000 may repeatedly perform evaluation, correction, and execution of one or more corrected tasks until a predetermined standard is met.

[0149] The information processing system 1000 may use a first agent to generate at least multiple tasks based on second instruction information different from the instruction information, use a second agent to execute each of the multiple tasks based on the second instruction information, use a third agent to evaluate the execution results of each of the multiple tasks based on the second instruction information, and output the processing results for the second instruction information generated based on the execution of the multiple tasks based on the second instruction information.

[0150] The fourth agent can also be a coding agent.

[0151] The second and third agents may operate using a working environment that allows them to access common resources.

[0152] The working environment may include memory areas for storing data used by the second agent and the third agent, respectively.

[0153] The work environment may include storage areas for storing data used to perform each of the multiple tasks.

[0154] The working environment may include a storage area for data shared by the second and third agents.

[0155] The second and third agents may each operate using independent contexts.

[0156] The second agent may execute each of the multiple tasks using an independent context.

[0157] The third agent may evaluate the execution results of each of the multiple tasks using independent contexts.

[0158] The first agent may generate validation tasks for external tools.

[0159] The information processing system 1000 may present the user with multiple tasks generated by the first agent.

[0160] The information processing system 1000 may accept modifications by users to one or more tasks included in a set of multiple tasks.

[0161] The information processing system 1000 may receive feedback information from users and, using the first agent, modify one or more tasks based on the feedback information.

[0162] The information processing system 1000 may input information indicating the evaluation criteria to the third agent.

[0163] The information processing system 1000 may use a third agent to generate a report showing the evaluation results for each of several tasks, and input the report into a machine learning model to determine whether or not to modify one or more tasks.

[0164] As a result, according to one embodiment of this disclosure, highly reliable processing results can be generated. For example, according to this embodiment, highly reliable processing results can be generated even for difficult, long-term tasks. Also, according to this embodiment, even with simple instructions, highly reliable processing results can be generated for difficult, long-term tasks. In one aspect, according to this embodiment, appropriate processing results can be generated for user instructions because the execution results of tasks based on instruction information are evaluated. In another aspect, according to this embodiment, the agent that executes the task and the agent that evaluates the execution results of the task do not share a context, so the execution results of the task can be evaluated objectively. In yet another aspect, according to this embodiment, the agent that executes the task and the agent that evaluates the execution results of the task operate in a common work environment, so they can easily refer to each other's operation results.

[0165] [Hardware configuration of information processing equipment] In the embodiments described above, some or all of the devices (information processing devices 10) may be composed of hardware, or they may be composed of information processing by software (programs) executed by a CPU (Central Processing Unit), GPU (Graphics Processing Unit), etc. If the information processing is composed of software, the software that realizes at least some of the functions of each device in the embodiments described above may be stored on a non-temporary storage medium (non-temporary computer-readable medium) such as a CD-ROM (Compact Disc-Read Only Memory) or USB (Universal Serial Bus) memory, and the software information processing may be executed by loading it into a computer. Alternatively, the software may be downloaded via a communication network. Furthermore, all or part of the software processing may be implemented in a circuit such as an ASIC (Application Specific Integrated Circuit) or FPGA (Field Programmable Gate Array), so that the information processing by the software is executed by hardware.

[0166] The storage medium for the software may be a removable medium such as an optical disc, or a fixed storage medium such as a hard disk or memory. Furthermore, the storage medium may be located inside the computer (main memory, auxiliary storage, etc.) or outside the computer.

[0167] Figure 7 is a block diagram showing an example of the hardware configuration of each device (information processing device 10) in the embodiment described above. Each device may be implemented as a computer 7, for example, comprising a processor 71, a main memory 72 (memory), an auxiliary memory 73 (memory), a network interface 74, and a device interface 75, all connected via a bus 76.

[0168] The computer 7 in Figure 7 has one of each component, but it may have multiple identical components. Also, although Figure 7 shows one computer 7, the software may be installed on multiple computers, and each of these multiple computers may execute the same or different parts of the software's processing. In this case, it may be a distributed computing configuration in which each computer communicates via a network interface 74 or the like to execute processing. In other words, each device (information processing device 10) in the above-described embodiment may be configured as a system that realizes its function by having one or more computers execute instructions stored in one or more storage devices. Alternatively, it may be configured so that information transmitted from a terminal is processed by one or more computers located on the cloud, and the processing results are transmitted to the terminal.

[0169] The various calculations performed by each device (information processing device 10) in the embodiments described above may be executed in parallel using one or more processors, or using multiple computers via a network. Alternatively, the various calculations may be distributed to multiple processing cores within a processor and executed in parallel. Furthermore, some or all of the processing and means of this disclosure may be implemented by at least one of a processor and a storage device located on a cloud that can communicate with a computer 7 via a network. Thus, each device in the embodiments described above may be in the form of parallel computing using one or more computers.

[0170] The processor 71 may be an electronic circuit (processing circuit, processing circuitry, CPU, GPU, FPGA, ASIC, etc.) that performs either control or calculations of a computer. The processor 71 may also be a general-purpose processor, a dedicated processing circuit designed to perform specific calculations, or a semiconductor device that includes both a general-purpose processor and a dedicated processing circuit. Furthermore, the processor 71 may include optical circuits or quantum computing-based calculation functions.

[0171] The processor 71 may perform calculations based on data and software input from various devices within the computer 7, and may output calculation results and control signals to these devices. The processor 71 may also control the various components of the computer 7 by executing the computer 7's OS (Operating System) or applications.

[0172] Each device (information processing device 10) in the above-described embodiment may be implemented by one or more processors 71. Here, the processor 71 may refer to one or more electronic circuits arranged on one chip, or one or more electronic circuits arranged on two or more chips or two or more devices. When multiple electronic circuits are used, each electronic circuit may communicate by wire or wireless.

[0173] The main memory 72 may store instructions executed by the processor 71 and various data, and the information stored in the main memory 72 may be read by the processor 71. The auxiliary memory 73 is a memory device other than the main memory 72. These memory devices refer to any electronic component capable of storing electronic information, and may be semiconductor memory. The semiconductor memory may be either volatile memory or non-volatile memory. In each of the devices (information processing device 10) in the above-described embodiment, the memory device for storing various data may be implemented by the main memory 72 or the auxiliary memory 73, or by the built-in memory of the processor 71. For example, each storage unit in the above-described embodiment may be implemented by the main memory 72 or the auxiliary memory 73.

[0174] In the embodiments described above, if each device (information processing device 10) consists of at least one storage device (memory) and at least one processor connected to (coupled with) this at least one storage device, then at least one processor may be connected to one storage device. Also, at least one storage device may be connected to one processor. Furthermore, the configuration may include at least one processor among a plurality of processors being connected to at least one storage device among a plurality of storage devices. This configuration may also be realized by storage devices and processors included in a plurality of computers. Moreover, the configuration may include a storage device integrated with a processor (for example, a cache memory including an L1 cache and an L2 cache).

[0175] The network interface 74 is an interface for connecting to the communication network 8 wirelessly or via a wired connection. The network interface 74 can be any appropriate interface, such as one conforming to existing communication standards. Information may be exchanged between the computer 7 and an external device 9A connected via the communication network 8 through the network interface 74. The communication network 8 may be a WAN (Wide Area Network), LAN (Local Area Network), PAN (Personal Area Network), or a combination thereof, as long as information is exchanged between the computer 7 and the external device 9A. An example of a WAN is the Internet, an example of a LAN is IEEE 802.11 or Ethernet (registered trademark), and an example of a PAN is Bluetooth (registered trademark) or NFC (Near Field Communication).

[0176] The device interface 75 is an interface such as USB that connects directly to the external device 9B.

[0177] External device 9A is a device connected to computer 7 via a network. External device 9B is a device directly connected to computer 7.

[0178] External device 9A or external device 9B may, for example, be an input device. The input device may be a camera, microphone, motion capture device, various sensors, keyboard, mouse, touch panel, etc., and provides the acquired information to the computer 7. Alternatively, it may be a device equipped with an input unit, memory, and processor, such as a personal computer, tablet terminal, or smartphone.

[0179] Furthermore, external device 9A or external device 9B may, for example, be an output device. The output device may be a display device such as an LCD (Liquid Crystal Display) or an organic EL (Electro Luminescence) panel, or a speaker that outputs sound, etc. It may also be a device equipped with an output unit, memory, and a processor, such as a personal computer, tablet terminal, or smartphone.

[0180] Furthermore, external device 9A or external device 9B may be a storage device (memory). For example, external device 9A may be network storage, and external device 9B may be storage such as an HDD.

[0181] Furthermore, the external device 9A or external device 9B may be a device that has some of the functions of the components of each device (information processing device 10) in the embodiment described above. In other words, the computer 7 may transmit some or all of the processing results to the external device 9A or external device 9B, or may receive some or all of the processing results from the external device 9A or external device 9B.

[0182] In this specification (including the claims), when the expression "at least one of a, b, and c" or "at least one of a, b, or c" (including similar expressions) is used, it includes any of a, b, c, ab, ac, bc, or abc. Furthermore, any element may have multiple instances, such as aa, abb, aabbcc, etc. In addition, it is also possible to add other elements other than the enumerated elements (a, b, and c), such as abcd which has d.

[0183] In this specification (including the claims), when expressions such as "using data as input / based on data / according to / in accordance with data" (including similar expressions) are used, unless otherwise specified, this includes using the data itself or using data that has been processed in some way (e.g., data with added noise, normalized data, features extracted from the data, intermediate representations of the data, etc.). Furthermore, when it is stated that some result is obtained "using data as input / based on data / according to / in accordance with data" (including similar expressions), unless otherwise specified, this includes cases where the result is obtained based solely on the data in question or where the result is influenced by other data, factors, conditions, and / or states other than the data in question. Furthermore, when it is stated that "data is output" (including similar expressions), unless otherwise specified, this includes cases where the data itself is used as output or where data that has been processed in some way (e.g., data with added noise, normalized data, features extracted from the data, intermediate representations of various types of data, etc.) is used as output.

[0184] In this specification (including the claims), the terms “connected” and “coupled” are intended to be non-restrictive terms that include any direct connection / coupling, indirect connection / coupling, electrical connection / coupling, communicative connection / coupling, operational connection / coupling, physical connection / coupling, etc. The terms should be interpreted as appropriate in the context in which they are used, but any form of connection / coupling that is not intentionally or naturally excluded should be interpreted non-restrictively as being included in the terms.

[0185] In this specification (including the claims), when the expression "A configured to B" is used, it may include that the physical structure of element A has a configuration capable of performing operation B, and that the permanent or temporary setting / configuration of element A is configured to actually perform operation B. For example, if element A is a general-purpose processor, it is sufficient that the processor has a hardware configuration capable of performing operation B, and that it is configured to actually perform operation B by the setting of a permanent or temporary program (instruction). Furthermore, if element A is a dedicated processor, dedicated arithmetic circuit, etc., it is sufficient that the circuit structure of the processor is implemented to actually perform operation B, regardless of whether control instructions and data are actually attached.

[0186] Wherever terms meaning "comprising" or "possessing" (e.g., "comprising / including," "having," etc.) are used herein, they are intended to be open-ended terms, including cases where the subject matter of such terms is not the object of the term. Where the object of such terms meaning "comprising" or "possessing" is an expression that does not specify a quantity or suggests a singular number (an expression with the article "a" or "an"), such expression should be interpreted as not being limited to a specific number.

[0187] In this specification (including the claims), even if expressions such as "one or more" or "at least one" are used in some places, and expressions that do not specify a quantity or suggest a singularity (expressions using the articles a or an) are used in other places, the latter expressions are not intended to mean "one." In general, expressions that do not specify a quantity or suggest a singularity (expressions using the articles a or an) should not necessarily be interpreted as not being limited to a specific number.

[0188] In this specification, if a particular configuration of an embodiment is described as having a specific advantage or result, it should be understood, unless otherwise stated, that the same advantage or result can also be obtained from one or more other embodiments having that configuration. However, it should be understood that the presence or absence of such advantage or result generally depends on various factors, conditions, and / or states, and that the configuration does not necessarily guarantee that the advantage or result can be obtained. The advantage or result can only be obtained from the configuration described in the embodiment when various factors, conditions, and / or states are met, and the advantage or result cannot necessarily be obtained in the invention claimed to define that configuration or a similar configuration.

[0189] In this specification (including the claims), when multiple hardware components perform a predetermined process, each component may cooperate to perform the predetermined process, or some components may perform all of the predetermined process. Alternatively, some components may perform part of the predetermined process, while other components perform the remainder. In this specification (including the claims), when expressions such as "one or more hardware components perform a first process, and the one or more hardware components perform a second process" (including similar expressions) are used, the hardware component performing the first process and the hardware component performing the second process may be the same or different. In other words, it is sufficient that the hardware component performing the first process and the hardware component performing the second process are included in the one or more hardware components. Hardware may include electronic circuits, devices containing electronic circuits, etc.

[0190] In this specification (including the claims), when multiple memory devices store data, each of the multiple memory devices may store only a portion of the data or the entire data. Furthermore, a configuration in which some of the multiple memory devices store data is also included.

[0191] In this specification (including the claims), terms such as “first,” “second,” etc., are used merely as a way of distinguishing between two or more elements and are not necessarily intended to impose technical meanings such as temporal, spatial, order, or quantity on the subject. Therefore, for example, references to a first element and a second element do not necessarily mean that only two elements can be employed therein, that the first element must precede the second element, or that the first element must exist for the second element to exist.

[0192] While embodiments of this disclosure have been described in detail above, this disclosure is not limited to the individual embodiments described above. Various additions, modifications, substitutions, and partial deletions are possible, provided that they do not depart from the conceptual idea and spirit of the present invention derived from the claims and their equivalents. For example, where numerical values ​​or mathematical formulas are used in the description of the embodiments described above, these are provided for illustrative purposes only and do not limit the scope of this disclosure. Similarly, the sequence of operations shown in the embodiments is also illustrative and does not limit the scope of this disclosure.

[0193] Furthermore, the following forms are possible for disclosure technology.

[0194] (Note 1) At least one memory location for storing program code, Equipped with at least one processor, The at least one processor executes at least a portion of the program code, thereby causing the at least one processor to: Using the first agent, at least multiple tasks are generated based on the instruction information. Using the second agent, each of the above multiple tasks is executed. Using a third agent, the execution results of each of the aforementioned tasks are evaluated. Using at least one of the second or fourth agents, modify one or more tasks included in the plurality of tasks based on the evaluation results of each of the plurality of tasks, Use at least one of the above second agent or the above fourth agent to perform one or more modified tasks. Outputting the processing results for the instruction information generated based on the execution of the aforementioned multiple tasks, Execute the process, The previous2 agent is a coding agent, The former second agent and the former third agent are different agents. Information processing system.

[0195] (Note 2) The aforementioned at least one processor, The evaluation, the correction, and the execution of one or more tasks after the correction are repeated until a predetermined criterion is met. The information processing system described in Appendix 1.

[0196] (Note 3) The aforementioned at least one processor, Using the first agent, at least several tasks are generated based on second instruction information that differs from the instruction information. Using the second agent, each of the multiple tasks based on the second instruction information is executed. Using the third agent, the execution results of each of the multiple tasks based on the second instruction information are evaluated. Outputting the processing results for the second instruction information, which are generated based on the execution of the multiple tasks based on the second instruction information. To execute the process The information processing system described in Appendix 1 or 2.

[0197] (Note 4) The fourth agent is a coding agent. An information processing system as described in any one of the items in Appendix 1 to Appendix 3.

[0198] (Note 5) The second agent and the third agent operate using a work environment that has access to common resources. An information processing system as described in any one of the items in Appendix 1 to Appendix 4.

[0199] (Note 6) The aforementioned work environment includes a storage area for storing data used by the second agent and the third agent, The information processing system described in Appendix 5.

[0200] (Note 7) The aforementioned work environment includes a storage area for storing data used to perform each of the aforementioned tasks, The information processing system described in Appendix 5 or 6.

[0201] (Note 8) The aforementioned work environment includes a storage area for storing data shared by the second agent and the third agent, An information processing system as described in any one of the items in Appendix 5 to Appendix 7.

[0202] (Note 9) The configured 2 agent and the configured 3 agent operate using independent contexts. An information processing system as described in any one of the items in Appendix 1 through Appendix 8.

[0203] (Note 10) The aforementioned second agent executes each of the aforementioned tasks using an independent context. An information processing system as described in any one of the items in Appendix 1 through Appendix 9.

[0204] (Note 11) The third agent evaluates the execution results of each of the multiple tasks using an independent context. An information processing system as described in any one of the items in Appendix 1 through Appendix 10.

[0205] (Note 12) The aforementioned first agent generates verification tasks for external tools. An information processing system as described in any one of the items in Appendix 1 through Appendix 11.

[0206] (Note 13) The aforementioned at least one processor, The above-mentioned first agent presents the multiple tasks it has generated to the user. To execute the process An information processing system as described in any one of the items in Appendix 1 to Appendix 12.

[0207] (Note 14) The aforementioned at least one processor, The system accepts modifications by the user to one or more tasks included in the multiple tasks, To execute the process The information processing system described in Appendix 13.

[0208] (Note 15) The aforementioned at least one processor, We receive feedback information from the aforementioned users, Using the first agent, the one or more tasks are modified based on the feedback information. To execute the process The information processing system described in Appendix 14.

[0209] (Note 16) The aforementioned at least one processor, Information indicating the evaluation criteria is input to the third agent. To execute the process An information processing system as described in any one of the items in Appendix 1 through Appendix 15.

[0210] (Note 17) The aforementioned at least one processor, Using the aforementioned third agent, a report showing the evaluation results for each of the multiple tasks is generated. By inputting the aforementioned report into the machine learning model, it is determined whether or not to modify the one or more tasks mentioned above. To execute the process An information processing system as described in any of the appendices 1 through 16.

[0211] (Note 18) Using the information processing system described in any one of the items from Appendix 1 to Appendix 17, Based on the instruction information, generate at least several tasks, Each of the aforementioned tasks is executed, Outputting the processing results for the instruction information generated based on the execution of the aforementioned multiple tasks, Information processing methods.

[0212] (Note 19) A program to cause at least one processor to execute the information processing method described in Appendix 18.

[0213] (Note 20) At least one processor executes at least a portion of program code stored in at least one memory, thereby enabling the at least one processor to: Using the first agent, at least multiple tasks are generated based on the instruction information. Using the second agent, each of the above multiple tasks is executed. Using a third agent, the execution results of each of the aforementioned tasks are evaluated. Using at least one of the second or fourth agents, modify one or more tasks included in the plurality of tasks based on the evaluation results of each of the plurality of tasks, Use at least one of the above second agent or the above fourth agent to perform one or more modified tasks. The system outputs the processing results for the instruction information generated based on the execution of the aforementioned multiple tasks. The previous2 agent is a coding agent, The former second agent and the former third agent are different agents. Information processing methods.

[0214] (Note 21) At least one processor executes at least a portion of program code stored in at least one memory, thereby causing the at least one processor to: Using the first agent, at least multiple tasks are generated based on the instruction information. Using the second agent, each of the above multiple tasks is executed. Using a third agent, the execution results of each of the aforementioned tasks are evaluated. Using at least one of the second or fourth agents, modify one or more tasks included in the plurality of tasks based on the evaluation results of each of the plurality of tasks, Use at least one of the above second agent or the above fourth agent to perform one or more modified tasks. Outputting the processing results for the instruction information generated based on the execution of the aforementioned multiple tasks, Execute the process, The previous2 agent is a coding agent, The former second agent and the former third agent are different agents. A computer-readable storage medium that stores the aforementioned program code. [Explanation of Symbols]

[0215] 10: Information Processing Devices 1000: Information Processing Systems AG1: Planning Agent AG2: Execution Agent AG3: Evaluation Agent AG4: Correction Agent M: Machine learning model T: External Tools

Claims

1. At least one memory for storing program code, It comprises at least one processor, The at least one processor executes at least a portion of the program code, thereby causing the at least one processor to: Using one or more first agents, at least multiple tasks are generated based on instruction information. Using one or more second agents, each of the above-mentioned tasks is executed. Using one or more third agents, the execution results of each of the multiple tasks are evaluated. Using at least one of the two second agents or one or more fourth agents, one or more tasks included in the multiple tasks are modified based on the evaluation results of each of the multiple tasks. Using at least one of the above two agents or one of the above four agents, execute one or more modified tasks. Outputting the processing results for the instruction information generated based on the execution of the aforementioned multiple tasks, Execute the process, The one or more second agents include at least a coding agent that generates code for performing a task and executes the generated code. The one or more second agents and the one or more third agents include different agents. Modifying the one or more tasks described above includes modifying at least a portion of the code used to execute the one or more tasks described above. Performing one or more of the modified tasks includes executing the modified code. The one or more second agents and the one or more third agents operate using a work environment that has access to common resources. The one or more second agents and the one or more third agents each operate using an independent context. The one or more third agents can access at least a portion of the execution results of each of the multiple tasks through the work environment. The at least one processor is repeatedly made to evaluate the execution results, modify one or more tasks, and execute one or more modified tasks until a predetermined criterion is met. Information processing system.

2. The aforementioned at least one processor, Using the one or more first agents described above, at least multiple tasks are generated based on second instruction information that differs from the instruction information described above. Using the one or more second agents, each of the multiple tasks based on the second instruction information is executed. Using the one or more third agents described above, the execution results of each of the multiple tasks based on the second instruction information are evaluated. Outputting the processing results for the second instruction information, which are generated based on the execution of the multiple tasks based on the second instruction information. To execute the process The information processing system according to claim 1.

3. The one or more fourth agents mentioned above include at least a coding agent. The information processing system according to claim 1.

4. The aforementioned work environment includes a storage area for storing data used by each of the one or more second agents and the one or more third agents. The information processing system according to claim 1.

5. The aforementioned work environment includes a storage area for storing data used to perform each of the aforementioned tasks, The information processing system according to claim 1.

6. The aforementioned work environment includes a storage area for storing data shared by the one or more second agents and the one or more third agents. The information processing system according to claim 1.

7. The one or more second agents described above execute each of the multiple tasks using an independent context. The information processing system according to claim 1.

8. The one or more third agents described above evaluate the execution results of each of the multiple tasks using independent contexts. The information processing system according to claim 1.

9. The one or more first agents mentioned above generate verification tasks for external tools. The information processing system according to claim 1.

10. The aforementioned at least one processor, The user is presented with the multiple tasks generated by the one or more first agents. To execute the process The information processing system according to claim 1.

11. The aforementioned at least one processor, The system accepts modifications by the user to one or more tasks included in the multiple tasks. To execute the process The information processing system according to claim 10.

12. The aforementioned at least one processor, We receive feedback information from the aforementioned users, Using the one or more first agents, the one or more tasks are modified based on the feedback information. To execute the process The information processing system according to claim 11.

13. The aforementioned at least one processor, Information indicating the evaluation criteria is input to the one or more third agents. To execute the process The information processing system according to claim 1.

14. The aforementioned at least one processor, Using the one or more third agents described above, a report showing the evaluation results for each of the multiple tasks is generated. By inputting the aforementioned report into the machine learning model, it is determined whether or not to modify one or more of the aforementioned tasks. To execute the process The information processing system according to claim 1.

15. The one or more second agents include an agent that performs the one or more modified tasks. The information processing system according to claim 1.

16. The one or more fourth agents include agents that perform the one or more modified tasks. The information processing system according to claim 1.

17. Modifying at least a portion of the code used to execute the one or more tasks includes generating code that is at least partially different from the code used to execute the one or more tasks, Executing the modified code involves executing code that is at least partially different from the generated code used to execute the one or more tasks. The information processing system according to claim 1.

18. Using one or more first agents to generate at least the plurality of tasks based on the instruction information includes using one or more first agents to generate a task list that includes at least the plurality of tasks based on the instruction information. The information processing system according to claim 1.

19. The aforementioned at least one processor, Using the one or more first agents described above, project information including at least the task list and project plan is generated based on the instruction information. Execute the process, The aforementioned project plan includes at least one of the following: the purpose of the process, the final goal, the background, the expected results, or the method of the process. The information processing system according to claim 18.

20. The aforementioned at least one processor, We accept feedback information from users. Based on the aforementioned feedback information, the project information is modified. To execute the process The information processing system according to claim 19.

21. Executing each of the multiple tasks using the one or more second agents includes executing two or more tasks included in the multiple tasks in parallel using the one or more second agents. The execution of the two or more tasks in parallel is determined by the one or more first agents. The information processing system according to claim 1.

22. The information processing system according to any one of claims 1 to 21 is Based on the instruction information, generate at least several tasks, Each of the aforementioned tasks is executed, Outputting the processing results for the instruction information generated based on the execution of the aforementioned multiple tasks, Information processing methods.

23. A program for causing at least one processor to execute the information processing method described in claim 22.

24. At least one processor executes at least a portion of program code stored in at least one memory, so that the at least one processor Using one or more first agents, at least multiple tasks are generated based on instruction information. Using one or more second agents, each of the above-mentioned tasks is executed. Using one or more third agents, the execution results of each of the multiple tasks are evaluated. Using at least one of the two second agents or one or more fourth agents, one or more tasks included in the multiple tasks are modified based on the evaluation results of each of the multiple tasks. Using at least one of the above two agents or one of the above four agents, execute one or more modified tasks. The system outputs the processing results for the instruction information generated based on the execution of the aforementioned multiple tasks. The one or more second agents include at least a coding agent that generates code for performing a task and executes the generated code. The one or more second agents and the one or more third agents include different agents. Modifying the one or more tasks described above includes modifying at least a portion of the code used to execute the one or more tasks described above. Performing one or more of the modified tasks includes executing the modified code. The one or more second agents and the one or more third agents operate using a work environment that has access to common resources. The one or more second agents and the one or more third agents each operate using an independent context. The one or more third agents can access at least a portion of the execution results of each of the multiple tasks through the work environment. The at least one processor repeatedly performs the following until a predetermined criterion is met: evaluation of the execution result, modification of one or more tasks, and execution of one or more modified tasks. Information processing methods.

25. At least one processor executes at least a portion of program code stored in at least one memory, thereby causing the at least one processor to: Using one or more first agents, at least multiple tasks are generated based on instruction information. Using one or more second agents, each of the above-mentioned tasks is executed. Using one or more third agents, the execution results of each of the multiple tasks are evaluated. Using at least one of the two second agents or one or more fourth agents, one or more tasks included in the multiple tasks are modified based on the evaluation results of each of the multiple tasks. Using at least one of the above two agents or one of the above four agents, execute one or more modified tasks. Outputting the processing results for the instruction information generated based on the execution of the aforementioned multiple tasks, Execute the process, The one or more second agents include at least a coding agent that generates code for performing a task and executes the generated code. The one or more second agents and the one or more third agents include different agents. Modifying the one or more tasks described above includes modifying at least a portion of the code used to execute the one or more tasks described above. Performing one or more of the modified tasks includes executing the modified code. The one or more second agents and the one or more third agents operate using a work environment that has access to common resources. The one or more second agents and the one or more third agents each operate using an independent context. The one or more third agents can access at least a portion of the execution results of each of the multiple tasks through the work environment. The at least one processor is repeatedly made to evaluate the execution results, modify one or more tasks, and execute one or more modified tasks until a predetermined criterion is met. A computer-readable storage medium that stores the aforementioned program code.

Images