Scoped task delegation system between AI agent nodes

Abstract

In an environment where multiple AI agents are used to support multiple users, the goal is to enable the automated execution of tasks that depend on information or permissions on external services of other users, without sharing the authentication information of other users. [Solution] The system comprises multiple AI agent nodes corresponding to each user, and a registry that manages agent description information, including capabilities, delegation-permitted operational scope, and connection destinations, in a searchable manner. The requesting node obtains the description information of the delegated node from the registry and sends a task request. The delegated node accesses an external service using the user's authentication information and executes the task only if it is within the operational scope, returns the result, and the requesting node aggregates and presents it.

Description

Technical Field

[0001] The present invention relates to an information processing system including artificial intelligence agents, and particularly to an information processing technology for delegating task requests between a plurality of artificial intelligence agent nodes corresponding to a plurality of users respectively, accessing an external service using the authentication information of the delegated user, and executing processing.

Background Art

[0002] In recent years, artificial intelligence agents that execute business processing including operations of external services in response to natural language input have been used. This type of artificial intelligence agent can access external services such as calendars, messaging, and ticket management using the user's authentication information and execute desired processing.

[0003] However, in an environment where a plurality of users in an organization use artificial intelligence agents respectively, there may be a situation where an artificial intelligence agent of a certain user executes processing that depends on information or operation authority on the external service of another user. For example, processing for querying the schedules of multiple people and extracting candidate dates and times, processing for referring to templates held by others, or processing that requires approval from others.

[0004] If such processing is attempted to be realized by a single artificial intelligence agent, it is necessary to share the authentication information of other users, which is not desirable from the viewpoints of authority management, auditability, and security. Also, in a configuration where the management of access tokens on the external service side is centralized to a single entity, it may violate the principle of least privilege.

[0005] On the other hand, technologies for sharing tasks among a plurality of automatic assistants, technologies for handling authority delegation, and technologies for managing access permissions have been proposed. However, a series of frameworks for mutually discovering a plurality of artificial intelligence agent nodes while maintaining the boundary of authentication information for each user, clearly defining an acceptable delegation range, executing only requests within the acceptable range on the receiving side, and aggregating and presenting the results to the requester have not been sufficiently established.

Prior Art Documents

[0006] [Patent Document 1] U.S. Patent No. 11,579,923 [Patent Document 2] U.S. Patent No. 8,776,204 [Patent Document 3] U.S. Patent No. 10,530,761 [Overview of the project] [Problems that the invention aims to solve]

[0007] In environments where multiple users each utilize artificial intelligence agents, a configuration in which a single artificial intelligence agent directly holds or shares the authentication information of other users when performing tasks involving the acquisition and manipulation of necessary information between users can increase the risk of excessive privilege granting and data breaches.

[0008] Furthermore, manually adjusting tasks that depend on other users' information or permissions each time can lead to processing delays and increased operational burden. Moreover, even when multiple artificial intelligence agents work together, if there is a lack of mechanisms to discover which agent can perform what, control the scope of delegation, and aggregate the execution results, automating collaborative processing can become difficult.

[0009] In addition, if the operations performed by the delegated entity change the state of an external service, it may be necessary to confirm the intentions of the user associated with that delegated entity. However, a configuration that does not incorporate this confirmation may leave a risk of error.

[0010] In view of the above circumstances, the present invention aims to provide an information processing technology that enables multiple artificial intelligence agent nodes to delegate tasks to each other while maintaining the boundaries of authentication information for each user, allows the receiving side to determine whether to execute a task based on the scope of delegation, obtains approval as necessary, and enables the requesting party to aggregate and present the execution results. [Means for solving the problem]

[0011] The information processing system according to the present invention comprises: a plurality of artificial intelligence agent nodes, each corresponding to a plurality of users and capable of accessing external services using the authentication information of each user; a storage unit that holds agent description information for each artificial intelligence agent node, including metadata indicating the processing capacity that the node can provide, the scope of operations that are permitted to be delegated to the node, and the connection destination to the node; a registry that manages the agent description information in a searchable manner; a delegation request unit that, when it is determined that delegation to another artificial intelligence agent node is necessary based on a task request from a user, obtains the agent description information of a target node from the registry and sends the task request to the target node; an execution unit that, at the target node, executes the received task request using the authentication information of the user associated with the target node only if the received task request falls within the scope of operations, and returns the execution result; and an aggregation unit that aggregates the returned execution results and presents them to the user.

[0012] Furthermore, the present invention can also be understood as an information processing method and program realized by the above system.

[0013] The present invention can be implemented by appropriately combining the following: a configuration that includes availability information in agent description information; a configuration that sends and receives task requests over an encrypted communication channel with mutual authentication; a configuration that executes task requests, including operations that involve writing or changing the state of an external service, after obtaining user approval; a configuration that performs parallel delegation to multiple nodes and generates integrated results; a configuration that provides streaming returns for requests in the form of procedural calls; a configuration that allows users to set the scope of operations; a configuration that provides mutual authentication and routing using containers and service meshes; a configuration in which organizational bots delegate to individual user nodes; and a configuration that transmits data via asynchronous queues. [Effects of the Invention]

[0014] According to the present invention, the requesting artificial intelligence agent node can delegate a task to another artificial intelligence agent node, and access an external service and execute processing using authentication information associated with the recipient user. This makes it possible to automate cross-user tasks while maintaining the boundaries of authentication information for each user.

[0015] Furthermore, according to the present invention, since the delegated artificial intelligence agent node executes the received task request only when it falls within its operational scope, operations that deviate from the permitted delegation scope can be suppressed, thereby reducing the risk of excessive authorization and erroneous operations.

[0016] Furthermore, according to the present invention, agent description information can be managed in a searchable manner by a registry, and a delegate can be selected based on processing capacity and connection destination, thereby improving the discoverability and operability of cooperative processing.

[0017] Furthermore, according to the present invention, by adopting a configuration in which task requests, including operations that involve writing or changing the status of external services, are executed only after obtaining approval from the user associated with the delegate, it is possible to incorporate confirmation of intent for high-risk operations and improve security and governance.

[0018] Furthermore, according to the present invention, by employing parallel delegation to multiple nodes, integrated result generation, and return in streaming format, responsiveness can be improved even when simultaneously acquiring multiple data sources or processing for long periods of time.

[0019] Furthermore, according to the present invention, by adopting a configuration that provides mutual authentication and routing using containers and a service mesh, as well as a configuration for transmission via asynchronous queues, secure communication and scalable delegation processing in a distributed environment can be realized. [Brief explanation of the drawing]

[0020] [Figure 1] This is a diagram illustrating the configuration of the present invention. [Figure 2] This is a functional block diagram of the present invention. [Figure 3]This is a flowchart of the present invention. [Figure 4] This is a sequence diagram of the present invention.

Embodiments for Carrying Out the Invention

[0021] Hereinafter, embodiments of the present invention will be described with reference to the drawings. The following embodiments are examples for embodying the present invention, and the technical scope of the present invention is determined by the invention described in the claims.

[0022] As shown in FIG. 1, the information processing system 10 may include a terminal 11 of a requesting user, a terminal 12 of an approver, a requesting artificial intelligence agent node 20, an agent registry 30, an entrusted artificial intelligence agent node 40, and an external service 50. The requesting artificial intelligence agent node 20 and the entrusted artificial intelligence agent node 40 correspond to different users respectively, and can access the external service 50 using the authentication information of the corresponding user.

[0023] Based on a task request from the requesting user, the requesting artificial intelligence agent node 20 determines whether it is necessary to delegate to other artificial intelligence agent nodes. If necessary, it acquires agent description information of candidate entrustees from the agent registry 30 and sends a task request to the entrusted artificial intelligence agent node 40.

[0024] The entrusted artificial intelligence agent node 40 executes the task request only when the received task request is within the operation range set for the node, and returns the execution result to the requesting artificial intelligence agent node 20. The requesting artificial intelligence agent node 20 aggregates the returned execution results and presents them to the terminal 11 of the requesting user.

[0025] The information processing system 10 is characterized in that when the delegated artificial intelligence agent node 40 operates the external service 50, it uses the authentication information of the user associated with the delegated service. As a result, the requesting artificial intelligence agent node 20 can perform processing that depends on the information or permissions on the external service of other users without sharing or holding the authentication information of other users.

[0026] Communication between the requesting artificial intelligence agent node 20 and the delegated artificial intelligence agent node 40 may take place over an encrypted communication channel with mutual authentication. Furthermore, mutual authentication and routing may be provided between each node running as a container via a service mesh 60.

[0027] As shown in Figure 2, the requesting artificial intelligence agent node 20 may include a task analysis unit 21, a delegation decision unit 22, a registry search unit 23, a delegation request unit 24, an aggregation unit 25, and a response generation unit 26. The task analysis unit 21 analyzes the input from the requesting user and identifies the processing content to be executed and the necessary external services. The delegation decision unit 22 determines whether the identified processing content can be completed using only the authentication information of its own node or whether it requires the authentication information of another user.

[0028] The registry search unit 23 searches the agent registry 30 to obtain agent description information of candidate delegates when it determines that delegation is necessary. The delegation request unit 24 sends a task request to the delegated artificial intelligence agent node 40 based on the connection destination included in the obtained agent description information.

[0029] The aggregation unit 25 receives the execution results returned from the delegated artificial intelligence agent node 40 and, if the execution is delegated in parallel to multiple nodes, generates an integrated result from the multiple execution results. The response generation unit 26 formats the integrated result into a format that can be presented to the requesting user's terminal 11 and outputs it.

[0030] The agent registry 30 manages agent description information for multiple artificial intelligence agent nodes in a searchable manner. The agent registry 30 may include an agent description information holding unit 31 and a search unit 32. The agent description information includes metadata indicating the processing capabilities that the node can provide, the scope of operations that are permitted to be delegated to the node, and the connection destination to the node.

[0031] The agent description information may further include availability information indicating the operational status or processing load of the node. In this case, the delegation decision unit 22 or the delegation request unit 24 may select a delegation destination based on the availability information.

[0032] The delegated artificial intelligence agent node 40 may include a receiving unit 41, an operation range determination unit 42, an approval acquisition unit 43, an execution unit 44, and a return unit 45. The receiving unit 41 receives a task request from the requesting artificial intelligence agent node 20. The operation range determination unit 42 determines whether the received task request falls within the operation range that allows delegation to the node.

[0033] If the task falls within the scope of operation, the execution unit 44 accesses the external service 50 using the user authentication information associated with the delegated artificial intelligence agent node 40 and executes the task request. The return unit 45 returns the execution result to the requesting artificial intelligence agent node 20. The execution result may be returned in streaming format for requests in the form of procedure calls.

[0034] If the received task request does not fall within the scope of operation, the delegated artificial intelligence agent node 40 may refuse to execute the task request and return a response including the reason for the refusal.

[0035] The approval acquisition unit 43 may adopt a configuration in which it only allows the execution unit 44 to execute a task request that includes an operation to write or change the status of an external service, after obtaining approval from the user associated with the delegate. Approval can be obtained by notifying the approver's terminal 12 and obtaining a response.

[0036] The user can pre-configure the range of requests that the delegated node will accept via a configuration user interface 70 that allows setting the scope of operations. The scope of operations may be defined, for example, as a read-only scope, a writable scope requiring approval, or a scope limited to specific external services.

[0037] Next, an example of the processing procedure of the present invention will be described with reference to Figure 3. The requesting artificial intelligence agent node 20 receives a task request from the requesting user (step S201). Subsequently, the task analysis unit 21 analyzes the task request and identifies the necessary processing content (step S202).

[0038] Next, the delegation determination unit 22 determines whether it is necessary to delegate the identified processing content to another artificial intelligence agent node (step S203). If delegation is necessary, the registry search unit 23 searches the agent registry 30 and obtains agent description information of candidate delegation destinations (step S204).

[0039] Next, the delegation request unit 24 selects a delegation destination artificial intelligence agent node 40 based on the acquired agent description information and sends a task request (step S205). In a configuration that performs parallel delegation, task requests can be sent in parallel to multiple delegation destinations.

[0040] The delegated artificial intelligence agent node 40 receives the task request (step S206). Subsequently, the operation range determination unit 42 determines whether the received task request falls within the operation range (step S207).

[0041] If a task request falls within the scope of operations, it can be determined whether the task request includes an operation to write or change the status of an external service (step S208). If it includes an operation to write or change, the approval unit 43 obtains approval from the user associated with the delegate (step S209). If approval is obtained, the execution unit 44 accesses the external service 50 using the authentication information of the user associated with the delegate and executes the task request (step S210).

[0042] The execution result is returned to the requesting artificial intelligence agent node 20 by the return unit 45 (step S211). In a configuration where the execution result is returned in streaming format, the requesting artificial intelligence agent node 20 can sequentially aggregate the partial results as they are received.

[0043] The requesting artificial intelligence agent node 20 aggregates the execution results using the aggregation unit 25, generates a response for user presentation using the response generation unit 26, and outputs it to the requesting user's terminal 11 (step S212).

[0044] Next, a time-series example of the present invention will be described with reference to Figure 4. The requesting user sends a task request from terminal 11, and the requesting artificial intelligence agent node 20 analyzes the task request and determines the necessity of delegation. If delegation is necessary, the requesting artificial intelligence agent node 20 obtains agent description information from the agent registry 30, including the destination of the delegation and the scope of operations.

[0045] The requesting artificial intelligence agent node 20 sends a task request to the delegated artificial intelligence agent node 40. The delegated artificial intelligence agent node 40 determines whether the task can be executed based on the scope of operation, and if it can be executed, it uses the authentication information of the user associated with the delegate to operate the external service 50 and generate the execution result.

[0046] If the operation involves high-risk operations such as writing data, the delegated artificial intelligence agent node 40 obtains approval from the user associated with the delegate via the approver terminal 12, and after approval, accesses the external service 50 to execute the process.

[0047] The requesting artificial intelligence agent node 20 aggregates the returned execution results and presents them to the requesting user. If task requests are sent in parallel to multiple delegates, the execution results from each delegate are integrated and presented.

[0048] In embodiments of the present invention, artificial intelligence agent nodes can be run as containers. In a container orchestration environment, the connection destination for each node can be represented as a service name or endpoint, and mutual authentication and routing can be provided by a service mesh 60.

[0049] Task requests may be sent and received over an encrypted communication channel with mutual authentication. The method of mutual authentication is not limited and may be certificate-based, token-based, or a combination thereof.

[0050] Task requests and execution results can be transmitted via an asynchronous queue 61. This allows for loose coupling between the requesting artificial intelligence agent node 20 and the delegated artificial intelligence agent node 40, facilitating load balancing and retry control.

[0051] As an example, we will describe a process that queries the schedules of multiple users to extract common available time slots. When a requesting user sends a task request stating "check the available time slots of multiple people for next week," the requesting artificial intelligence agent node 20 determines that the process depends on the calendar information of other users and decides that delegation is necessary.

[0052] The requesting artificial intelligence agent node 20 obtains agent description information for the delegated artificial intelligence agent node 40 corresponding to the target user from the agent registry 30, and sends task requests for obtaining free time in parallel to multiple delegates.

[0053] Each delegated artificial intelligence agent node 40 verifies that the received task request is within the scope of a read operation, accesses an external service 50 (calendar service) using the authentication information of the user associated with that delegate, obtains free time information, and returns the execution result.

[0054] The requesting artificial intelligence agent node 20 aggregates multiple execution results, extracts common available time slots, and presents them to the requesting user. This enables the automation of cross-user scheduling without sharing other users' authentication information with the requesting party.

[0055] As another embodiment, the process by which a bot node corresponding to an organizational bot delegates tasks to an individual user's node is described. When an organizational bot receives a task request that depends on personal resources such as templates held by an individual user, the bot node refers to the agent registry 30 and delegates the task request to the individual user's delegated artificial intelligence agent node 40.

[0056] The delegated artificial intelligence agent node 40 executes the task within the scope of read operations if the scope of operation is limited to read operations and returns the result, and if it includes write operations, it obtains approval from the individual user via the approval acquisition unit 43 before executing. This can prevent the bot node from operating beyond the authority of the individual user.

[0057] <Example 1> The agent registry 30 is not limited to a centralized configuration, but can also be configured in a distributed manner. Furthermore, the agent description information is not limited to being obtained through a predetermined search API; a configuration may be adopted in which description information located in a standard path is obtained.

[0058] This configuration allows for flexibility in the agent discovery method, depending on the size of the organization and network isolation requirements.

[0059] <Modification 2> The communication format for task requests is not limited; requests in the form of procedural calls can use JSON format RPC, gRPC, etc. The return of execution results is not limited to single responses; they can also be returned in streaming format via server-sent events, etc.

[0060] This configuration allows for the sequential presentation of partial results and progress notifications even for tasks with long processing times.

[0061] <Variation 3> The expression of the scope of operation is not limited and can be defined as a policy that includes an API permission list for each external service, a read-only / write-enabled classification, or an approval-required classification. The scope of operation determination unit 42 can determine whether or not to execute by comparing the operation type included in the task request with the policy.

[0062] This configuration clarifies the acceptable scope of the delegated node and can prevent abuse of authority.

[0063] <Modification 4> When sending task requests in parallel to multiple delegates, the aggregation unit 25 is not limited to simple joining, but can apply integration logic such as deduplication, inconsistency detection, or integration based on reliability. Dynamic delegate selection based on availability information may also be combined.

[0064] This configuration can improve responsiveness and result quality in distributed environments.

[0065] <Modification 5> In a configuration using an asynchronous queue 61, the requesting artificial intelligence agent node 20 can place a task request into the queue, and the delegated artificial intelligence agent node 40 can retrieve it from the queue and process it. Similarly, the execution result can also be placed into a queue and retrieved by the requesting agent.

[0066] This configuration ensures processing continuity through retries and buffering, even in the event of instantaneous load fluctuations or temporary communication interruptions. [Explanation of Symbols]

[0067] 10 Information Processing Systems 11. The requesting user's terminal 12. Approver's terminal 20 Requesting Artificial Intelligence Agent Nodes 21 Task Analysis Department 22 Delegation Decision Department 23 Registry Search Section 24. Delegation Request Department 25. Aggregation Section 26 Response generation unit 30 Agent Registry 31 Agent Description Information Storage Unit 32 Search Section 40 Delegation recipient artificial intelligence agent nodes 41 Receiving unit 42 Operation range determination unit 43 Approval Department 44 Execution Unit 45. Return Department 50 External Services 60 service meshes 61 Asynchronous Queue 70. Configuration User Interface

Claims

1. Multiple AI agent nodes, each corresponding to a different user, can access external services using each user's authentication information, Each of the aforementioned AI agent nodes has a holding unit that holds agent description information including metadata indicating the processing capacity that the node can provide, the range of operations that are permitted to be delegated to the node, and the connection destination to the node. A registry that manages the aforementioned agent description information in a searchable manner, A delegation request unit, which, when it determines that delegation to another AI agent node is necessary based on a task request from a user, obtains the agent description information of the target node from the registry and sends the task request to the target node, An execution unit in the target node executes the received task request using the authentication information of the user associated with the target node only if the received task request falls within the scope of the operation, and returns the execution result. An aggregation unit that aggregates the returned execution results and presents them to the user, An information processing system characterized by comprising the following features.

2. In the information processing system described in claim 1, The agent description information includes availability information indicating the operating status or processing load of the node. An information processing system characterized in that the delegation request unit selects a delegation destination node based on the availability information.

3. In the information processing system described in claim 1, An information processing system characterized in that the sending and receiving of the aforementioned task requests are performed over an encrypted communication channel with mutual authentication.

4. In the information processing system described in claim 1, An information processing system characterized in that the execution unit executes a task request, which includes an operation to write or change the status of an external service, only after obtaining approval from the user associated with the target node.

5. In the information processing system described in claim 1, The delegation request unit sends task requests in parallel to multiple target nodes, An information processing system characterized in that the aggregation unit generates an integrated result from multiple execution results.

6. In the information processing system described in claim 1, The aforementioned task request is a request in the form of a procedure call, An information processing system characterized in that the execution results are returned in streaming format.

7. In the information processing system described in claim 1, An information processing system further comprising a setting UI that allows the user to set the aforementioned operating range.

8. In the information processing system described in claim 1, The aforementioned AI agent node is run as a container, An information processing system characterized by the provision of mutual authentication and routing through a service mesh.

9. In the information processing system described in claim 1, An information processing system characterized in that a bot node corresponding to an organizational bot delegates task requests to the AI ​​agent node corresponding to an individual user.

10. In the information processing system described in claim 1, An information processing system characterized in that the transmission of the task request and the execution result is performed via an asynchronous queue.

Images