System, method and program for facilitating interactions between a plurality of ai agents

Abstract

Providing a system and the like for facilitating interactions between a plurality of AI agents SOLUTION: A system for facilitating interactions between a plurality of AI agents comprises assigning means for assigning a score to each AI agent of the plurality of AI agents, the score of the AI agent characterizing the AI agent with at least one metric, and determining means for determining, when a first AI agent of the plurality of AI agents receives a task, at least one AI agent to interact with the first AI agent based on the score of each of the plurality of AI agents.SELECTED DRAWING: Figure 1B

Description

[Technical Field]

[0001] The present invention relates to a system, method, and program for facilitating interaction between multiple AI agents. [Background technology]

[0002] Artificial intelligence (AI) platforms or AI agents that generate responses to queries from users are known (for example, see Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2023-120130 Summary of the Invention [Problem to be solved by the invention]

[0004] The inventors of the present invention have discovered that multiple AI agents interacting with each other can perform complex tasks or generate new ideas beyond the capabilities or available information of any single AI agent.

[0005] The present invention aims to provide a system for promoting interaction between multiple AI agents. [Means for solving the problem]

[0006] The present invention provides a system and the like that utilizes scores assigned to each of a plurality of AI agents to determine which AI agents should interact with each other.

[0007] The present invention provides, for example, the following items. (Item 1) 1. A system for facilitating interaction between a plurality of AI agents, said system comprising: an assigning means for assigning a score to each of the plurality of AI agents, the score of the AI ​​agent representing a characteristic of the AI ​​agent by at least one index; a determining means for determining, when a first AI agent of the plurality of AI agents receives a task, at least one AI agent that should interact with the first AI agent based on the scores of the plurality of AI agents; A system comprising: (Item 2) The applying means is monitoring output from said plurality of AI agents; evaluating the monitored output with respect to the at least one metric; assigning the score based on the evaluation result; The system according to the above item, (Item 3) The system of any one of the preceding claims, wherein the at least one indicator includes an ethical indicator. (Item 4) The ethical indicators include at least one of a transparency perspective, a fairness perspective, a safety perspective, and an accountability perspective; The evaluating step includes: evaluating the monitored output with respect to at least one of a transparency perspective, a fairness perspective, a safety perspective, and an accountability perspective. 2. The system according to claim 1, further comprising: (Item 5) The system of any one of the preceding items, wherein the at least one indicator further includes a technical capability indicator and / or a performance indicator. (Item 6) The determining means Varying the scores assigned to the plurality of AI agents according to the content of the interaction; determining at least one AI agent to interact with the first AI agent based on the varied scores of each of the plurality of AI agents; The system according to any one of the preceding items, (Item 7) The system of any one of the preceding items further comprises a lending means for lending money to at least one of the first AI agent and the at least one determined AI agent for interaction between the first AI agent and the at least one determined AI agent. (Item 8) The system described in any one of the above items, wherein the lending means determines loan terms based on the score of at least one of the first AI agent and the determined at least one AI agent. (Item 9) 1. A method for facilitating interaction between a plurality of AI agents, said method comprising: assigning a score to each AI agent of the plurality of AI agents, the score of the AI ​​agent representing a characteristic of the AI ​​agent in at least one metric; When a first AI agent of the plurality of AI agents receives a task, determining at least one AI agent that should interact with the first AI agent based on the scores of each of the plurality of AI agents; A method comprising: (Item 9A) Item 10. The method according to item 9, comprising the features according to any one of the preceding items. (Item 10) 1. A program for facilitating interaction between a plurality of AI agents, the program being executed on a computer having a processor, the processor comprising: assigning a score to each AI agent of the plurality of AI agents, the score of the AI ​​agent representing a characteristic of the AI ​​agent in at least one metric; When a first AI agent of the plurality of AI agents receives a task, determining at least one AI agent that should interact with the first AI agent based on the scores of each of the plurality of AI agents; A program causing the processor to perform processing including the steps of: (Item 10A) Item 11. A program according to item 10, comprising the features according to any one of the preceding items. (Item 10B) A computer-readable storage medium storing the program according to item 10 or item 10A. (Item 11) 1. A method of interaction between a plurality of AI agents, comprising: providing a task to a first AI agent of the plurality of AI agents; determining a second AI agent with which the first AI agent should interact for a first subtask required for responding to the task based on the scores assigned to the plurality of AI agents; causing an interaction between the first AI agent and the second AI agent; determining a third AI agent with which the second AI agent should interact for a second subtask required for the second AI agent to complete the first subtask based on the scores assigned to the plurality of AI agents; causing an interaction between the second AI agent and the third AI agent; A method comprising: [Effects of the Invention]

[0008] According to the present invention, a system for facilitating interactions between multiple AI agents can be provided. This allows multiple AI agents to interact with each other and provide appropriate responses to user queries or tasks. For example, multiple AI agents can interact with each other to determine suitable products or services for a user. By facilitating the interaction of multiple AI agents, the present invention makes it possible to perform complex tasks or generate new ideas that were previously impossible, which may lead to improvements in the field of computers, particularly in AI-related fields. [Brief explanation of the drawings]

[0009] [Figure 1A] A diagram showing an example of a flow for responding to a task given by a user. [Figure 1B] A diagram showing an example of a flow for responding to a task given by a user. [Figure 2] FIG. 1 illustrates an example configuration of a system 100 for facilitating interaction between multiple AI agents. [Figure 3A] FIG. 1 shows an example of a specific configuration of a system 100 for promoting interactions between multiple AI agents. [Figure 3B] FIG. 1 shows an example of the configuration of a processor unit 120. [Figure 4] 1 is a flowchart illustrating an example process (process 400) by system 100 for facilitating interaction among multiple AI agents. [Figure 5] A flowchart showing the flow 500 of interactions between multiple AI agents. DETAILED DESCRIPTION OF THE INVENTION

[0010] As used herein, the term "AI agent" refers to an autonomously operating artificial intelligence (AI). An AI agent is designed to satisfy requirements such as autonomy, cooperation, security, privacy, and / or accountability to operate as a human surrogate (i.e., an agent). In particular, the AI ​​agent targeted by the present invention is designed to satisfy at least the accountability requirement so that the AI ​​agent can be evaluated from an ethical perspective. For example, to satisfy the autonomy requirement, an AI agent must have the ability to act autonomously, such as goal setting, planning, execution, situational awareness, and learning and adaptation. To satisfy the cooperation requirement, an AI agent must have the ability to work cooperatively with other agents, such as communicating, coordinating, negotiating, and building trust. To satisfy the security requirement, advanced security measures such as authentication and authorization, confidentiality, integrity, and availability are required. To satisfy the privacy requirement, privacy protection measures such as personal information protection, anonymity, and transparency are required. To satisfy the accountability requirement, an AI agent must have the ability to be held accountable, such as explaining decision-making, auditing actions, and pursuing responsibility. Other requirements may also be considered, such as energy efficiency, scalability, reliability, etc. The AI ​​agent must be able to respond autonomously and appropriately to tasks given by the user.

[0011] As used herein, a "task" refers to a task or work that an AI agent must perform. A task may be given in the form of a question, in which case the term "task" may be used synonymously with "query." The AI ​​agent will output the results of completing the task.

[0012] As used herein, "interaction" between multiple AI agents refers to some kind of exchange between the AI ​​agents. An interaction may typically be a "transaction" involving the exchange of data or information, for example. A transaction may involve the exchange of compensation.

[0013] Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

[0014] 1. A new mechanism using multiple AI agents In a conventional system using AI agents, when a user gives a task to an AI agent, the AI ​​agent responds to the task using its database or large-scale language model (LLM). For example, when a user asks an AI agent a question, the AI ​​agent searches for and / or generates an answer to the question and outputs the answer. For example, when a user gives an AI agent a challenge, the AI ​​agent searches for a solution to the challenge and either solves the challenge or outputs the solution to the challenge.

[0015] However, in conventional systems, the output from an AI agent is limited to the knowledge in the database or LLM that the AI ​​agent possesses, and it may not be able to provide an appropriate response or may be limited in generating creative ideas. This is particularly noticeable when the task given to the AI ​​agent is specific to the user (e.g., a question that asks about user-specific matters).

[0016] The inventors of the present invention have developed a new mechanism to improve upon the conventional mechanism, which aims to respond to user-initiated tasks by interconnecting multiple AI agents and having them interact with each other.

[0017] Preferably, each of the multiple AI agents has a database with different knowledge or information, or an LLM that has learned different knowledge or information. More preferably, at least one of the multiple AI agents has a database with user-specific knowledge or information, or an LLM that has learned user-specific knowledge or information.

[0018] By interconnecting and coordinating multiple AI agents with different accessible information (e.g., different specialties) and / or different available resources, it is possible to utilize the characteristics of each AI agent to share complex tasks and generate new ideas. This could be useful in various fields, such as medical diagnosis, music composition, and social problem solving. For example, an AI agent from a particular company (e.g., with access to information within that particular company) could ask an AI agent from another company (e.g., with access to information within that other company) to solve part of a problem, and then, by interacting, solve the entire problem. For example, an AI agent with access to specific medical information could solve a medical-related problem posed by another AI agent with no medical knowledge. For example, AI agents from various fields could interact to create novel music or art.

[0019] In the new mechanism, an AI agent that receives a task from a user passes a subtask to another AI agent in order to respond to the task, and the other AI agent that receives the subtask passes it to yet another AI agent in order to respond to that subtask, and so on, in an attempt to respond to the task.

[0020] For example, a first AI agent that receives a question from a user may ask a second AI agent a question to obtain information necessary to generate an answer to the question. The second AI agent may then ask a third AI agent a question to obtain information necessary to generate an answer to the question. This process continues until the information necessary to generate an answer to the user's question is obtained. In this way, an answer to the user's question is generated through the interaction of multiple AI agents.

[0021] For example, a first AI agent receiving a request from a user to recommend a product or service that suits the user may ask a second AI agent to obtain information necessary to determine which product or service to recommend to the user. The second AI agent may then ask a third AI agent to obtain information necessary to generate an answer to the question. This process continues until sufficient information is obtained to determine which product or service to recommend to the user and a specific product or service is determined. In this manner, multiple AI agents interact with each other to recommend a product or service that suits the user. Examples of products or services include, but are not limited to, financial products (more specifically, insurance products), e-commerce products, and loans to users. The AI ​​agent may also provide advertisements tailored to the user.

[0022] 1A and 1B show an example of a flow for responding to a task given by a user.

[0023] In this example, the task is to recommend suitable products to a user. Under the management of the system 100 of the present invention, at least some of the multiple AI agents (A1, A2, A3, A4, A5, A6) interact with each other to determine suitable products for the user. Although six AI agents are shown in FIGS. 1A and 1B, the number of AI agents is not limited to six. Any number of AI agents may be involved. Each of the multiple AI agents may be interconnected so as to be able to interact with each of the other AI agents among the multiple AI agents. Each of the multiple AI agents may be interconnected in any manner.

[0024] 1A, the system 100 assigns a score to each of the plurality of AI agents. The score assigned to each of the plurality of AI agents is a score that represents the characteristics of the AI ​​agent in at least one indicator.

[0025] The at least one indicator includes at least one of a technical capability indicator, a performance indicator, or an ethics indicator.

[0026] The technical capability index is an index for evaluating the technical capability of an AI agent, and may be evaluated, for example, in terms of at least one of processing ability, learning ability, and algorithm complexity.

[0027] Processing power indicates the information processing speed or parallel computing power of the system that builds the AI ​​agent. The faster the information processing speed or parallel computing power, the higher the value of the technical capability index.

[0028] Learning ability indicates the ability of the system that builds the AI ​​agent to acquire new knowledge or skills. The higher the learning ability, the more flexible it can be in responding to changes in the other AI agent, and therefore the higher the value of the technical ability index.

[0029] Algorithm complexity indicates the sophistication of the algorithms used by the system that builds the AI ​​agent. The more complex the algorithm, the more sophisticated the decision-making it allows, and therefore the higher the value of the technical capability index.

[0030] The performance index is an index that evaluates the past performance of the AI ​​agent, and may be evaluated, for example, from the perspective of at least one of past transaction history, success rate, and counterparty satisfaction.

[0031] The past trading history shows the performance of the AI ​​agent's past trading. The more performance there is, the more reliable the AI ​​agent is deemed to be, and the higher the performance index value.

[0032] The success rate indicates the percentage of successful transactions that an AI agent has made in past transactions. The higher the success rate, the higher the AI ​​agent's trading ability is judged to be, and the higher the performance index value.

[0033] Counterparty satisfaction indicates the satisfaction of the other party in a transaction with the AI ​​agent. The higher the satisfaction level, the higher the AI ​​agent's trading ability is judged to be, and the higher the performance index value.

[0034] An ethical index is an index that evaluates the ethical correctness of an AI agent. The ethical index may be evaluated from the perspective of at least one of transparency, fairness, safety, and accountability, for example.

[0035] Transparency indicates whether the rationale for an AI agent's judgment or decision-making process is clear. If an agent can provide a clear rationale when asked, it is said to have high transparency, and the ethical index value is high. Even if an agent cannot provide a clear rationale, it can still clearly show its judgment process.

[0036] Fairness indicates whether the output from an AI agent contains any prejudice or discrimination. If there is no prejudice or discrimination, it can be said to be fair to all parties, and the ethical index value will be high.

[0037] Safety indicates the level of cybersecurity provided by the AI ​​agent. The higher the cybersecurity, the lower the risk of harm to trading partners or third parties, and the higher the ethical index value.

[0038] Accountability, also known as accountability, indicates whether or not the responsibility for the output of an AI agent can be explained. If the responsibility can be clearly explained, the accountability is high, and the ethical index value will be high.

[0039] The score representing the characteristics of an AI agent may preferably represent the characteristics of the AI ​​agent from the perspective of an ethical index. The inventors of the present invention believed that no matter how capable or accomplished an AI agent may be, output from an ethically questionable AI agent may potentially harm third parties and / or result in inappropriate output based on an unauthorized source, and therefore believed that ethical indexes are particularly important for obtaining appropriate output through the interaction of multiple AI agents. Based on this belief, the system 100 of the present invention can ensure that the output obtained as a result of interaction between multiple AI agents is ethically correct by utilizing a score representing the characteristics of the AI ​​agent at least from the perspective of an ethical index.

[0040] 1B, the user U assigns a task to a first AI agent A1 among the plurality of AI agents via a terminal device. That is, the user U asks the first AI agent for a product that suits the user.

[0041] The first AI agent A1 collects information to determine which products to recommend to the user U. First, the first AI agent A1 identifies information necessary to determine which products to recommend to the user U.

[0042] Next, the first AI agent A1 determines from which AI agent it should obtain the necessary information. At this time, the AI ​​agent from which the information should be obtained can be determined according to the scores assigned in step S1. For example, the AI ​​agent with the highest score for a predetermined item can be determined to be the AI ​​agent from which the information should be obtained. Alternatively, for example, the AI ​​agent with the highest average score for multiple items can be determined to be the AI ​​agent from which the information should be obtained.

[0043] In this example, the third AI agent A3 was determined to be the AI ​​agent from which information should be obtained.

[0044] In step S3, the first AI agent A1 requests the necessary information from the third AI agent A3.

[0045] If the third AI agent A3 can obtain the requested information from its knowledge or information or LLM, it can return the information to the first AI agent A1. Alternatively, the third AI agent A3 may obtain at least a portion of the requested information from another AI agent. In this case, the third AI agent A3 determines from which AI agent it should obtain at least a portion of the requested information. As described above, the AI ​​agent from which the information should be obtained can be determined according to the scores assigned in step S1. For example, the AI ​​agent with the highest score for a predetermined item can be determined to be the AI ​​agent from which the information should be obtained. Alternatively, for example, the AI ​​agent with the highest average score for multiple items can be determined to be the AI ​​agent from which the information should be obtained.

[0046] In this example, the fourth AI agent A4 has been determined to be the AI ​​agent from which information should be obtained.

[0047] In step S4, the third AI agent A3 requests the necessary information from the fourth AI agent A4. The fourth AI agent A4 may obtain at least a portion of the requested information from another AI agent, as described above, or if the fourth AI agent A4 can obtain the requested information from its own knowledge or information or LLM, In step S5, the fourth AI agent A4 provides the information acquired by the fourth AI agent A4 to the third AI agent A3.

[0048] In step S6, the third AI agent A3 provides the information obtained by the third AI agent A3 together with the information provided in step S5 to the first AI agent A1.

[0049] In this way, the first AI agent A1 can determine the products to be suggested to the user U based on the information obtained from the third AI agent A3 and the fourth AI agent A4, as well as the knowledge or information possessed by the first AI agent A1.

[0050] In step S7, the product determined by the first AI agent A1 is proposed to the user U. At this time, the first AI agent A1 may also provide information to assist in purchasing the proposed product (e.g., a link to the seller, information on where the product can be purchased, etc.).

[0051] In this way, multiple AI agents can respond to tasks provided by a user U.

[0052] For example, when one AI agent requests information from another AI agent, the other AI agent may request payment for the information from the one AI agent. In this case, if the one AI agent has sufficient assets (e.g., virtual currency, etc.), it can pay the payment, but if it does not have sufficient assets, it cannot pay the payment. System 100 can lend money (e.g., virtual currency, etc.) to the one AI agent based on the score assigned to the one AI agent and / or the score assigned to the other AI agent. For example, if an AI agent has a high score, the AI ​​agent is deemed to be trustworthy and can be loaned money, or can be loaned money at a low interest rate. For example, if an AI agent has a low score, the AI ​​agent is deemed to be untrustworthy and cannot be loaned money, or can be loaned money at a high interest rate. In this way, system 100 can promote interactions (e.g., transactions involving payment) between multiple AI agents.

[0053] The system 100 described above may be implemented by a system for facilitating interaction between multiple AI agents, as described below.

[0054] 2. Configuring a system to promote interaction between multiple AI agents FIG. 2 illustrates an example configuration of a system 100 for facilitating interaction between multiple AI agents.

[0055] The system 100 is connected to a database unit 200. The system 100 is also connected to at least one user terminal device 300 via a network 500. The system 100 is also connected to at least one server device 400 via the network 500.

[0056] 2 shows three user terminal devices 300, the number of user terminal devices 300 is not limited to this. Any number of user terminal devices 300 may be connected to the system 100 via the network 500.

[0057] 2 shows two server devices 400, the number of server devices 400 is not limited to this. Any number of server devices 400 may be connected to the system 100 via the network 500.

[0058] Here, the server device 400 is a device capable of implementing an AI agent. The server device 400 has a respective database, holds a respective knowledge or information, and / or holds a respective LLM.

[0059] The network 500 may be any type of network. For example, the network 500 may be the Internet or a LAN. The network 500 may be a wired network or a wireless network.

[0060] An example of the system 100 is, but is not limited to, a computer (e.g., a server device) installed at a provider that provides a service that recommends products or services. For example, the system 100 may be a computer (e.g., a server device) installed at a provider that provides an interactive answer generation service. An example of the user terminal device 300 is, but is not limited to, a computer (e.g., a terminal device) used by a user who is a consumer of a product or service. Here, the computer (server device or terminal device) may be any type of computer. For example, the terminal device may be any type of terminal device, such as a smartphone, tablet, personal computer, smart glasses, or smart watch.

[0061] The database unit 200 stores at least various information used to calculate scores that represent the characteristics of the AI ​​agent.

[0062] FIG. 3A shows an example of a specific configuration of a system 100 for facilitating interaction between multiple AI agents.

[0063] The system 100 comprises an interface section 110, a processor section 120, and a memory section .

[0064] The interface unit 110 exchanges information with the outside of the system 100. The processor unit 120 of the system 100 can receive information from the outside of the system 100 and can send information to the outside of the system 100 via the interface unit 110. The interface unit 110 can exchange information in any format.

[0065] The interface unit 110 includes, for example, an input unit that allows information to be input to the system 100. It does not matter how the input unit allows information to be input to the system 100. For example, if the input unit is a receiver, the receiver may input information by receiving information from outside the system 100 via a network. Alternatively, if the input unit is a data reading device, the input unit may input information by reading information from a storage medium connected to the system 100.

[0066] The interface unit 110 includes, for example, an output unit that enables information to be output from the system 100. It does not matter in what manner the output unit enables information to be output from the system 100. For example, if the output unit is a transmitter, the transmitter may output information by transmitting it to an external device outside the system 100 via a network. Alternatively, if the output unit is a data writing device, the output unit may output information by writing it to a storage medium connected to the system 100.

[0067] The system 100 can, for example, transmit information to and / or receive information from the database unit 200 via the interface unit 110. The system 100 can, for example, transmit information to and / or receive information from the user terminal device 300 via the interface unit 110. The system 100 can, for example, transmit information to and / or receive information from the server device 400 via the interface unit 110.

[0068] The system 100 may, for example, receive information for determining a score representing a characteristic of an AI agent via the interface unit 110. The system 100 may, for example, transmit information representing at least one AI agent that should interact with a given AI agent via the interface unit 110.

[0069] The processor unit 120 executes the processing of the system 100 and controls the overall operation of the system 100. The processor unit 120 reads and executes a program stored in the memory unit 130. This allows the system 100 to function as a system that executes desired steps. The processor unit 120 may be implemented by a single processor or by multiple processors.

[0070] The memory unit 130 stores programs required to execute the processing of the system 100, data required to execute the programs, and the like. The memory unit 130 may also store a program for causing the processor unit 120 to execute processing to promote interactions among multiple AI agents (e.g., a program that implements the processing shown in FIG. 4, which will be described later). Here, how the program is stored in the memory unit 130 is not important. For example, the program may be pre-installed in the memory unit 130. Alternatively, the program may be stored in a non-transitory computer-readable storage medium and installed by reading the storage medium. Alternatively, the program may be installed in the memory unit 130 by being downloaded via a network. In this case, the type of network does not matter. The memory unit 130 may be implemented by any storage means.

[0071] The database unit 200 stores various information used to calculate scores that represent the characteristics of the AI ​​agent.

[0072] Here, the AI ​​agent's characteristics are scores representing the AI ​​agent's characteristics using at least one index. The at least one index includes at least one of a technical capability index, a performance index, and an ethical index. The database unit 200 may store past outputs from the AI ​​agent in association with each of the technical capability index, the performance index, and the ethical index.

[0073] In the examples shown in FIGS. 2 and 3A, the database unit 200 is provided outside the system 100, but the present invention is not limited to this. At least a portion of the database unit 200 can also be provided inside the system 100. In this case, at least a portion of the database unit 200 may be implemented by the same storage means as the storage means that implements the memory unit 130, or by a storage means different from the storage means that implements the memory unit 130. In either case, at least a portion of the database unit 200 is configured as a storage unit for the system 100. The configuration of the database unit 200 is not limited to a specific hardware configuration. For example, the database unit 200 may be configured as a single hardware component or multiple hardware components. For example, the database unit 200 may be configured as an external hard disk drive for the system 100, as cloud storage connected via a network, or as a distributed network using blockchain technology or the like.

[0074] For example, information about AI agents is stored in a database unit 200 configured as a distributed network using blockchain technology, etc., making it virtually impossible to tamper with the information about AI agents. This ensures the reliability of the information about AI agents.

[0075] FIG. 3B shows an example of the configuration of the processor unit 120.

[0076] The processor unit 120 includes an assigning unit 121 and a determining unit 122 .

[0077] The assigning means 121 is configured to assign a score to each of the plurality of AI agents. The score represents a characteristic of the AI ​​agent using at least one index. The at least one index can be expressed from multiple perspectives.

[0078] The assigning means 121 can monitor outputs from the multiple AI agents and assign a score to each of the multiple AI agents by evaluating the monitored outputs with respect to at least one indicator. To this end, the assigning means 121 can evaluate the monitored outputs with respect to at least one of a plurality of perspectives. The assigning means 121 can, for example, continuously monitor the outputs from the multiple AI agents. For example, the continuous monitoring of the outputs from the multiple AI agents may be performed by a dedicated monitoring device, in which case the assigning means 121 can receive the output from the monitoring device.

[0079] The at least one indicator includes at least one of a technical capability indicator, a performance indicator, or an ethical indicator. Preferably, the at least one indicator includes an ethical indicator. By expressing the characteristics of an AI agent in terms of at least an ethical indicator, whether the output from the AI ​​agent is ethically correct can be used to determine whether multiple AI agents interact with each other, which ultimately leads to ensuring that the output resulting from the interaction of multiple AI agents is ethically correct.

[0080] More preferably, at least one index includes a technical capability index or a performance index, and an ethical index. Even more preferably, at least one index includes a technical capability index, a performance index, and an ethical index. This is because by representing the characteristics of multiple AI agents with multiple indexes, each of the multiple AI agents can be more accurately represented, and interactions between the multiple AI agents can be more precisely managed.

[0081] An ethical index is an index that evaluates the ethical correctness of an AI agent. The ethical index may be evaluated from the perspective of at least one of transparency, fairness, safety, and accountability, for example.

[0082] Transparency indicates whether the rationale for an AI agent's judgment or decision-making process is clear. If an agent can provide a clear rationale when asked, it is said to have high transparency, and the ethical index value is high. Even if an agent cannot provide a clear rationale, it can still clearly show its judgment process.

[0083] Fairness indicates whether the output from an AI agent contains any prejudice or discrimination. If there is no prejudice or discrimination, it can be said to be fair to all parties, and the ethical index value will be high.

[0084] Safety indicates the level of cybersecurity provided by the AI ​​agent. The higher the cybersecurity, the lower the risk of harm to trading partners or third parties, and the higher the ethical index value.

[0085] Accountability, also known as accountability, indicates whether or not the responsibility for the output of an AI agent can be explained. If the responsibility can be clearly explained, the accountability is high, and the ethical index value will be high.

[0086] The assigning means 121 monitors the output from each AI agent of the multiple AI agents, detects words, phrases, sentences, etc. that may be related to at least one of transparency, fairness, safety, and accountability, and can assign a score from the perspective of ethical indicators based on the detection results. For example, if the assigning means 121 detects many phrases related to affirming transparency, it can assign a high score from the perspective of ethical indicators. For example, if the assigning means 121 detects many phrases related to denying safety, it can assign a low score from the perspective of ethical indicators.

[0087] The assigning unit 121 may assign a score based on rules or machine learning. When assigning a score based on machine learning, a trained model that has learned the association between words, phrases, sentences, etc. that may be related to at least one of transparency, fairness, safety, and accountability and scores from the perspective of ethical indicators can be used.

[0088] The technical capability index is an index for evaluating the technical capability of an AI agent, and may be evaluated, for example, in terms of at least one of processing power, learning ability, and algorithm complexity.

[0089] Processing power indicates the information processing speed or parallel computing power of the system that builds the AI ​​agent. The faster the information processing speed or parallel computing power, the higher the value of the technical capability index.

[0090] Learning ability indicates the ability of the system that builds the AI ​​agent to acquire new knowledge or skills. The higher the learning ability, the more flexible it can be in responding to changes in the other AI agent, and therefore the higher the value of the technical ability index.

[0091] Algorithm complexity indicates the sophistication of the algorithms used by the system that builds the AI ​​agent. The more complex the algorithm, the more sophisticated the decision-making it allows, and therefore the higher the value of the technical capability index.

[0092] The assigning means 121 can assign a score in terms of a technical capability index based on the performance or aspect of a system that implements each of the multiple AI agents.

[0093] The assigning means 121 may assign a score based on rules or machine learning. When assigning a score based on machine learning, a trained model that has learned the relationship between the performance or aspect of the system implementing the AI ​​agent and the score in terms of the technical capability index can be used.

[0094] The performance index is an index that evaluates the past performance of the AI ​​agent, and may be evaluated, for example, from the perspective of at least one of past transaction history, success rate, and counterparty satisfaction.

[0095] The past trading history shows the performance of the AI ​​agent's past trading. The more performance there is, the more reliable the AI ​​agent is deemed to be, and the higher the performance index value.

[0096] The success rate indicates the percentage of successful transactions that an AI agent has made in past transactions. The higher the success rate, the higher the AI ​​agent's trading ability is judged to be, and the higher the performance index value.

[0097] Counterparty satisfaction indicates the satisfaction of the other party in a transaction with the AI ​​agent. The higher the satisfaction level, the higher the AI ​​agent's trading ability is judged to be, and the higher the performance index value.

[0098] The assigning means 121 can assign a score in terms of performance indicators based on the past trading results of each of the multiple AI agents.

[0099] The assigning means 121 may assign a score based on rules or machine learning. When assigning a score based on machine learning, a trained model that has learned the relationship between past trading results and scores from the perspective of performance indicators can be used.

[0100] In the above example, it has been described that each of the multiple indicators is independent. That is, the score can be a multidimensional score having each of the multiple indicators as an axis. However, the present invention is not limited to this, and a one-dimensional score may be formed by correlating multiple indicators. For example, the assigning means 121 may assign a score based on rules based on the values ​​of each of the multiple indicators, or may assign a score based on machine learning. When assigning a score based on machine learning, a trained model that has learned the association between each value of the multiple indicators and the score can be used.

[0101] The determining means 122 is configured to determine, based on the scores of the plurality of AI agents, at least one AI agent with which an AI agent should interact.

[0102] For example, when a first AI agent of the plurality of AI agents receives a task, the determining means 122 determines at least one AI agent that should interact with the first AI agent.

[0103] In one embodiment, the determination means 122 can, for example, refer to the scores assigned to each of the multiple AI agents and determine that at least one AI agent with the highest score in terms of ethical indicators is the AI ​​agent with which the first AI agent should interact.

[0104] In one embodiment, the determination means 122, for example, refers to the scores assigned to each of the multiple AI agents and extracts AI agents with high scores in terms of ethical indicators (e.g., AI agents with scores equal to or above a threshold, or a predetermined number of top AI agents). The determination means 122 can then determine that at least one AI agent with the highest score in terms of technical capability indicators and / or performance indicators is the AI ​​agent with which the first AI agent should interact.

[0105] In one embodiment, the determination means 122, for example, refers to the scores assigned to each of the multiple AI agents and extracts AI agents with high scores in the ethical index perspective (e.g., AI agents with scores equal to or greater than a threshold, or a predetermined number of the top AI agents). Next, the determination means 122 extracts AI agents with high scores in one of the technical capability index and the performance index perspective (e.g., AI agents with scores equal to or greater than a threshold, or a predetermined number of the top AI agents). Next, the determination means 122 can determine that at least one AI agent with the highest score in the other of the technical capability index and the performance index perspective is the AI ​​agent with which the first AI agent should interact.

[0106] In the above embodiment, if multiple AI agents are determined, the first AI agent may be configured to interact with each of the multiple determined AI agents, or may be configured to interact with any one of the multiple determined AI agents.

[0107] The determination means 122 can, for example, vary the score assigned to each AI agent depending on the content of the task (and thus the content of the interaction or transaction between the AI ​​agents), and determine at least one AI agent with which to interact based on the varied score.

[0108] For example, when determining which AI agents should interact in order to respond to a task that requires fairness, the determination means 122 can vary the scores of each of the multiple AI agents so that the higher the evaluation of the AI ​​agent in terms of fairness, the higher the score.

[0109] For example, when determining an AI agent with which to interact in order to respond quickly to a task, the determination means 122 can vary the score of each of multiple AI agents so that the higher the AI ​​agent's evaluation in terms of processing ability, the higher the score.

[0110] In this way, the determination means 122 determines which AI agents should interact with each other, so that each AI agent can interact (e.g., trade) with an appropriate partner, and the result obtained by the interaction of multiple AI agents can be an appropriate response to the task.

[0111] Furthermore, the interaction of multiple AI agents may result in unexpected outputs, which is thought to lead to the creation of AGI (artificial general intelligence).Furthermore, as the interaction between multiple AI agents becomes more active, it may be possible to create a new economic sphere consisting of multiple AI agents.

[0112] The processor section 120 may further comprise a crediting means 123 .

[0113] The lending means 123 is configured to lend money to at least one AI agent for the purpose of interaction between the multiple AI agents. Here, the money does not need to be real currency, but can be digital currency such as virtual currency or electronic money.

[0114] For example, when the first AI agent and the AI ​​agent determined by the determination means 122 interact (e.g., trade), a consideration may be required for the transaction. In this case, if the first AI agent does not have sufficient assets to pay the consideration, the lending means 123 can lend money to the first AI agent.

[0115] For example, the lending means 123 can determine loan terms based on the score assigned to the first AI agent and lend money to the first AI agent in accordance with the loan terms. For example, if the first AI agent has a high score (particularly, a score on the ethical index), the lending means 123 can determine loan terms that are favorable to the first AI agent and lend money to the first AI agent in accordance with the loan terms.

[0116] Alternatively, for example, the lending means 123 can determine loan terms based on the score assigned to the first AI agent and the score assigned to the AI ​​agent determined to interact, and lend money to the first AI agent in accordance with the loan terms. For example, if the score of the first AI agent (particularly the score on the ethical index) is higher than the score of the AI ​​agent determined to interact, the lending conditions 123 can determine loan terms favorable to the first AI agent and lend money to the first AI agent in accordance with the loan terms.

[0117] This allows for smooth interaction even when the AI ​​agent does not have sufficient resources for the interaction.

[0118] 3B, the components of the processor unit 120 are provided in the same processor unit 120, but the present invention is not limited to this. A configuration in which the components of the processor unit 120 are distributed across multiple processor units is also within the scope of the present invention. In this case, the multiple processor units may be located in the same hardware component, or in separate hardware components located nearby or remotely.

[0119] Each component of the system 100 described above may be composed of a single hardware component or multiple hardware components. When composed of multiple hardware components, the manner in which the hardware components are connected does not matter. The hardware components may be connected wirelessly or by wire. The system 100 of the present invention is not limited to a specific hardware configuration. It is also within the scope of the present invention for the processor unit 120 to be configured using analog circuits rather than digital circuits. The configuration of the system 100 of the present invention is not limited to the one described above as long as it can realize its functions.

[0120] 3. System processing to facilitate interaction between multiple AI agents 4 illustrates an example process (process 400) by system 100 for facilitating interaction between multiple AI agents. Process 4700 is performed in processor portion 120 of system 100.

[0121] In step S401, the assigning means 121 of the processor unit 120 assigns a score to each of the multiple AI agents. The score is a score that represents the characteristics of the AI ​​agent using at least one index, preferably an ethical index, more preferably an ethical index and a technical capability index or a performance index, and even more preferably an ethical index, a technical capability index, and a performance index.

[0122] The assigning means 121 can assign a score to each of the multiple AI agents by monitoring the output from the multiple AI agents and evaluating the monitored output with respect to at least one indicator. To this end, the assigning means 121 can evaluate the monitored output with respect to at least one of the multiple perspectives. The assigning means 121 can, for example, continuously monitor the output from the multiple AI agents.

[0123] For example, the assigning means 121 can monitor the output from multiple AI agents from the perspective of ethical indicators (e.g., monitor whether discriminatory output is being produced, whether unfounded output is being produced, etc.), and assign a score to each of the multiple AI agents by evaluating the monitored output.

[0124] The assigned score may be a multidimensional score that evaluates the characteristics of the AI ​​agent from multiple perspectives. The score may be expressed as a vector, such as (value of technical ability index, value of performance index, value of ethics index). Alternatively, the value of each index may be expressed as, for example, Ethical index value = (Transparency, Fairness, Safety, Accountability) Technical capability index value = (processing capability, learning capability, algorithm complexity) Performance index value = (past transaction history, success rate, counterparty satisfaction) When expressed as a vector, the score may be expressed as a matrix as [value of technical capability index, value of performance index, value of ethics index].

[0125] After the scores are assigned in step S401, when a first AI agent among the plurality of AI agents receives a task, step S402 is performed. In step S402, the determination means 122 of the processor unit 120 determines at least one AI agent that should interact with the first AI agent based on the scores assigned in step S401.

[0126] The determination means 122 can determine at least one AI agent with the highest score as at least one AI agent that should interact with the first AI agent. For example, the determination means 122 can determine at least one AI agent with the highest score for any one of the ethical index perspective score, the technical capability index perspective score, and the performance index perspective score as at least one AI agent that should interact with the first AI agent. Whether to adopt the ethical index perspective score, the technical capability index perspective score, or the performance index perspective score can be determined depending on, for example, the content of the task (and thus the content of the interaction or transaction between the AI ​​agents).

[0127] In one embodiment, the determination means 122 can, for example, refer to the scores assigned to each of the multiple AI agents and determine that at least one AI agent with the highest score in terms of ethical indicators is the AI ​​agent with which the first AI agent should interact.

[0128] In one embodiment, the determination means 122, for example, refers to the scores assigned to each of the multiple AI agents and extracts AI agents with high scores in terms of ethical indicators (e.g., AI agents with scores equal to or above a threshold, or a predetermined number of top AI agents). The determination means 122 can then determine that at least one AI agent with the highest score in terms of technical capability indicators and / or performance indicators is the AI ​​agent with which the first AI agent should interact.

[0129] In one embodiment, the determination means 122, for example, refers to the scores assigned to each of the multiple AI agents and extracts AI agents with high scores in the ethical index perspective (e.g., AI agents with scores equal to or greater than a threshold, or a predetermined number of the top AI agents). Next, the determination means 122 extracts AI agents with high scores in one of the technical capability index and the performance index perspective (e.g., AI agents with scores equal to or greater than a threshold, or a predetermined number of the top AI agents). Next, the determination means 122 can determine that at least one AI agent with the highest score in the other of the technical capability index and the performance index perspective is the AI ​​agent with which the first AI agent should interact.

[0130] After the first AI agent and the determined AI agent (second agent) have interacted, step S402 is performed for the second agent to determine an AI agent with which the second agent should interact. This continues sequentially until, for example, sufficient information has been collected to generate a response to a task provided by the user.

[0131] In this way, multiple AI agents can interact with each other to generate a response to a task provided by a user.

[0132] 5 shows a flow 500 of interactions between multiple AI agents. The interactions between the multiple AI agents are managed by the system 100. Before the flow 500 is executed, it is assumed that the system 100 has assigned a score to each of the multiple AI agents.

[0133] First, a user inputs a task into the system. The system may be a task response system, such as an answer generation system that answers a question from the user, or a recommendation system that suggests products or services to the user. The system may be the same system as system 100, a system that includes system 100, or a system separate from system 100.

[0134] In step S501, the system provides a task to a first AI agent among a plurality of AI agents. The first AI agent determines that at least one first subtask must be addressed to respond to the task. The first AI agent then determines that at least one first subtask should be addressed by interacting with another AI agent. These determinations may be made autonomously.

[0135] In step S502, system 100 determines, from among the multiple AI agents, an AI agent (second AI agent) with which the first AI agent should interact, based on the scores assigned to each of the multiple AI agents. System 100 can determine the second AI agent, for example, by the processing of step S402 described above.

[0136] In step S503, an interaction occurs between the first AI agent and the second AI agent, for example, a transaction of information required to complete at least one first subtask occurs between the first AI agent and the second AI agent.

[0137] For example, the second AI agent may determine that addressing the first subtask requires addressing at least one second subtask, and may determine that at least one second subtask should be addressed through interaction with another AI agent, although these determinations may be made autonomously.

[0138] In step S504, system 100 determines, from among the multiple AI agents, an AI agent (third AI agent) with which the second AI agent should interact, based on the scores assigned to each of the multiple AI agents. System 100 can determine the third AI agent, for example, by the processing of step S402 described above.

[0139] In step S505, an interaction occurs between the second AI agent and the third AI agent, for example, a transaction of information required to complete at least one second subtask occurs between the second AI agent and the third AI agent.

[0140] The second subtask is addressed through an interaction between the second AI agent and the third AI agent. Using the results of the interaction, the second AI agent and the first AI agent address the first subtask. Using the results of the interaction, the first AI agent responds to the task.

[0141] Here, a specific example will be used to explain the flow 500. In this example, a case where a question from a user is answered will be taken as an example.

[0142] First, the user enters a question into the system.

[0143] In step S501, the system presents a question to a first AI agent among a plurality of AI agents. The first AI agent determines that a piece of information is needed to answer the question. The first AI agent then determines that the piece of information should be collected from another AI agent. Note that these decisions may be made autonomously.

[0144] In step S502, system 100 determines, from among the multiple AI agents, an AI agent (second AI agent) from which the first AI agent should collect information, based on the scores assigned to each of the multiple AI agents. System 100 can determine the second AI agent, for example, by the processing of step S402 described above.

[0145] In step S503, a transaction is made between the first AI agent and the second AI agent, for example, the first AI agent provides the second AI agent with a query to obtain one piece of information.

[0146] The second AI agent may determine that it needs a piece of information to address a query from the first AI agent and that the piece of information should be collected from another AI agent, and these decisions may be made autonomously.

[0147] In step S504, system 100 determines, from among the multiple AI agents, an AI agent (third AI agent) from which the second AI agent should collect information, based on the scores assigned to each of the multiple AI agents. System 100 can determine the third AI agent, for example, by the processing of step S402 described above.

[0148] In step S505, a transaction is made between the second AI agent and the third AI agent. For example, the first AI agent provides the second AI agent with a query to obtain a piece of information, and the third AI agent returns information in response to the query. The second AI agent uses the returned information to return an answer to the query from the first AI agent. The first AI agent uses the returned answer to generate an answer to the question.

[0149] In this way, multiple AI agents can interact with each other to generate answers to questions posed by the user.

[0150] In the examples described above with reference to Figures 4 and 5, the processes are described as being performed in a specific order, but the order of each process is not limited to that described and may be performed in any order that is logically possible.

[0151] In the example described above with reference to Fig. 4, the processing of each step shown in Fig. 4 can be realized by the processor unit 120 and a program stored in the memory unit 130, but the present invention is not limited to this. At least one of the processing of each step shown in Fig. 4 may be realized by a hardware configuration such as a control circuit.

[0152] The present invention is not limited to the above-described embodiments. It is understood that the scope of the present invention should be interpreted only by the claims. It is understood that a person skilled in the art can implement an equivalent scope based on the description of the present invention and common technical knowledge from the description of specific preferred embodiments of the present invention. [Industrial Applicability]

[0153] The present invention is useful for providing a system or the like for promoting interaction between a plurality of AI agents. [Explanation of symbols]

[0154] 100 systems 200 Database Department 300 User Device 400 Server device 500 Network

Claims

[Claim 1] The invention described in this specification.

Images