Information processing method, information processing device, and computer program

Abstract

The present invention provides an information processing method for using a plurality of AIs to provide services.　The information processing method comprises: a reception step for receiving an input that includes a purpose of a task and a constraint condition; an identification step for identifying a delegation destination model and delegation content on the basis of the input; a transmission step for transmitting the delegation content to the identified delegation destination model; a result reception step for receiving an execution result of the delegation content from the delegation destination model; and a presentation step for presenting information based on the execution result. In the identification step, the delegation destination model is identified from among the plurality of AI models on the basis of execution history data together with the delegation content.

Description

Information processing method, information processing device, and computer program

[0001] The technology disclosed in this specification (hereinafter referred to as "the present disclosure") relates to an information processing method, an information processing device, and a computer program that perform processing using an AI model.

[0002] The development of machine learning has been remarkable in recent years, and the introduction of AI (Artificial Intelligence) is progressing in various fields. In particular, the progress of generative AI such as ChatGPT has been remarkable. In addition, various products and services using generative AI have appeared, including image-generating AI such as Stable Diffusion and Midjourney. To date, generative AI products and services have been limited to automatically generating content itself, such as images, videos, music, and text.

[0003] On the other hand, advances in AI have led to an increase in cases of interaction between AIs. Examples of interaction between AIs include collaboration and cooperation between AI assistants, AI training of AI, AI monitoring of AI, and AI control of AI. For example, a device has been proposed that acquires output values ​​for a target task using multiple task models (see Patent Document 1). While interaction between AIs is becoming increasingly important, such interaction should be carried out under appropriate monitoring and control.

[0004] Japanese Patent Application Laid-Open No. 2023-39615

[0005] An object of the present disclosure is to provide an information processing method, an information processing device, and a computer program that perform processing utilizing an AI model.

[0006] The present disclosure has been made in consideration of the above-mentioned problems, and a first aspect thereof is an information processing method including: a receiving step of receiving an input including a task purpose and constraints; a specifying step of identifying a delegatee model and delegation content based on the input; a transmitting step of transmitting the delegation content to the identified delegatee model; a result receiving step of receiving an execution result of the delegation content from the delegatee model; and a presenting step of presenting information based on the execution result. Each step is performed using a multimodal AI model. In this specification, this multimodal AI model is referred to as a "supervisor AI model."

[0007] In the identifying step, the delegatee model is identified from among a plurality of AI models based on the delegation content and execution history data. The execution history data includes at least one of the cost and time required to perform the task, performance, achievements, real-world sensing results (comment information such as criticisms and word-of-mouth posted on social media, ethics check results, etc.), and expertise. Each AI model identified as the delegatee model performs the task through autonomous and collaborative operation.

[0008] A second aspect of the present disclosure is an information processing device including: a receiving unit that receives input including a task purpose and constraints; an identifying unit that identifies a delegatee model and delegation content based on the input; a transmitting unit that transmits the delegation content to the identified delegatee model; a result receiving unit that receives an execution result of the delegation content from the delegatee model; and a presentation unit that presents information based on the execution result.

[0009] Furthermore, a third aspect of the present disclosure is a computer program written in a computer-readable format to cause a computer to function as: a receiving unit that receives input including a task purpose and constraints; an identifying unit that identifies a delegatee model and delegation content based on the input; a transmitting unit that transmits the delegation content to the identified delegatee model; a result receiving unit that receives execution results of the delegation content from the delegatee model; and a presentation unit that presents information based on the execution results.

[0010] A computer program according to a third aspect of the present disclosure defines a computer program written in a computer-readable format to perform predetermined processing on a computer. The computer program can be provided to a computer capable of executing various program codes in a computer-readable format via a storage medium or communication medium, such as an optical disk, a magnetic disk, or a semiconductor memory, or a communication medium such as a network. By installing the computer program according to the third aspect of the present disclosure on a computer via any of these media, a cooperative effect is exerted on the computer, and the same effects as those of the information processing device according to the second aspect of the present disclosure can be obtained.

[0011] FIG. 1 is a diagram showing a schematic configuration of a system 100 to which the present disclosure is applied. FIG. 2 is a diagram showing a specific operation example of the system 100. FIG. 3 is a diagram showing a schematic operation of the system 100 according to the present disclosure. FIG. 4 is a diagram showing an operation example when the system 100 is applied to a company's business activities. FIG. 5 is a diagram showing an example when the system 100 is applied to creative activities. FIG. 6 is a diagram showing the input / output relationship of a multimodal AI model. FIG. 7 is a diagram showing a specific configuration example of the system 100 according to the present disclosure. FIG. 8A is a diagram showing an example of a processing sequence for task execution performed in the system 100 shown in FIG. 7. FIG. 8B is a diagram showing an example of a processing sequence for task execution performed in the system 100 shown in FIG. 7. FIG. 9A is a diagram showing a method for configuring a simple UI. FIG. 9B is a diagram showing another simple method for configuring a UI. FIG. 9C is a diagram showing a UI configuration method in which the product of a supervisor AI model is directly presented to a user and the user directly inputs data to the supervisor AI model. Fig. 10 is a diagram showing an example of the configuration of a UI screen on which a user inputs instructions to the supervisor AI model. Fig. 11 is a diagram showing an example of the configuration of a UI screen on which the supervisor AI model presents task execution results to the user. Fig. 12 is a diagram showing an example of a UI that displays important indicators in a graph. Fig. 13 is a diagram showing an example of the configuration of a UI screen customized for a user. Fig. 14 is a diagram showing an example of the hardware configuration of the information processing device 2000.

[0012] Hereinafter, embodiments of the present disclosure will be described in the following order with reference to the drawings.

[0013] A. Overview A-1. System configuration overview A-2. Compliance with legal regulations and ethics in each country A-3. System operation B. Implementation examples B-1. Examples of application to corporate activities B-2. Examples of application to creative activities C. Prerequisite AI model C-1. Input / output relationships of the AI ​​model C-2. Specific examples of input / output of the AI ​​model D. Specific examples D-1. Specific system configuration D-2. Specific system operation G. Creation of UI screen G-1. UI configuration method G-2. Specific examples of UI screen G-2-1. UI for input from the user G-2-2. UI for output to the user H. Configuration of information processing device

[0014] A. Overview To date, generative AI products and services have been limited to automatically generating content such as images, videos, music, and text. In contrast, the present disclosure proposes a technology that goes beyond generative AI that generates content itself and provides services using a large number of AI models that act as the creators and artists who create the content—in other words, AI models that act as people and perform tasks. This disclosure also makes it possible to utilize digital twins in addition to AI models. Digital twins are a technology that recreates a real space in a virtual space based on the results of analyzing and processing information in the real space using an AI model. In the embodiments of the present disclosure described below, many of the functional modules implemented using an "AI model" can be equivalently implemented by replacing them with a "digital twin."

[0015] In the present disclosure, the generation AI not only sequentially generates person AI models that perform specific tasks, such as creator AI models and artist AI models, but also bundles a large number of person AI models that each perform different tasks to generate an event AI model, a company AI model, a society AI model, and a city AI model that correspond to a single event (such as a concert), company, society, and city, respectively.

[0016] In the system according to the present disclosure, a "supervisor AI model" is used as a supervisory function that bundles and comprehensively supervises a group of AI models and coordinates between the AI ​​models. When the supervisor AI model receives a task request from a customer, it selects one or more AI models from the group of AI models to carry out the requested task and delegates the task to each AI model. Then, each AI model selected by the supervisor AI model operates autonomously and cooperatively to provide a service to the requester. The "customer" referred to here is the source of the service request, but there is no particular limitation on the type of person.

[0017] The supervisor AI model supervises the AI ​​models by evaluating their performance and, based on the results of each evaluation, continuing to use the AI ​​models (equivalent to "continued employment"), deleting (equivalent to "dismissal") or not selecting (equivalent to "reassignment") AI models that are found to be unnecessary, relearning (equivalent to "retraining") AI models, and adding other AI models, thereby striving to improve the performance of the entire system. The supervisor AI model may distribute revenue earned by the AI ​​models it delegates to perform tasks to each AI model according to their performance. The supervisor AI model and each AI model under its supervision are assumed to be foundation models. The foundation model can be built, for example, based on a Transformer.

[0018] A-1. System Overview Figure 1 shows a schematic configuration of a system 100 to which the present disclosure is applied. The system 100 uses AI technology to achieve a user's objectives, and is configured such that an AI model that acts as a supervisor comprehensively oversees a group of AI models consisting of multiple AI models that each operate autonomously and cooperatively to carry out requested tasks.

[0019] 1 is composed of a supervisor AI model 101, an AI model group 110, and an external AI model group 120. The supervisor AI model 101 is directly connected to the AI ​​model group 110, and is externally connected to the external AI model group 120.

[0020] The supervisor AI model 101 comprehensively supervises the AI ​​model group 110 and the external AI model group 120. If the system 100 is likened to corporate activities, the AI ​​model group 110 corresponds to a corporate AI model, and the AI ​​model group 110 includes numerous employee AI models belonging to various departments, such as human resources, accounting, public relations, research and development, design, manufacturing, sales, and materials. Each AI model included in the AI ​​model group 110 is autonomous, but can operate in cooperation with each other, just like employees within a company cooperate with each other in their work. Some of the employee AI models may also be digital twins. Meanwhile, the external AI model group 120 corresponds to a staffing agency AI model or an affiliated company AI model, and includes AI models or digital twins that can replace employee AI models in at least one of the departments of the above-mentioned corporate AI model.

[0021] A user such as an investor makes a request to the corporate AI model. The request includes the task content, task objectives, and constraints (such as cost and delivery date). The supervisor AI model 101 selects or generates, from the AI ​​model group 110, multiple AI models corresponding to employees suitable for carrying out the request (however, "generating" an AI model corresponds to generating an instance of the AI ​​model). In addition, the supervisor AI model 101 selects AI models and digital twins that are insufficient in the AI ​​model group 110 to carry out the request from the external AI model group 120 corresponding to staffing agency AI models and affiliated company AI models.

[0022] The supervisor AI model 101 distributes tasks for realizing the request from the client to each AI model selected as described above (some of which may be digital twins). Each AI model operates autonomously and cooperatively to carry out the task delegated via the supervisor AI model 101, just like an employee and a temporary worker working together to realize a project.

[0023] The supervisor AI model 101, which has a supervisory function for the AI ​​models, evaluates the performance of each AI model and, based on the results of each evaluation, continues to use the AI ​​model (equivalent to "continued employment"), deletes the AI ​​model (equivalent to "dismissal") or does not select it (equivalent to "transfer"), retrains the AI ​​model (equivalent to "retraining"), or adds other AI models, thereby striving to improve the performance of the entire system 100.

[0024] Then, the service realized by each AI model selected by the supervisor AI model 101 is provided to the requester. In response, the requester returns an evaluation of the provided service (including gratitude and complaints). Each AI model and the supervisor AI model 101 can receive feedback and learn. For example, depending on whether the evaluation from the requester is favorable or unfavorable, the supervisor AI model 101 can refer to the current feedback content when selecting an AI model from the AI ​​model group 110 and the external AI model group 120 the next time it receives a request from the same requester or a similar request from another requester. Furthermore, each AI model (and digital twin) of the AI ​​model group 110 and the AI ​​model group 120 can learn the current feedback content and perform the task when it is assigned a task based on a request from the same requester or a similar request from another requester by the supervisor AI model 101.

[0025] In addition, the supervisor AI model 101 distributes the profits earned by the AI ​​models selected by the supervisor AI model 101 performing the tasks to each AI model according to their performance.

[0026] The system 100 shown in Figure 1 is basically designed for a corporate AI model, but even for social AI models or urban AI models that are larger than a company, the supervisory function of the supervisor AI model 101 can select appropriate AI models from the AI ​​model group 110 and external AI model group 120, making it possible to respond with high performance to requests made by society or cities.

[0027] FIG. 2 outlines a specific example of operation of the system 100 according to the present disclosure. In the example shown in FIG. 2, the corporate AI model is assumed to be a company to which creators and artists belong, such as a painter, rock singer, pop singer, dancer, and content creator. Therefore, the AI ​​model group 110 is composed of a collection of creator AI models, artist AI models, and digital twins. Each AI model included in the AI ​​model group 110 is autonomous, but can operate in a cooperative manner, such as when artists cooperate with each other or when artists cooperate with dancers and other support members during a joint concert. Each AI model may also autonomously generate advertisements.

[0028] On the other hand, the external AI model group 120 includes AI models of various human resources, not limited to creators or artists. For example, it is composed of a doctor AI model, a teacher AI model, AI models of various instrument players, band member AI models, AI models of various concert staff who handle cameras, lighting, sound, and other equipment, support member AI models, and a collection of digital twins. Each AI model in the external AI model group 120 can automatically generate advertisements based on performance records and feedback.

[0029] A user such as an investor requests a task of hosting an event, such as a tour or concert, from a corporate AI model. The request also includes constraints such as the timing and cost of the event. The supervisor AI model 101 selects or generates an AI model or digital twin of an artist (e.g., rock singer, pop singer, dancer) suitable for hosting the requested event from the AI ​​model group 110. The supervisor AI model 101 also selects various concert staff AI models, support member AI models, and digital twins that are lacking in the AI ​​model group 110 from the AI ​​model group 120, which corresponds to a staffing agency AI model or an affiliated company AI model.

[0030] The AI ​​models or digital twins of artists (rock singers, pop singers, dancers, etc.) selected and generated by the supervisor AI model 101, and the various concert staff AI models, support member AI models and digital twins summoned from the external AI model group 120, work autonomously and cooperatively to fulfill their respective roles in order to hold and successfully execute the requested event.

[0031] The supervisor AI model 101, which has a supervisory function for the AI ​​models, evaluates the performance of each AI model and, based on the results of each evaluation, continues to use the AI ​​model (equivalent to "continued employment"), deletes the AI ​​model (equivalent to "dismissal") or does not select the AI ​​model (equivalent to "transfer"), retrains the AI ​​model (equivalent to "retraining"), or adds other AI models, thereby striving to improve the performance of the entire system 100 (events in the example shown in Figure 2).

[0032] An event is held using each AI model selected by the supervisor AI model 101. In response, the requester returns an evaluation of the held event (including praise and criticism). Each AI model and the supervisor AI model 101 can receive feedback and learn. For example, depending on whether the evaluation of the held event was favorable or unfavorable, the supervisor AI model 101 can refer to the current feedback when selecting an AI model from the AI ​​model group 110 and the external AI model group 120 the next time the supervisor AI model 101 receives a request to hold an event from the same requester or a request to hold a similar event from another requester. Furthermore, each AI model (and digital twin) of the AI ​​model group 110 and the AI ​​model group 120 can learn the current feedback and achieve better performance when summoned to the same or similar event at the request of the supervisor AI model 101.

[0033] A-2. Compliance with national laws and regulations and ethics When managing and operating the above AI model group, AI models that have learned national laws and regulations and European AI regulations can be used to function in accordance with national laws and ethics. When selecting and generating an AI model from the AI ​​model group 110 and the external AI model group 120, the supervisor AI model 101 can also select or generate an "ethics supervision AI model" that has learned the laws, regulations, and ethics of the country in which the requested task will be performed.

[0034] In countries and regions such as Europe, the United States, and Japan, it is expected that the following obligations will be imposed on providers of platform models:

[0035] - Process and incorporate only datasets for which appropriate data governance measures for the underlying model have been implemented (in particular, ensuring the appropriateness of the data source and mitigation of bias). - Register the generated AI model in a database specified by laws and regulations. - After the AI ​​model is generated or its use begins, technical documentation related to the AI ​​model must be stored so that it can be freely accessed by designated government agencies in the relevant country or region for a specified period of time. - Ensure the transparency of AI models (explainability, reproducibility, ethics, reliability, and fairness). - Train, design, and develop AI models in a way that ensures appropriate protection measures against the generation of content that violates laws and regulations. - Document a sufficiently detailed outline of the use of training data protected by copyright law and make it publicly available.

[0036] Therefore, it is desirable that the supervisor AI model learns the processes for complying with the above obligations to generate an ethics supervision AI model or other AI models and select an external AI model. The ethics supervision AI model can be generated by learning the laws, regulations, and ethics of each country, for example, as follows.

[0037] <Step 1> Data collection: Collect information on AI-related laws, regulations, and guidelines in Europe, the United States, and Japan to create a dataset. For example, collect data such as official documents, legal precedents, and academic papers, and use a crowdsourcing platform to have legal experts scrutinize the data.

[0038] <Step 2> Model training: The collected learning data is used to train the AI ​​model. The model learns information about each country's laws and regulations and generates output based on that information.

[0039] <Step 3> Output comparison and conformance: The output generated by the trained AI model is compared with each country's laws, regulations, guidelines, etc., and its conformance is evaluated.

[0040] <Step 4> Adjusting the output: If necessary, adjust the generated output to comply with local laws, regulations, etc. This may include output modifications and additional training.

[0041] <Step 5> Test and verify: Test the adjusted output and verify whether it complies with the laws and regulations of each country. Make corrections and retrain as necessary.

[0042] <Step 6> Continuous monitoring and updating: Regularly monitor the AI ​​model in response to amendments to laws and regulations in each country and the issuance of new guidelines, and update and learn as necessary.

[0043] Because the training data is legal documents with specialized language and complexities, language processing, understanding of legal context, management of bias and fairness, and transparency of interpretation are important when training the ethics oversight AI model.

[0044] Language processing: Legal documents use specialized and complex language, so natural language processing is used. When developing AI models based on national or regional legal regulations, large-scale language models (such as GPT and BERT) are used to deepen understanding of legal documents.

[0045] Legal context understanding: Legal context understanding utilizes the construction of a legal knowledge graph. When an AI model interprets a legal document, it uses the legal knowledge graph, which shows relevant legal concepts and relationships, to provide an accurate interpretation.

[0046] Bias and fairness management: Bias and fairness management involves incorporating data normalization and fairness constraints. When AI models make legal decisions, data bias is mitigated and fairness constraints are introduced to avoid biased predictions toward specific attributes.

[0047] Transparency of interpretation: To ensure transparency of interpretation, we will utilize interpretable AI models. When an AI model reaches a legal conclusion, we will use methods such as attention mechanisms and SHAP to transparently show how the AI ​​model has reached its judgment.

[0048] A-3. System Operation The operation of the system 100 according to the present disclosure will be described with reference to FIG.

[0049] The supervisor AI model 101 comprehensively controls the operation of the entire system 100 in order to accomplish the purpose specified by the user. Specifically, the supervisor AI model 101 performs the following processes (1) to (4).

[0050] (1) Objective receiving process: Receives input from the user, including the objective of the task and the constraints associated with achieving the objective. (2) Identification process: Selects an AI model to delegate to accomplish the objective of the task under the constraints, and identifies the delegation content to be delegated to the AI ​​model for accomplishing the objective of the task. (3) Delegation sending process: Sends a request to the delegated AI model to accomplish the delegated content. (4) Report receiving process: Receives a report of the results of accomplishing the delegated content from the delegated AI model. (5) Presentation process: Generates a report for the user based on the report received from the delegated AI model and presents it to the user.

[0051] Here, "selecting" an AI model to delegate the accomplishment of a goal means selecting an AI model to delegate to, and includes creating a new AI model and deleting or not selecting an AI model that is found to be unnecessary. "Creating" an AI model in this specification corresponds to creating an instance of the relevant model.

[0052] In short, the supervisor AI model 101 can comprehensively supervise each AI model of the AI ​​model group 110 within the corporate AI model and the external AI model group 120 within the affiliated company AI model and the dispatch company AI model. The supervisor AI model 101 is a multimodal and multitasking AI model. Because it is multimodal, the user can input task objectives and constraints using data of any modality, and feedback to the user (such as a report on the results of accomplishing the task objective) can also be presented in any modality. Details of the multimodal and multitasking AI model will be described later.

[0053] The AI ​​model group 110 within the corporate AI model and the external AI model group 120 within the affiliated company AI model and the temporary staffing agency AI model are made up of a collection of AI models to which the supervisor AI model 101 can delegate task execution. Each AI model included in the AI ​​model group 110 and the external AI model group 120 is capable of autonomous and cooperative operation, and the AI ​​models delegated by the supervisor AI model 101 interact with each other, cooperate with each other, and perform their assigned roles toward a goal, taking action in the real world. The AI ​​model group 110 and the external AI model group 120 also sense real-world reactions to the actions they perform and report them to the supervisor AI model 101. Real-world reactions may include comment information such as criticism and word-of-mouth posted on social media.

[0054] Furthermore, the AI ​​model group 110 and the external AI model group 120 may sense ethical issues caused by task execution results and real-world actions. The AI ​​model group 110 and the external AI model group 120 may include an ethics supervision AI model (see section A-2 above) that has trained processes for complying with the laws and regulations of the relevant country or region and resolving ethical issues. The ethics supervision AI model checks for different ethical issues depending on the country, region, religion, etc. Furthermore, the supervisor AI model 101 may delegate to the AI ​​model group 110 and the external AI model group 120 the creation of an ethics supervision AI model appropriate for the relevant country, region, or cultural sphere.

[0055] For example, if the system 100 is intended for corporate activities, the AI ​​model group 110 corresponds to the AI ​​models of each employee within the company, and the external AI model group 120 corresponds to the employee AI models of a staffing agency or affiliated company. Furthermore, if the system 100 is intended for content production such as music or for holding events such as tours or concerts, the AI ​​model group 110 corresponds to a collection of creator AI models and artist AI models. Furthermore, the external AI model group 120 in this case corresponds to a collection of various concert staff AI models and support member AI models that handle cameras, lighting, sound, and other equipment. Furthermore, the AI ​​model group 110 and the external AI model group 120 may include not only AI models but also digital twins (for example, an artist digital twin may be used instead of an artist AI model).

[0056] The AI ​​model group 110 and the external AI model group 120, i.e., the AI ​​instances created, deleted, or not selected by the supervisor AI model 101, may be any of the following: The AI ​​model group 110 and the external AI model group 120 may also include digital twins.

[0057] (1) Another instance of the same multimodal AI model as the supervisor AI model 101 on the system 100. (2) An instance of an AI model different from the supervisor AI model 101 on the system 100. (3) An AI model or AI instance on a system different from the system 100 (e.g., on the Internet).

[0058] B. Implementations There are various implementations of the system 100 according to the present disclosure, each of which has a different purpose indicated by a user.

[0059] B-1. Example of Application to Corporate Activities Figure 4 shows an example of the operation of the system 100 according to the present disclosure when applied to the corporate activities of a company. In this case, the user who specifies the objective corresponds to an "investor" who invests in the company or an investor who provides capital. The supervisor AI model 101 corresponds to the company's management team, and the AI ​​model group 110 of the corporate AI model and the external AI model group 120 correspond to the company organization.

[0060] The user instructs the supervisor AI model 101 on the task objective and the constraints associated with accomplishing the objective. For example, the user instructs the objective "to fill the world with emotion through the power of creativity and technology." The user may impose constraints when instructing the objective. For example, the user may impose constraints such as budget, deadline, sales (or other indicators that can quantify results). Because the supervisor AI model 101 is a multimodal AI model, the user can instruct the objective and constraints in any modality.

[0061] The supervisor AI model selects an AI model to delegate the execution of the above task from among the AI ​​model group 110 and the external AI model group 120, specifies the delegation content for each AI model, and sends an execution request.

[0062] Among the AI ​​model group 110 and the external AI model group 120, each AI model delegated with the task objective cooperates with others to accomplish its assigned role toward the objective. The actions that the AI ​​models that accomplish the task objective take on the real world depend on the industry of the corporate AI model. In the manufacturing industry, the AI ​​model group 110 and the external AI model group 120 output products, while in the service industry, the AI ​​model group 110 and the external AI model group 120 output services. In the content business, the AI ​​model group 110 and the external AI model group 120 output content promotions. In the home appliance and AV product manufacturing industry, the AI ​​model group 110 and the external AI model group 120 output product promotions. In the medical industry, for example, a pathologist AI model outputs image diagnosis results. In the culinary industry, for example, a food coordinator AI model or a food advisor AI model outputs new menu development and advice. Furthermore, the AI ​​model group may be a group of AI models for a specific department rather than for the entire company. For example, a human resources AI model outputs information on new and mid-career recruitment, work management, employee evaluations, etc. Furthermore, the AI ​​model group may be a product AI model rather than the above-mentioned corporate AI model. For example, an automobile AI model outputs vehicle anomaly detection results by a vehicle body AI model and collision prediction results based on camera sensing results by a collision prediction AI model.

[0063] Then, each delegated AI model among the AI ​​model group 110 and the external AI model group 120 senses real-world reactions to the performed action and reports to the supervisor AI model 101. Any method can be used to sense real-world reactions. As real-world reactions, feedback including comment information such as criticisms and word-of-mouth posted on social media may be sensed. Multiple AI models operating autonomously and cooperatively cooperate with each other to create a report on the results of task objective accomplishment for the supervisor AI model 101.

[0064] The supervisor AI model 101 may reselect an AI model to which to delegate the performance of the task objective based on reports received from the AI ​​model group 110 and the external AI model group 120. Reselection of an AI model includes deleting or deselecting a previously selected (i.e., created) AI model, and adding a new AI model. In this case, the supervisor AI model 101 specifies the delegation details for each reselected AI model, and sends a new performance request to the AI ​​model group 110 and the external AI model group 120.

[0065] The supervisor AI model 101 then reports to the user who specified the objective based on the reports received from the AI ​​model group 110 and the external AI model group 120. The report from the supervisor AI model 101 to the user is just like a management team reporting performance to investors.

[0066] B-2. Example of Application to Creative Activities Figure 5 shows an example in which the system 100 according to the present disclosure is applied to creative activities. "Creation" here includes the creation of artwork, content (such as movies, TV shows such as dramas), and games. In this case, the user who specifies the objective corresponds to an "investor" who invests in the creation. The supervisor AI model 101 corresponds to the management of a company engaged in creative activities, and the AI ​​model group 110 and the external AI model group 120 correspond to groups of creator AI models and artist AI models.

[0067] The user instructs the supervisor AI model 101 on the task objective and the constraints associated with accomplishing the objective. For example, the user instructs the objective as "a pioneering group of artists who discover new genres of art." The user may impose constraints when instructing the objective. For example, the user may impose constraints such as budget, deadline, and sales (or other indicators that can quantify results). Because the supervisor AI model 101 is a multimodal AI model, the user can instruct the objective and constraints in any modality.

[0068] The supervisor AI model 101 selects an AI model to delegate the execution of the above task from among the AI ​​model group 110 and the external AI model group 120, specifies the delegation content for each AI model, and sends an execution request.

[0069] Among the AI ​​model group 110 and the external AI model group 120, each AI model delegated to accomplish the task goal cooperates with each other to accomplish its assigned role toward the goal. The AI ​​models accomplishing the task goal output new artwork, new content, new games, etc. as actions taken in the real world by the AI ​​models. The AI ​​model group 110 and the external AI model group 120 then sense real-world reactions to the actions they have taken and report them to the supervisor AI model 101. Any method can be used to sense real-world reactions. Feedback, including comment information such as criticisms and word-of-mouth posted on social media, may be sensed as real-world reactions. Multiple AI models operating autonomously and cooperatively cooperate with each other to create a report on the results of accomplishing the task goal for the supervisor AI model 101.

[0070] The supervisor AI model 101 may reselect an AI model to delegate the performance of the task objective based on reports received from the AI ​​model group 110 and the external AI model group 120. Reselection of an AI model includes deleting or deselecting a previously selected (i.e., created) AI model, and adding a new AI model. In this case, the supervisor AI model 101 specifies the delegation details for each reselected AI model and sends a performance request. Then, the supervisor AI model 101 reports to the user who specified the objective based on reports received from the AI ​​model group 110 and the external AI model group 120.

[0071] C. Prerequisite AI Model AI models vary from general-purpose, powerful general-purpose types to specialized types that achieve high quality at low cost for specific purposes, and types with low performance and significantly lower cost. In the system 100 according to the present disclosure, the prerequisite AI model for the supervisor AI model 101 is a multimodal, multitask AI model.

[0072] In this disclosure, the multimodal AI model applied as the supervisor AI model is assumed to be an interactive AI model that can input and interpret diverse information and generate and output similarly diverse information. Transformer is a model architecture originally designed for processing text (GPT-4, when first released, was a text-specific AI model that utilized Transformer), but recent Transformer-based AI models can handle the input of diverse types of data by converting and inputting various data, such as characters, words, images, audio, and video fragments, into tokens, and can also output diverse types of data.

[0073] C-1. Input / Output Relationships of the AI ​​Model Figure 6 illustrates the input / output relationships of the supervisor AI model applied to the present disclosure. The supervisor AI model shown in the figure can utilize and create tools itself through program code output, etc., and can also influence the external world, including the real world. The supervisor AI model applied to the present disclosure is an AI model that, like humans who work every day, inputs and processes various information such as various information displayed on a monitor, text contained in audio played from a speaker, images, audio, files, sensor information from cameras, and handwritten instructions, processes and judges it, freely operates a computer, and outputs various information such as text, diagrams (images), and audio to communicate with users and other AIs. In other words, the supervisor AI model applied to the present disclosure is a versatile AI model that can flexibly process and judge like a human.

[0074] The supervisor AI model applied to the present disclosure carries out the objectives instructed by the user, and the following three can be listed as the main external functions (functions that can be instructed to the AI ​​model group) that can be used in this case.

[0075] Create AI: Create an AI to handle the work Delegate AI: Request a task from an AI Delete AI: Delete or deselect an AI to handle the work

[0076] The user can input data of various modalities, such as the role of the AI ​​model, text, video, images, audio, files, sensor information such as camera, and handwritten instructions, as input prompts for the supervisor AI model. The user can instruct the supervisor AI model to execute the above-mentioned external functions Create AI, Delegate AI, and Delete AI, as well as other functions typically including internet access and program code execution. In response to the user's instructions, the supervisor AI model can generate output to the AI ​​models that includes data of various modalities, such as text, images, video, and audio, as well as instructions to call external functions.

[0077] C-2. Specific examples of input and output of AI models For example, a user can input instructions for the creation of game content into the supervisor AI model along with text such as the following, along with data of various modalities such as related videos (images of users enjoying the game) and files.

[0078] User instructions: "Create game content that fosters intellectual curiosity and release it on the XXX platform. We will send you an image of users enjoying the game. Monetize it without compromising the gaming experience, and maximize MAU (Monthly Active Users) as a KPI (Key Performance Indicator). The budget for monetization is 100,000 yen."

[0079] In response to the user's instructions, the supervisor AI model uses the external function Create AI to create AI models that fulfill the roles of planner, screenwriter, and artist.The supervisor AI model then uses the external function Delegate AI to request the AI ​​models fulfilling the roles of planner, screenwriter, and artist to carry out a task of discussing and planning a game among the three AI models.

[0080] Create AI: Planning Create an AI that will play the role of a game planner. Create AI: Scriptwriter Create an AI that will play the role of a scriptwriter who comes up with the game's story. Create AI: Artist Create an AI that will play the role of an artist who will create assets such as images and videos to be used in the game. The Artist AI will follow the instructions of the Project Leader and carry out the task of preparing the assets necessary for the game devised by the Planner AI. Delegate AI: The Planner AI, Scriptwriter AI, and Artist AI will consult together to come up with a game plan.

[0081] The Supervisor AI model can also generate the following text to report to the user about calling the above external functions Create AI and Delegate AI:

[0082] "In order to achieve this goal, I have formed a team consisting of a planner, a scriptwriter, and an artist. I have instructed the three parties to consult together and consider the game's plans."

[0083] D. Specific Example D-1. Specific System Configuration Figure 7 shows a specific configuration example of a system 100 according to the present disclosure. In addition to a supervisor AI model 101, a group of AI models 110 that operate autonomously and cooperatively, and a group of external AI models 120, the system 100 further includes a UI device 103, an execution result recording device 104, and a resource management device 105.

[0084] The UI (User Interface) device 103 is a device that accepts input from a user and presents information to the user. Through the UI device 103, the user can input instructions regarding task objectives and constraints (budget, cost, deadline, etc.) to the system 100 and receive reports on the results of task execution from the system 100. The input and output of the UI device 103 is multimodal data. The UI device 103 basically receives the UI generated by the supervisor AI model 101 via the communication path 106 and presents it to the user.

[0085] The supervisor AI model 101 and the UI device 103 are interconnected via a communication path 106. The communication path 106 may be configured using either a wired or wireless communication medium. The communication path 106 may also include a wide area network such as the Internet. The UI device 103 transmits the task objective and constraints input by the user to the supervisor AI model 101 via the communication path 106.

[0086] When the supervisor AI model 101 receives the task objective and constraint conditions input to the UI device 103 via the communication path 106, it selects an AI model to delegate the execution of the task objective under the constraint conditions from the AI ​​model group 110 and the external AI model group 120, specifies the delegation content to be delegated to the AI ​​model for the execution of the task objective, and transmits a request for execution of the delegation content to the AI ​​model group 110 and the external AI model group 120. The delegation content transmitted by the supervisor AI model 101 to the AI ​​model group 110 and the external AI model group 120 includes Create AI, which instructs the creation of an AI that will perform the work, Delegate AI, which requests a task to an AI, and Delete AI, which instructs the deletion or deselection of an AI that will perform the work. The AI ​​models that Create AI instructs to create may include AI models that perform external access to the Internet, etc., and ethics supervision AI models (see section A-2 above) that have learned the processes required to comply with the laws and regulations of the relevant country or region and resolve ethical issues.

[0087] The supervisor AI model 101 can read past task execution history data from the execution result recording device 104, and can read resource information such as the cost spent by each AI model on task processing from the resource management device 105. The supervisor AI model 101 can pre-learn resource information such as past task execution history data by AI models and the cost required for task processing, and when a user instructs on a new task or constraints such as budget or cost, it can estimate the optimal AI model to perform the instructed task and instruct the creation, addition, deletion, or non-selection of an AI based on the estimation result.

[0088] The supervisor AI model 101 also receives task execution result data by the AI ​​models delegated to carry out the task purpose from the AI ​​model group 110 and the external AI model group 120. The supervisor AI model 101 creates a UI for reporting the task execution result data received from the AI ​​model group 110 and the external AI model group 120 to the user, transmits the UI to the UI device 103 via the communication path 106, and records the task execution result data in the execution result recording device 104. The task execution result data and execution history data may include the cost and period required to execute the task, performance such as MAU, business performance, real-world sensing results (comment information such as criticism and word-of-mouth posted on SNS media, ethics check results, etc.), expertise, etc.

[0089] The AI ​​model group 110 and the external AI model group 120 include AI model groups consisting of multiple AI models that operate autonomously and cooperatively. When the AI ​​model group 110 and the external AI model group 120 receive from the supervisor AI model 101 a Create AI instruction to create an AI that will perform a practical task or a Delete AI instruction to delete or deselect an AI that will perform a practical task, they add, delete, or deselect an AI model to the AI ​​model group 110 and the external AI model group 120. When the AI ​​model group 110 and the external AI model group 120 receive a Delegate AI instruction from the supervisor AI model 101 that requests an AI to perform a task, they cause the AI ​​model created or added to the AI ​​model group by Create AI to perform the delegated task. The delegated multiple AI models interact with each other, cooperate with each other, fulfill their assigned roles toward a goal, and take action in the real world. The AI ​​model group 110 and the external AI model group 120 may perform external access to the Internet, etc., to accomplish delegated tasks. The AI ​​model group may include an AI model that performs external access to the Internet, etc., and the supervisor AI model 101 may instruct the creation of an AI for external access.

[0090] The AI ​​model group 110 and the external AI model group 120 also sense real-world reactions to actions performed in the real world and include the sensing results in a report to the supervisor AI model 101. Real-world reactions may be sensed by comment information such as criticism or word-of-mouth posted on social media. The AI ​​model group 110 and the external AI model group 120 may also sense task execution results or ethical issues caused by real-world actions. The AI ​​model group 110 and the external AI model group 120 then store resource information, such as the costs incurred by the AI ​​models to process delegated tasks, in the resource management device 105.

[0091] The execution result recording device 104 records the execution result data of the task received from the supervisor AI model 101. The supervisor AI model 101 can read the execution history data of past tasks recorded in the execution result recording device 104.

[0092] The resource management device 105 stores resource information such as the costs incurred by the AI ​​models in the AI ​​model group 110 and the external AI model group 120 in processing tasks delegated by the supervisor AI model 101. The supervisor AI model 101 can read the resource information stored in the resource management device 105 during task processing.

[0093] In addition, the supervisor AI model 101, the AI ​​model group 110, the external AI model group 120, the UI device 103, the execution result recording device 104, and the resource management device 105 may each be an individual information processing device that is physically separated, or at least two or more of these devices may be configured as a physically integrated information processing device.

[0094] For example, the supervisor AI model 101, the AI ​​model group 110, the external AI model group 120, the execution result recording device 104, and the resource management device 105 may be physically integrated into an information processing device, with only the UI device 103 being an information terminal outside the information processing device. Such an information processing device may be a server on the cloud, and the UI device 103 may be an information terminal such as a smartphone or personal computer (PC) owned by a user. A user can input the purpose and constraints of a task using their smartphone and instruct the server to execute the task. In addition, a user can receive UI information presenting task execution history data from the server using their smartphone and display it on the smartphone screen for confirmation.

[0095] D-2. Specific System Operation Figure 8 shows an example of a processing sequence for performing a task, which is implemented in the system 100 shown in Figure 7. However, Figure 8A shows the first half of the processing sequence, and Figure 8B shows the first half of the processing sequence.

[0096] First, the user inputs the objective of the desired task and constraints for performing the task using the UI device 103. The user can input using various modalities, such as text, video, images, audio, files, sensor information from a camera, or handwritten instructions (SEQ 601). The UI device 103 transmits the objective and constraints of the task input by the user to the supervisor AI model 101 via the communication path 106 (not shown in FIG. 6) (SEQ 602).

[0097] When the supervisor AI model 101 receives the task objective and constraints input to the UI device 103 via the communication path 106, it selects an AI model to delegate the task objective execution under the constraints and specifies the delegation content to be delegated to the AI ​​model for the task objective execution (SEQ 611). The delegation content includes Create AI, which instructs the creation of an AI to perform the work, Delegate AI, which requests a task from an AI, and Delete AI, which instructs the deletion or deselection of an AI to perform the work. The supervisor AI model 101 then transmits a request to execute the delegation content to the AI ​​model group 110 and the external AI model group 120 (SEQ 612). The AI ​​models instructed to be created by Create AI may include an AI model that performs external access to the Internet, etc., and an ethics supervision AI model (see section A-2 above) that has learned processing to comply with the laws and regulations of the relevant country or region and resolve ethical issues.

[0098] When selecting an AI model, past task execution history data of the AI ​​model read from the execution result recording device 104 and resource information of the AI ​​model read from the resource management device 105 (such as the cost spent to perform the task) may be taken into consideration (SEQ 610). The task execution history data may include the cost and period required to perform the task, performance such as MAU, business performance, real-world sensing results (comment information such as criticism and word-of-mouth posted on SNS media, ethics check results, etc.), expertise, etc.

[0099] The AI ​​model group 110 and the external AI model group 120 create AI models based on the delegation content received from the supervisor AI model 101 (SEQ 621). In the example shown in FIG. 8 , the AI ​​model group 110 and the external AI model group 120 create AI models A to C based on the delegation content from the supervisor AI model 101. The AI ​​model group 110 and the external AI model group 120 accomplish the purpose of the delegated task through autonomous cooperative operation between the created AI models (SEQ 622). Each AI model may operate sequentially or in parallel. The supervisor AI model 101 may generate an ethics supervision AI model (see section A-2 above). The ethics supervision AI model supervises other AI models created at the same time to ensure compliance with the laws and regulations of the relevant country or region when performing tasks based on the delegation content, thereby preventing ethical issues from arising.

[0100] The AI ​​model group 110 and the external AI model group 120 take action in the real world and sense reactions from the real world (SEQ 623). As reactions to the real world, the AI ​​model group 110 and the external AI model group 120 may sense comment information such as criticism and word-of-mouth posted on social media. Here, the AI ​​model group 110 and the external AI model group 120 may sense task execution results and ethical issues caused by real-world actions. The AI ​​model group 110 and the external AI model group 120 then transmit task execution result data and real-world sensing results to the supervisor AI model 101 (SEQ 624). A summary process may be inserted into the execution result data as needed to reduce the burden on the supervisor AI model 101 for understanding the data. The execution result data may be summarized in a report or the like and included in the execution history data.

[0101] The supervisor AI model 101 creates a UI for reporting the task execution result data received from the AI ​​model group 110 and the external AI model group 120 to the user (SEQ 613) and transmits it to the UI device 103 via the communication path 106 (SEQ 614). The UI device 103 presents the UI received from the supervisor AI model 101 to the user (SEQ 603). In addition, the supervisor AI model 101 records the task execution result data and real-world sensing results received from the AI ​​model group 110 and the external AI model group 120 in the execution result recording device 104 (SEQ 615).

[0102] Next, in response to a response from the user to whom the UI is presented (SEQ604), or based on the supervisor AI model 101's own evaluation of the task execution result data (SEQ616), the supervisor AI model 101 decides to reconfigure the AI ​​model that will execute the task, specifies the delegation content to be delegated to the reconfigured AI model (SEQ617), and resends a request to execute the delegation content to the AI ​​model group 110 and the external AI model group 120 (SEQ618). Reconfiguration of an AI model includes deleting or deselecting an AI that will perform the work, and adding an AI that will perform the work. When deleting or deselecting an AI that will perform the work, the delegation content includes Delete AI, and when adding an AI that will perform the work, the delegation content includes Create AI. The AI ​​model group 110 and the external AI model group 120 create, delete, or deselect an AI model based on the delegation content received from the supervisor AI model 101 (SEQ625). In the example shown in FIG. 6, the AI ​​model group 110 and the external AI model group 120 delete or deselect AI model C from the existing AI models A to C and create an additional AI model D based on the new delegation content from the supervisor AI model 101. The AI ​​model group 110 and the external AI model group 120 accomplish the purpose of the delegated task through autonomous cooperative operation between the created AI models (SEQ 626). Each AI model may operate sequentially or in parallel. The supervisor AI model 101 may generate an ethics supervision AI model (see section A-2 above). The ethics supervision AI model supervises other AI models generated at the same time to ensure compliance with the laws and regulations of the relevant country or region when performing tasks based on the delegation content, thereby preventing ethical issues from arising.

[0103] The presentation of the task execution result data to the user and the reconfiguration of the AI ​​model that executes the task are repeated until an event occurs that ends the task. Events that end the task include, for example, the user being satisfied with the task execution result data presented by the UI device 103, or the supervisor AI model 101 determining that the task has been completed based on the evaluation result of the task execution result data.

[0104] G. Generation of UI Screen The UI device 103 displays a UI screen that allows the user to input instructions regarding task objectives and constraints (budget, cost, deadline, etc.) to the system 100 and receive reports on task execution results from the system 100. The input and output of the UI device 103 are multimodal data. The supervisor AI model 101 generates a UI and applies it to the UI device 103.

[0105] G-1. UI Configuration Methods Figure 9 shows three UI configuration methods. Figure 9A shows a simple UI configuration method. In this method, the supervisor AI model 101 may present text and UI to the user through a pre-prepared Web UI.

[0106] 9B shows another simple method for configuring a UI. In this method, the supervisor AI model 101 may generate a program that realizes a user interface for communicating with the user and output this to the UI device 103. In this case, the UI device 103 displays a UI screen generated by the supervisor AI model 101, and the user can input to the supervisor AI model 101 via this UI screen.

[0107] FIG. 9C also shows an ultimate UI configuration method in which the product of the supervisor AI model 101 is directly presented to the user and the user directly inputs data to the supervisor AI model 101. However, more strictly speaking, the product of the supervisor AI model 101 is a product generated by the supervisor AI model 101 through delegation to the AI ​​model group 110 and the external AI model group 120. The UI configuration method shown in FIG. 9C differs from the examples shown in FIGS. 9A and 9B in that no Web UI or program is interposed between the supervisor AI model 101 and the user. In this case, the user can directly input various information of different modalities, including images input from a camera and audio input from a microphone, to the supervisor AI model 101. Furthermore, the supervisor AI model 101 can directly output data including video, audio, and UI to the user via a UI screen.

[0108] G-2. Specific Examples of UI Screens G-2-1. UI for User Input FIG. 10 shows an example of the configuration of a UI screen through which a user inputs instructions to the supervisor AI model 101.

[0109] The illustrated UI screen 800 has, in its upper half, a camera input toggle button 801 for switching camera input on and off, an audio input toggle button 802 for switching audio input on and off, and a file input section 803 for specifying a file. A desired file can be specified by dragging its icon into the file input section 803. In addition, the lower half of the UI screen 800 has a text input field 804 for entering text. When a send button 805 at the bottom right of the UI screen 800 is clicked, the input content of the UI screen 800 is sent from the UI device 103 to the supervisor AI model 101 (supervisor AI model).

[0110] The user can send instructions to the supervisor AI model 101 via the UI screen shown in Figure 10 using multimodal data such as text input, audio input, video input, and files.

[0111] Note that the voice input is not converted into text after being recognized, but is input directly as voice data. This allows data that is difficult to express as text, such as music and sound effects, to be directly input to the supervisor AI model 101.

[0112] In addition, with video input, for example, a user's webcam image can be directly input, allowing instructions to be given to the supervisor AI model 101 through gestures and other means.

[0113] In addition, in the file input, useful materials and other reference information for instructing the supervisor AI model 101 can be input as a file.

[0114] G-2-2. UI for Output to User Figure 11 shows an example of the configuration of a UI screen that presents a response from the supervisor AI model 101 to the user.

[0115] The illustrated UI screen 900 has a text display field 901 in the upper half that presents text responses, and an AV playback field 902 in the lower half that presents image / video and audio responses. Clicking the play button in the AV playback field 902 starts video playback. Also, a download button 903 that instructs downloading of related material files is located in the lower right portion of the UI screen 900. Clicking the download button 903 downloads the file to the UI device 103 (or a specified folder).

[0116] The supervisor AI model 101 can respond to the user with multimodal data such as images, audio, video, and files in addition to text via the UI screen shown in FIG.

[0117] Each AI model to which the supervisor AI model 101 delegates task execution executes the delegated task asynchronously, and actively asks questions and provides information to the user without waiting for user input.

[0118] The supervisor AI model 101 can generate and present a UI to the user regarding the report of task execution results so that the user can easily understand the report or so that the user can give instructions to the supervisor AI model 101 with simpler operations. For example, the supervisor AI model 101 can generate a UI based on user instructions, such as graphing and displaying important indicators such as MAU and income and expenditure, as shown in FIG. 12 , and displaying the UI on the UI device 103, which allows typical instructions to be instantly called up with a button. Clicking the "More detailed MAU report" button on the illustrated UI screen switches to a UI screen displaying a more detailed MAU report. Clicking the "More detailed income and expenditure report" button also switches to a UI screen displaying a more detailed income and expenditure report.

[0119] The supervisor AI model 101 can also generate a UI customized to suit the user and the purpose. That is, the supervisor AI model 101 can infer the task purpose and constraints entered by the user into the UI device 103, the content delegated to the AI ​​model group 110 and the external AI model group 120, the execution results returned from the AI ​​model group 110 and the external AI model group 120, and the user's state, and based on the inference results, identify the output format and UI format for presenting the task execution results to the user, and generate a UI in a format that suits the user's state and the user's input.

[0120] For example, the supervisor AI model 101 can generate a UI that directly presents a UI screen including a user input interface to the user in the form of a real-time video avatar with audio, and directly accepts feedback from the user. The UI screen customized for the user is preferably configured so that the supervisor AI model directly accepts not only user text input and file upload, but also touch screen and mouse operations, speech input from a microphone, and gestures input from a camera or sensor.

[0121] An example of one configuration of a UI screen customized for a user is shown in Figure 13. The UI screen shown is the exact output of the supervisor AI model, presenting avatar video and audio along with a MAU graph and a "More MAU Report" button.

[0122] H. Configuration of Information Processing Device The supervisor AI model 101, the AI ​​model group 110, the external AI model group 120, the UI device 103, the execution result recording device 104, and the resource management device 105, which are components of the system 100, may each be physically separated individual information processing devices, or at least two or more of these devices may be configured as a physically integrated information processing device. For example, the supervisor AI model 101, the AI ​​model group 110, the external AI model group 120, the execution result recording device 104, and the resource management device 105 may be physically integrated into an information processing device.

[0123] 14 shows an example of the hardware configuration of an information processing device 2000 applied to the present disclosure. This information processing device 2000 includes a CPU (Central Processing Unit) 2001, a ROM (Read Only Memory) 2002, a RAM (Random Access Memory) 2003, a host bus 2004, a bridge 2005, an expansion bus 2006, an interface unit 2007, an input unit 2008, an output unit 2009, a storage unit 2010, a drive 2011, and a communication unit 2013. The information processing device 2000 is configured, for example, by a personal computer, but some of its functions may be configured by an information terminal such as a tablet or a smartphone.

[0124] The CPU 2001, ROM 2002, and RAM 2003 are interconnected by a host bus 2004, which is composed of a CPU bus and other components. The CPU 2001 controls the overall operation of the information processing device 2000 in accordance with various programs. The ROM 2002 stores programs (such as a basic input / output system) and calculation parameters used by the CPU 2001 in a non-volatile manner. The RAM 2003 is used to load programs to be executed by the CPU 2001 and to temporarily store parameters such as work data that change as appropriate during program execution. The CPU 2001 can execute various application programs in an execution environment provided by an operating system (OS) through the cooperative operation of the ROM 2002 and RAM 2003, thereby realizing a variety of functions and services.

[0125] If the information processing device 2000 is a PC, the OS may be, for example, Microsoft Windows (registered trademark), Unix (registered trademark), or a successor OS. The programs loaded into the RAM 2003 and executed by the CPU 2001 include an OS and various application programs. The application programs or some modules in the application programs may use existing libraries that are stored, shared, or made public through, for example, a source code management service. For example, programs for operating as the supervisor AI model 101, the AI ​​model group 110, the external AI model group 120 (corporate AI model), the execution result recording device 104, and the resource management device 105 shown in FIG. 1 are executed on the information processing device 2000.

[0126] When performing computationally intensive processing such as learning an AI (Artificial Intelligence) model on the information processing device 2000, it is desirable that the CPU 2001 be a multi-core CPU (for example, Apple M1 Max, etc.), and that the information processing device 2000 further be equipped with a multi-core processor such as a GPU or GPGPU (General-purpose computing on graphics processing units) (for example, NVIDIA's "Quadro A6000"). However, for convenience, these will be collectively referred to as the CPU 2001 below.

[0127] The host bus 2004 is connected to an expansion bus 2006 via a bridge 2005. The expansion bus 2006 is, for example, a PCI (Peripheral Component Interconnect) bus or PCI Express, and the bridge 2005 is based on the PCI standard. However, the information processing device 2000 does not need to be configured so that the circuit components are separated by the host bus 2004, bridge 2005, and expansion bus 2006, and may be implemented so that almost all circuit components are interconnected by a single bus (not shown).

[0128] The interface unit 2007 connects peripheral devices such as an input unit 2008, an output unit 2009, a storage unit 2010, a drive 2011, and a communication unit 2013 in accordance with the standards of the expansion bus 2006. However, not all of the peripheral devices shown in Fig. 14 are necessarily required, and the information processing device 2000 may further include peripheral devices not shown. Furthermore, the peripheral devices may be built into the main body of the information processing device 2000, or some of the peripheral devices may be externally connected to the main body of the information processing device 2000.

[0129] The input unit 2008 is composed of an input control circuit that generates an input signal based on input from a user and outputs it to the CPU 2001. When the information processing device 2000 is a personal computer, the input unit 2008 may include a keyboard, a mouse, a touch panel, a camera, and a microphone. The output unit 2009 includes, for example, a display device such as a liquid crystal display (LCD), an organic electroluminescence (EL) display, or an LED (light emitting diode), as well as an audio output device such as a speaker. The input unit 2008 receives video to be processed, and the output unit 2009 displays GUI screens such as those shown in FIGS. 10 to 13. For example, when the information processing device 2000 operates as the UI device 103, the input unit 2008 and the output unit 2009 can be used to present a UI screen including a user input interface and a video of an avatar with audio to the user in real time, thereby directly receiving feedback from the user. In addition to user text input and file uploads, the device can be configured to directly accept touches on the screen, mouse operations, speech input from a microphone, and gestures input from a camera or sensor.

[0130] The storage unit 2010 stores files such as programs (applications, OS, etc.) executed by the CPU 2001 and various data. The storage unit 2010 is configured with a large-capacity storage device such as an SSD (Solid State Drive) or an HDD (Hard Disk Drive), but may also include an external storage device.

[0131] The removable storage medium 2012 is a storage medium configured as a cartridge, such as a microSD card. The drive 2011 performs read and write operations on the loaded removable storage medium 2012. The drive 2011 outputs data read from the removable storage medium 2012 to the RAM 2003 or the storage unit 2010, and writes data on the RAM 2003 or the storage unit 2010 to the removable storage medium 2012.

[0132] The communication unit 2013 is a device that performs wireless communication such as Wi-Fi (registered trademark), Bluetooth (registered trademark), and cellular communication networks such as 4G and 5G. The communication unit 2013 may also include terminals such as a Universal Serial Bus (USB) and a High-Definition Multimedia Interface (HDMI) (registered trademark), and may further include a function for performing HDMI (registered trademark) communication with USB devices such as scanners and printers, displays, etc. Programs executed on the information processing device 2000 are installed from the outside, for example, via the communication unit 2013.

[0133] The present disclosure has been described in detail above with reference to specific embodiments. It should be fully understood that the present disclosure can provide an information processing method, an information processing device, and a computer program that provide services through the autonomous cooperative operation of multiple AI models. Note that the effects described in this specification are merely examples, and the effects brought about by the present disclosure are not limited thereto. Furthermore, the present disclosure may achieve additional effects in addition to the above-described effects. Further objects, features, and advantages of the present disclosure will become apparent from the above-described embodiments and a more detailed description based on the accompanying drawings.

[0134] The present disclosure should not be interpreted as being limited to the above-described embodiments, and it is obvious that those skilled in the art can modify or substitute the embodiments without departing from the spirit of the present disclosure. Furthermore, the effects described in this specification are merely examples, and the effects brought about by the present disclosure are not limited thereto, and additional effects not described in this specification may exist.

[0135] Although this specification introduces examples in which the present disclosure is applied to corporate activities and creative activities, the scope of application of the present disclosure is not limited thereto. The present disclosure can address the objectives of various tasks in the real world, select an appropriate AI model taking into consideration imposed constraints, perform tasks optimally through autonomous cooperative operation between the AI ​​models, and present a UI showing task execution history data to the user.

[0136] In short, the present disclosure has been described in the form of examples, and the contents of the specification should not be interpreted as limiting. To determine the gist of the present disclosure, the claims should be taken into consideration.

[0137] The series of processes described in this specification can be executed by hardware, software, or a configuration that combines hardware and software. When executing processes by software, a program recording a processing sequence related to realizing the present disclosure is installed in memory in a computer incorporated in dedicated hardware and executed. It is also possible to install the program in a general-purpose computer capable of executing various processes and execute the processes related to realizing the present disclosure.

[0138] The program can be stored in advance on a recording medium installed in the computer, such as a HDD, SSD, or ROM. Alternatively, the program can be temporarily or permanently stored on a removable recording medium such as a flexible disk, CD-ROM (Compact Disc Read Only Memory), MO (Magneto Optical) disk, DVD (Digital Versatile Disc), BD (Blu-Ray Disc (registered trademark)), magnetic disk, or USB (Universal Serial Bus) memory. Using such a removable recording medium, a program related to the realization of the present disclosure can be provided as so-called package software.

[0139] The program may also be transferred wirelessly or via a wire from a download site to a computer via a network such as a wide area network (WAN) typified by cellular, a local area network (LAN), the Internet, etc. The computer can receive the program transferred in this manner and install it in a large-capacity storage device such as an HDD or SSD within the computer.

[0140] The present disclosure may also be configured as follows.

[0141] (1) An information processing method comprising: a receiving step of receiving an input including a task purpose and constraints; a specifying step of identifying a delegatee model and delegation content based on the input; a transmitting step of transmitting the delegation content to the identified delegatee model; a result receiving step of receiving an execution result of the delegation content from the delegatee model; and a presenting step of presenting information based on the execution result.

[0142] (2) The information processing method according to (1), wherein the identifying step identifies the delegate model from among a plurality of AI models.

[0143] (3) The information processing method according to any one of (1) or (2), wherein in the specifying step, the delegated model is specified from among a plurality of AI models based on the delegation content and execution history data.

[0144] (4) The information processing method described in (3) above, wherein the execution history data includes at least one of the cost and time required to execute the task, performance, achievements, real-world sensing results (comment information such as criticisms and word-of-mouth posted on social media, ethics check results, etc.), and expertise.

[0145] (5) The information processing method according to any one of (1) to (4), wherein each AI model identified as the delegate model performs a task through autonomous cooperative operation.

[0146] (6) The information processing method according to any one of (1) to (5), wherein the receiving step receives input including a task objective and constraints in a multimodal format, and the presenting step presents information based on the execution result in a multimodal format.

[0147] (7) In the information processing method described in (6) above, in the presentation step, a format for presenting information based on the execution result to the user is identified based on at least one of the content of the input received in the reception step, the execution result received in the result receiving step, and the result of estimating the state of the user who made the input, and the information is presented in a format that is compatible with the state of the user and the user's input.

[0148] (8) The information processing method according to any one of (1) to (7), wherein, when the activity of a specific company is specified as the purpose of the task, in the identification step, an AI model that performs the role of each employee who performs the task in the company is identified as the delegate model.

[0149] (9) The information processing method according to any one of (1) to (8), wherein the specifying step specifies an AI model to be generated as the delegated model and an AI model to be deleted or not selected.

[0150] (10) The information processing method according to any one of (1) to (9), wherein each step is performed using a multimodal AI model.

[0151] (11) An information processing device comprising: a receiving unit that receives input including a task purpose and constraints; an identifying unit that identifies a delegatee model and delegation content based on the input; a transmitting unit that transmits the delegation content to the identified delegatee model; a result receiving unit that receives an execution result of the delegation content from the delegatee model; and a presenting unit that presents information based on the execution result.

[0152] (12) A computer program written in a computer-readable format to cause a computer to function as: a receiving unit that receives input including the purpose and constraints of a task; an identifying unit that identifies a delegatee model and delegation content based on the input; a transmitting unit that transmits the delegation content to the identified delegatee model; a result receiving unit that receives the execution results of the delegation content from the delegatee model; and a presenting unit that presents information based on the execution results.

[0153] 100...system, 101...supervisor AI model, 103...UI device, 104...execution result recording device, 105...resource management device, 106...communication path, 110...AI model group, 120...external AI model group, 2000...information processing device, 2001...CPU, 2002...ROM, 2003...RAM, 2004...host bus, 2005...bridge, 2006...expansion bus, 2007...interface unit, 2008...input unit, 2009...output unit, 2010...storage unit, 2011...drive, 2012...removable recording medium, 2013...communication unit

Claims

1. An information processing method comprising: a receiving step of receiving input including a task purpose and constraints; a specifying step of identifying a delegatee model and delegation content based on the input; a sending step of sending the delegation content to the identified delegatee model; a result receiving step of receiving the execution results of the delegation content from the delegatee model; and a presentation step of presenting information based on the execution results.

2. The information processing method according to claim 1, wherein the identifying step identifies the delegate model from among a plurality of AI models.

3. The information processing method according to claim 1, wherein in the specifying step, the delegated model is specified from among a plurality of AI models based on the delegation content and execution history data.

4. The information processing method according to claim 3, wherein the execution history data includes at least one of the cost and time required to execute a task, performance, achievements, real-world sensing results, and expertise.

5. The information processing method according to claim 1, wherein each AI model identified as the delegate model performs a task through autonomous cooperative operation.

6. The information processing method according to claim 1, wherein the receiving step receives input including the purpose and constraints of the task in a multimodal format, and the presenting step presents information based on the execution result in a multimodal format.

7. The information processing method according to claim 6, wherein in the presentation step, a format for presenting information based on the execution result to the user is identified based on at least one of the content of the input received in the reception step, the execution result received in the result reception step, and the result of estimating the state of the user who made the input, and the information is presented in a format that is compatible with the state of the user and the user's input.

8. The information processing method of claim 1, wherein, when a specific company activity is specified as the purpose of the task, the identification step identifies an AI model that performs the role of each employee who performs the task in the company as the delegate model.

9. The information processing method according to claim 1, wherein the specifying step specifies an AI model to be generated as the delegated model and an AI model to be deleted or not selected.

10. The information processing method of claim 1, wherein each step is performed using a multimodal AI model.

11. An information processing device comprising: a receiving unit that receives input including a task purpose and constraints; an identifying unit that identifies a delegatee model and delegation content based on the input; a transmitting unit that transmits the delegation content to the identified delegatee model; a result receiving unit that receives execution results of the delegation content from the delegatee model; and a presentation unit that presents information based on the execution results.

12. A computer program written in a computer-readable format to cause a computer to function as: a receiving unit that receives input including the purpose and constraints of a task; an identifying unit that identifies a delegatee model and delegation content based on the input; a transmitting unit that transmits the delegation content to the identified delegatee model; a result receiving unit that receives the execution results of the delegation content from the delegatee model; and a presentation unit that presents information based on the execution results.

Images