Agent capability description method, task execution method, device and agent

Abstract

The present disclosure provides an agent capability description method, a task execution method, an apparatus and an agent, and relates to the technical field of artificial intelligence, in particular to the technical fields of deep learning, large model, human-computer interaction, intelligent education, video generation and the like. The agent capability description method comprises: obtaining a task example for a to-be-tested agent; detecting, by a master agent, an example execution result to obtain an example detection result, the example execution result being determined based on the to-be-tested agent executing the task example, and the example detection result representing a matching degree between the example execution result and a task requirement condition for the task example; and describing, by the master agent, a task execution capability of the to-be-tested agent based on the example detection result and the task example to obtain capability description information for the to-be-tested agent.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of artificial intelligence, and particularly relates to the technical fields of deep learning, large models, human-computer interaction, intelligent education, video generation, etc. BACKGROUND

[0002] An artificial intelligence agent (AI agent for short) can be an intelligent entity for performing a specific task. With the rapid development of artificial intelligence technology, tasks such as video generation, image editing, and copywriting can be performed based on an agent. SUMMARY

[0003] The present disclosure provides an agent capability description method, a task execution method, an agent, a device, an agent, and a storage medium.

[0004] According to an aspect of the present disclosure, an agent capability description method is provided, including: obtaining a task example for a to-be-tested agent; detecting, by a master agent, an example execution result to obtain an example detection result, the example execution result being determined based on the to-be-tested agent performing the task example, and the example detection result representing a matching degree between the example execution result and a task requirement condition for the task example; and describing, by the master agent, a task execution capability of the to-be-tested agent based on the example detection result and the task example to obtain capability description information for the to-be-tested agent.

[0005] According to another aspect of the present disclosure, a task execution method based on an agent is provided, including: obtaining task requirement information; performing task intent detection on the task requirement information to obtain task description information; determining a target agent from an execution agent based on the task description information and capability description information of the execution agent, wherein the capability description information is determined according to the agent capability description method provided in an embodiment of the present disclosure; and controlling the target agent to perform a target task related to the task requirement information to obtain a task execution result.

[0006] According to another aspect of the present disclosure, an agent capability description device is provided, including: a first obtaining module configured to obtain a task example for a to-be-tested agent; an example detection result obtaining module configured to detect, by a master agent, an example execution result to obtain an example detection result, the example execution result being determined based on the to-be-tested agent performing the task example, and the example detection result representing a matching degree between the example execution result and a task requirement condition for the task example; and a capability description information obtaining module configured to describe, by the master agent, a task execution capability of the to-be-tested agent based on the example detection result and the task example to obtain capability description information for the to-be-tested agent.

[0007] According to another aspect of the present disclosure, there is provided an agent-based task execution apparatus, comprising: a second acquisition module configured to acquire task requirement information; a task description information obtaining module configured to perform task intention detection on the task requirement information to obtain task description information; a target agent determining module configured to determine a target agent from execution agents based on the task description information and capability description information of the execution agents, wherein the capability description information is determined according to the agent capability description method provided in the embodiments of the present disclosure; and a task execution result obtaining module configured to control the target agent to execute a target task related to the task requirement information to obtain a task execution result.

[0008] According to another aspect of the present disclosure, there is provided an intelligent agent of artificial intelligence, configured to execute the method provided in the embodiments of the present disclosure.

[0009] According to another aspect of the present disclosure, there is provided an electronic device, comprising: at least one processor; and a memory connected to the at least one processor in communication; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the method provided in the embodiments of the present disclosure.

[0010] According to another aspect of the present disclosure, there is provided a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause the computer to execute the method provided in the embodiments of the present disclosure.

[0011] According to another aspect of the present disclosure, there is provided a computer program product comprising a computer program which, when executed by a processor, implements the method provided in the embodiments of the present disclosure.

[0012] It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of the present disclosure, nor is it used to limit the scope of the present disclosure. Other features of the present disclosure will become apparent from the following description. BRIEF DESCRIPTION OF DRAWINGS

[0013] The accompanying drawings are used to better understand the present scheme, and do not limit the present disclosure. Among them:

[0014] Figure 1 An exemplary system architecture to which the agent capability description method and apparatus according to the embodiments of the present disclosure can be applied is schematically shown;

[0015] Figure 2 A flowchart of the agent capability description method according to the embodiments of the present disclosure is schematically shown;

[0016] Figure 3An application scenario diagram of the capability description method of the intelligent agent according to an embodiment of the present disclosure is schematically shown.

[0017] Figure 4 An application scenario diagram of the capability description method of the intelligent agent according to an embodiment of the present disclosure is schematically shown.

[0018] Figure 5 A flowchart of the intelligent agent-based task execution method according to an embodiment of the present disclosure is schematically shown.

[0019] Figure 6 An application scenario diagram of the intelligent agent-based task execution method according to an embodiment of the present disclosure is schematically shown.

[0020] Figure 7 A block diagram of the capability description device of the intelligent agent according to an embodiment of the present disclosure is schematically shown.

[0021] Figure 8 A block diagram of the intelligent agent-based task execution device according to an embodiment of the present disclosure is schematically shown.

[0022] Figure 9 A structural block diagram of the intelligent agent of artificial intelligence according to an embodiment of the present disclosure is schematically shown.

[0023] Figure 10 A schematic block diagram of an example electronic device 1000 that can be used to implement the capability description method of the intelligent agent, the intelligent agent-based task execution method of embodiments of the present disclosure is shown. DETAILED DESCRIPTION

[0024] Exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings, which include various details of embodiments of the present disclosure to assist in understanding, and should be considered as merely exemplary. Therefore, those of ordinary skill in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Also, in order to be clear and concise, descriptions of well-known functions and structures are omitted in the following description.

[0025] In the technical solutions of the present disclosure, the acquisition, storage and application of user personal information involved comply with relevant legal regulations, necessary security measures are taken, and do not violate public order and good customs.

[0026] The inventors have found that, with the rapid development of artificial intelligence technology, intelligent agents based on large models such as large language models can perform relatively complex tasks, but for personalized or diversified task requirements, intelligent agents with task execution capabilities may not be able to accurately generate task execution results, reducing the efficiency of task execution.

[0027] Embodiments of the present disclosure provide an agent capability description method, an agent task execution method, an agent, an apparatus, an agent, an electronic device, and a storage medium. The agent capability description method includes: obtaining a task example for a to-be-tested agent; detecting, by a master agent, an example execution result to obtain an example detection result, the example execution result being determined based on the to-be-tested agent executing the task example, and the example detection result representing a matching degree between the example execution result and a task requirement condition for the task example; and describing, by the master agent, a task execution capability of the to-be-tested agent based on the example detection result and the task example to obtain capability description information for the to-be-tested agent.

[0028] According to embodiments of the present disclosure, by obtaining a task example and detecting, by a master agent, an example execution result obtained by a to-be-tested agent executing the task example, an example detection result representing a matching degree between the example execution result and a task requirement condition for the task example is determined, so that the execution capability of the to-be-tested agent for the task example can be more accurately represented according to the example detection result. By describing the task execution capability of the to-be-tested agent by processing the task example and the example detection result by the master agent, the capability description information can more accurately represent the matching degree between the execution result of the to-be-tested agent in executing a target task and the task requirement condition, so that the to-be-tested agent that matches the capability requirement of the target task can be more accurately called by the capability description information to accurately execute the target task, so that an execution result that matches the task requirement condition of the target task is obtained, thereby improving the scheduling accuracy and scheduling efficiency for the agent, and improving the execution efficiency of the target task and the accuracy of the task execution result.

[0029] Figure 1 An exemplary system architecture to which the agent capability description method and apparatus according to embodiments of the present disclosure can be applied is schematically shown.

[0030] It should be noted that Figure 1 The system architecture shown is only an example of a system architecture to which embodiments of the present disclosure can be applied, to help those skilled in the art understand the technical content of the present disclosure, but does not mean that embodiments of the present disclosure cannot be applied to other devices, systems, environments, or scenarios. For example, in another embodiment, the exemplary system architecture to which the agent capability description method and apparatus can be applied can include a terminal device, but the terminal device can not need to interact with a server to implement the agent capability description method and apparatus provided by embodiments of the present disclosure.

[0031] As Figure 1As shown, the system architecture 100 according to this embodiment can include terminal devices 101, 102, 103, a network 104, and a server 105. The network 104 is a medium for providing communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 can include various connection types, such as wired and / or wireless communication links, and the like.

[0032] A user can use the terminal devices 101, 102, 103 to interact with the server 105 through the network 104 to receive or send messages, and the like. Various communication client applications can be installed on the terminal devices 101, 102, 103, such as knowledge reading applications, web browser applications, search applications, instant messaging tools, email clients, and / or social platform software, and the like (only as examples).

[0033] The terminal devices 101, 102, 103 can be various electronic devices with display screens and supporting web browsing, including but not limited to smartphones, tablet computers, laptop computers, desktop computers, and the like.

[0034] The server 105 can be a server providing various services, such as a background management server providing support for content browsed by a user using a terminal device 101, 102, 103 (only as an example). The background management server can analyze and process received user requests and the like, and feed back the processing results (such as web pages, information, or data, and the like obtained or generated according to user requests) to the terminal device.

[0035] The server 105 can be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in a cloud computing service system, and solves the defects of large management difficulty and weak business scalability in traditional physical hosts and VPS services ("Virtual Private Server", or simply "VPS"). The server 105 can also be a server of a distributed system, or a server combined with a blockchain.

[0036] It should be noted that the method for describing the capabilities of an agent provided by the embodiments of the present disclosure can generally be executed by the server 105. Accordingly, the apparatus for describing the capabilities of an agent provided by the embodiments of the present disclosure can generally be disposed in the server 105. The method for describing the capabilities of an agent provided by the embodiments of the present disclosure can also be executed by a server or a server cluster different from the server 105 and capable of communicating with the terminal devices 101, 102, 103 and / or the server 105. Accordingly, the apparatus for describing the capabilities of an agent provided by the embodiments of the present disclosure can also be disposed in a server or a server cluster different from the server 105 and capable of communicating with the terminal devices 101, 102, 103 and / or the server 105.

[0037] For example, when a user is reading an electronic book online, the terminal device 101, 102, 103 can acquire target content in the electronic book to which the user's line of sight points, and then send the acquired target content to the server 105, and the server 105 analyzes the target content to determine characteristic information of the target content; predicts content of interest to the user according to the characteristic information of the target content; and extracts the content of interest to the user. Or the target content is analyzed by a server or a server cluster capable of communicating with the terminal device 101, 102, 103 and / or the server 105, and finally the content of interest to the user is extracted.

[0038] It should be understood that Figure 1 The number of terminal devices, networks and servers in the above embodiment is only illustrative. Any number of terminal devices, networks and servers can be provided according to the needs of implementation.

[0039] Figure 2 A flowchart of a capability description method of an agent according to an embodiment of the present disclosure is schematically shown.

[0040] As Figure 2 The capability description method of the agent of this embodiment includes operations S210-S230.

[0041] In operation S210, a task example for a to-be-tested agent is acquired.

[0042] In operation S220, the example execution result is detected by the master agent to obtain an example detection result.

[0043] In operation S230, the master agent describes the task execution capability of the to-be-tested agent based on the example detection result and the task example to obtain capability description information for the to-be-tested agent.

[0044] According to an embodiment of the present disclosure, the execution subject of operation S210 can be the master agent, or the execution subject of operation S210 can also be a terminal or other server in communication connection with the server in which the master agent is set.

[0045] According to an embodiment of the present disclosure, the execution subjects of operation S220 and operation S230 can be the master agent, and the master agent is configured on a server or a terminal device. The to-be-tested agent can be configured on the same or different electronic device as the master agent, such as a server or a terminal device, as long as it can satisfy the communication between the master agent and the to-be-tested agent.

[0046] The main agent and the to-be-tested agent can both be capable of perceiving an environment and taking actions to achieve a specific goal. The main agent and the target agent can both be designed to be highly scalable, for example, designed as agents configured with large models. The types of large models are not limited, for example, the large models can include one or a combination of a large language model (LLM), a large vision model (LVM), or a multimodal large model (MLM). The main agent and the to-be-tested agent differ from each other in that the configured large models have different large model knowledge and are capable of performing different types of tasks.

[0047] According to an embodiment of the present disclosure, the task examples can be executed by the to-be-tested agent, and the task examples can include any number or type, for example, the task examples can be video generation task examples, copywriting generation task examples. However, the task examples can also be multiple tasks with dependency relationships, and the to-be-tested agent can execute the multiple task examples according to the dependency relationships to obtain respective example execution results of the multiple task examples.

[0048] According to an embodiment of the present disclosure, the example execution result is determined based on the to-be-tested agent executing the task example, for example, the example execution result can be obtained by the main agent controlling the to-be-tested agent to execute the task example. The example execution result can include any type of information output by the to-be-tested agent, such as copywriting, images, videos, etc. Embodiments of the present disclosure do not limit the specific type of the example execution result.

[0049] According to an embodiment of the present disclosure, the example detection result represents the matching degree between the example execution result and the task requirement condition for the task example. The task requirement condition for the task example can indicate the attribute requirements of the example execution result, such as the type, quality, etc. of the example execution result. For example, the task requirement condition can represent the text word quantity requirement, copywriting logical coherence requirement, copywriting structure requirement, etc. of the copywriting generation example. For another example, the task requirement condition can also represent the correctness requirement, number of problem solving steps requirement, problem solving knowledge point requirement, etc. of the problem solving example task. Embodiments of the present disclosure do not limit the requirement type represented by the task requirement condition for the example execution result.

[0050] The example detection result can represent the matching degree based on non-quantitative information such as character identification or text, for example, the example detection result can represent the matching degree between the example execution result and the task requirement condition for the task example based on words such as "match" and "not match". But not limited to this, the example detection result can also represent the matching degree based on quantitative data, for example, the matching degree between the example execution result and the task requirement condition for the task example can be represented based on numerical values, matrices and other information. The embodiments of the present disclosure do not limit the information type of the example detection result.

[0051] In some embodiments, the task examples for the agent to be tested can include multiple task examples with dependency relationships, and the main agent can control the agent to be tested to execute the multiple task examples according to the dependency relationships to obtain multiple example execution results.

[0052] For example, the multiple task examples with dependency relationships can be multiple task examples for generating a problem solving video. The multiple task examples can include a problem solving task example, a problem statement generation example, a problem illustration generation example, and a video synthesis task example. The agent to be tested executes the multiple task examples based on the dependency relationships to obtain example execution results of the multiple task examples, which can be a problem solving result, a problem statement, a problem illustration data, and a problem solving video, respectively.

[0053] In some embodiments, the main agent can be used to detect the example execution result, which is suitable for executing a general detection task. But not limited to this. The main agent can also be provided with multiple sub-specialist agents, and the sub-specialist agents are suitable for executing an individualized detection task on the example execution result of a specified type of task example, so as to improve the detection accuracy of the example detection result.

[0054] The model structure configured in the general agent or the sub-specialist agent is not limited, for example, it can include a deep learning model of one or more of attention mechanism, convolution network, recurrent network, and long short-term memory network, but not limited to this, it can also be a large language model with more powerful functions. As long as it is a model that can be used to execute a detection task. The difference is that multiple sub-specialist agents can form a mixture of experts mechanism (Mixture of Experts), which combines multiple sub-specialist agents with pertinence and individualization to detect different example execution results for different task requirement conditions.

[0055] In some embodiments, detecting the example execution result by the master agent can include processing the example execution result and the task requirement condition by the master agent. Thus, the master agent can be prompted to understand the ability and analyze the matching degree between the example execution result and the task requirement condition based on the task requirement condition, so as to improve the detection accuracy of the example detection result for the example execution result.

[0056] According to an embodiment of the present disclosure, describing the task execution ability of the to-be-tested agent by the master agent based on the example detection result and the task example can include processing the example detection result and the task example by the master agent. Thus, the master agent can accurately describe the execution ability and the execution effect of the to-be-tested agent for the task example by understanding the matching degree between the example execution result represented by the example detection result and the corresponding task requirement condition, so as to accurately represent the execution ability of the to-be-tested agent for the target task to be executed by the ability description information. Then, the agent suitable for executing the target task to be processed can be selected based on the ability description information, so as to improve the execution efficiency and the task execution result accuracy of the target task.

[0057] In one example, the plurality of task examples can be task examples for detecting different dimension or different attribute abilities of the to-be-tested agent. The task execution ability of the to-be-tested agent is described by processing a large number of task examples and example detection results by the master agent, so as to describe the multi-dimensional ability of the execution agent and output the multi-dimensional ability description information of the to-be-tested agent. It should be noted that the dimension or attribute of the task example can include the attribute of the requirement information represented by the task requirement condition, such as the text word quantity attribute of the generated text, the complexity attribute of the generated text, and the like.

[0058] In some embodiments, the task examples for the to-be-tested intelligent agent can include a plurality of task examples having dependency relationships, the to-be-tested intelligent agent includes a plurality of sub-to-be-tested intelligent agents arranged according to execution logic, the plurality of sub-to-be-tested intelligent agents can be respectively used to execute the plurality of task examples to obtain a plurality of example execution results. The example detection result can represent a matching degree between an example execution result obtained by the sub-to-be-tested intelligent agent executing the task example and a task demand condition for the task example, so that the execution capability of the sub-to-be-tested intelligent agent for the task example can be represented based on the example detection result. By processing the plurality of example detection results and the plurality of task examples by using the master intelligent agent, the obtained capability description information can more accurately describe the execution capability of each sub-to-be-tested intelligent agent in the to-be-tested intelligent agent for the corresponding to-be-processed target task, and the execution capability of the to-be-tested intelligent agent for executing the plurality of task examples having dependency relationships as a whole. Therefore, the execution capability of each sub-to-be-tested intelligent agent in the to-be-tested intelligent agent for the task can be more accurately evaluated based on the method provided in the embodiments of the present disclosure, and the cooperation capability among the plurality of sub-to-be-tested intelligent agents can be more accurately evaluated, so that the plurality of sub-to-be-tested intelligent agents that match the target task having the dependency relationship can be selected to cooperate, so as to improve the execution efficiency of the complex target task.

[0059] It should be noted that the plurality of task examples having dependency relationships in the embodiments can be understood as a task example topology. The to-be-tested intelligent agent can execute the plurality of task examples based on the dependency relationships between the plurality of task examples indicated by the task example topology, and can obtain an execution result of the task example topology. For example, the task example topology can include a problem solving task example, a problem presentation script generation example, a problem presentation audio synthesis task example, and a video generation task example. The execution result of the task example topology can be a generated problem presentation video.

[0060] The capability description method of the intelligent agent provided in the embodiments of the present disclosure will be explained and described below in combination with the drawings and specific embodiments.

[0061] In some embodiments, the task examples include at least one of a video generation task example, an image editing task example, and a script generation task example.

[0062] The to-be-tested intelligent agent or the sub-to-be-tested intelligent agent can generate an advertisement video, an animation video, a problem presentation video, or any type of video by executing the video generation task example. The to-be-tested intelligent agent or the sub-to-be-tested intelligent agent can generate a video by processing video material data such as a problem presentation script and a virtual human image.

[0063] The image editing task example can represent editing tasks such as color editing and picture object adding on an image. The example execution result of the image editing task example can be an edited image.

[0064] The text generation task example can represent a task example for generating a novel, a news report, a speech text, and the like. The agent under test or the sub-agent under test can generate the text by performing the text generation task example according to the requirement condition.

[0065] It should be noted that the task example can also include examples for performing other tasks, for example, the task example can also be an image detection task example, a problem solving task example for solving a problem, a code generation task example for generating code, and the like. Embodiments of the present disclosure do not limit the specific task type of the task example.

[0066] In some embodiments, the main agent can include any one or more of the first sub-agent, the second sub-agent, the third sub-agent, and the fourth sub-agent. The first sub-agent, the second sub-agent, the third sub-agent, and the fourth sub-agent can be expert agents respectively for performing different specified tasks, so that the capability description method of the agent provided by the embodiments of the present disclosure can be performed through the cooperation between the multiple sub-agents in the main agent.

[0067] In some embodiments, the example detection result is obtained by detecting the example execution result using the main agent, including: processing the multiple example execution results using the fourth sub-agent of the main agent, and obtaining multiple example detection results for the task requirement condition of each of the multiple example execution results.

[0068] According to embodiments of the present disclosure, the multiple example detection results can represent the execution capabilities of the agent under test or the sub-agent under test for performing different task examples, so that the fourth sub-agent can process the multiple task examples and the multiple example detection results by obtaining multiple different task examples of different task types, different task parameters, and the like, so that the capability description information can represent the execution capabilities of the agent under test or the sub-agent under test for the target task of the specific task attribute of multiple dimensions, so as to accurately detect and evaluate the execution capabilities of the agent under test, and improve the efficiency and accuracy of subsequent agent scheduling for the target task.

[0069] In some embodiments, the execution result of a preceding task example in the multiple task examples having a dependency relationship can be used as task element data of a subsequent task example. For example, the preceding task example is a speech text generation task example, and the example execution result of the speech text generation task example is a speech text. The subsequent task example is a speech synthesis task example, and the agent under test can perform the speech synthesis task example by processing the speech text to generate an example execution result representing oral broadcast speech data of the speech text.

[0070] By utilizing the master agent to process the multiple task examples with dependency relationship and the example detection results of the multiple task examples respectively, the example execution result can be detected to be abnormal in the process of executing a specific task example in the multiple task examples based on the powerful understanding capability of the master agent, so as to more accurately analyze that the execution capability of the to-be-tested agent for the specific task example has defects, so that the capability description information can accurately describe the execution capability of the to-be-tested agent for the complex multi-step task.

[0071] In some embodiments, the to-be-tested agent includes multiple sub-to-be-tested agents arranged according to execution logic, and the multiple sub-to-be-tested agents can be respectively used to execute multiple task examples with dependency relationship. By utilizing the master agent to process the multiple task examples with dependency relationship and the example detection results of the multiple task examples respectively, the capability description information can more accurately describe the collaboration capability between the multiple sub-to-be-tested agents, so as to improve the accuracy of the capability boundary description of the to-be-tested agent as a workflow execution component for the collaboration of the multiple sub-agents.

[0072] In some embodiments, obtaining the task examples for the to-be-tested agent includes: obtaining multiple task examples for describing the capability of the agent from a task example library, and the multiple task examples can have different task attributes and different task requirement conditions, so as to generate the capability description information accurately describing the capability of the to-be-tested agent through the example execution results of the multiple task examples respectively.

[0073] In some embodiments, obtaining the task examples for the to-be-tested agent includes: utilizing the master agent to process the agent attributes of the to-be-tested agent to obtain the task examples.

[0074] According to embodiments of the present disclosure, the agent attribute represents attribute information for describing a specific function of the to-be-tested agent or the sub-to-be-tested agent, and the agent attribute can be obtained in a pre-test stage or determined based on the model performance or model capability of the large model for constructing the agent. The agent attribute can include, for example, information such as the type of task that the to-be-tested agent or the sub-to-be-tested agent can execute and the task attribute parameters.

[0075] In some examples, the master agent can be utilized to process the agent task attribute to obtain the task examples for at least one sub-to-be-tested agent in the to-be-tested agent. Thus, the multiple task examples generated by the master agent can be executed based on the dependency relationship between the multiple to-be-tested agents arranged, so as to accurately describe the execution capability boundary of each sub-to-be-tested agent for different tasks, to improve the accuracy of the capability description information in describing the capability boundary of each sub-to-be-tested agent in the to-be-tested agent as a workflow component, and to further improve the accuracy of the capability boundary description of the to-be-tested agent for a complex multi-task.

[0076] In some embodiments, processing the agent attribute of the to-be-tested agent by the main agent to obtain the task example includes: processing the agent attribute of the to-be-tested agent by a first sub-agent of the main agent to obtain initial capability description information related to the to-be-tested agent; and processing the initial capability description information by a third sub-agent of the main agent to obtain the task example.

[0077] According to embodiments of the present disclosure, processing the agent attribute of the to-be-tested agent by the first sub-agent of the main agent can include processing the agent attribute of the to-be-tested agent by the first sub-agent to obtain initial capability description information of the to-be-tested agent and initial capability description information of each of a plurality of sub-to-be-tested agents in the to-be-tested agent. The initial capability description information can represent a capability boundary range of the to-be-tested agent or the sub-to-be-tested agent in performing a task, for example, the initial capability description information can represent a word number range, a logical coherence score, etc. of the to-be-tested agent or the sub-to-be-tested agent in performing a copywriting generation task, or the initial capability description information can represent a knowledge point range involved in the to-be-tested agent or the sub-to-be-tested agent in performing a problem solving task, etc. Embodiments of the present disclosure do not limit the specific type of the capability boundary range, as long as it is suitable for the task requirement condition for which the task example needs to be generated.

[0078] The initial capability description information can be used as an initial prompt word of the capability boundary range to control the third sub-agent to generate diversified task examples in combination with an example prompt word representing a demand intention for the task example, so that the to-be-tested agent or the sub-to-be-tested agent can be controlled to perform the task example of different task requirement conditions. For example, the initial capability description information and the example prompt word can be used to prompt the third sub-agent to generate a boundary-out task example that exceeds the execution capability boundary range of the sub-to-be-tested agent, or the initial capability description information and the example prompt word can be used to prompt the third sub-agent to generate a boundary-in task example that belongs to the execution capability boundary range of the sub-to-be-tested agent. For example, the boundary-out task example can be to generate an explanatory copy with 100,000 words, and the boundary-in task example can be to generate an explanatory copy with 500 words.

[0079] Since the initial capability description information is usually not accurate in describing the execution capability boundary of the to-be-tested agent or the sub-to-be-tested agent for a task, and the quality of the task execution result of the agent fluctuates greatly, the initial capability description information and the example prompt word can be processed by the third sub-agent to generate a plurality of different boundary-in task examples and a plurality of boundary-out task examples to test the to-be-tested agent or the sub-to-be-tested agent in multiple dimensions, thereby improving the execution capability dimension represented by the subsequent example detection result, and improving the accuracy of the capability description information.

[0080] In an embodiment, the example prompt word can be obtained by the first sub-agent by processing the agent attribute, so as to control the diversity of the task examples by the understanding ability of the first sub-agent.

[0081] According to an embodiment of the present disclosure, the first sub-agent and the third sub-agent are agents with different functions and suitable for performing different types of tasks. By generating the task examples through the cooperation of the first sub-agent and the third sub-agent, the task examples can be generated in combination with the example prompt word and the agent attribute to generate the task examples for evaluating the task execution ability of the to-be-tested agent, so as to improve the data scale and diversity of the task examples, and then the ability description information that can accurately represent the task execution ability of the to-be-tested agent can be obtained by detecting the example execution result by using the main agent and performing the ability description. Thus, the task execution abilities of the multiple sub-to-be-tested agents and the cooperation abilities between the multiple sub-to-be-tested agents can be more accurately represented by the ability description information for the to-be-tested agent obtained by logically arranging the multiple sub-to-be-tested agents.

[0082] In some embodiments, based on the example detection result and the task example, the ability description of the to-be-tested agent by using the main agent includes: updating the task example by using the first sub-agent of the main agent based on the example detection result to obtain an intermediate task example; and processing the intermediate task example and the intermediate detection result by using the second sub-agent of the main agent to obtain the ability description information.

[0083] According to an embodiment of the present disclosure, the intermediate detection result is determined based on the intermediate execution result, for example, can be obtained by detecting the intermediate execution result by using the main agent, and the intermediate execution result is determined by the to-be-tested agent executing the intermediate task example. The intermediate detection result can represent the matching degree between the intermediate task requirement condition for the intermediate task example and the intermediate execution result.

[0084] According to an embodiment of the present disclosure, updating the task example by using the first sub-agent can include processing the example detection result and the task example by using the agent with the task example generation function to obtain an updated intermediate task example. In this way, the execution capability of the agent under test or the sub-agent under test for the task example can be understood based on the first sub-agent by processing the example detection result, and the intermediate task example obtained by updating the task example can represent the new in-bound task example and the out-bound task example. In this way, the execution capability of the agent under test or the sub-agent under test for the task example of each task attribute can be further understood in depth by processing the intermediate task example and the intermediate detection result by using the second sub-agent with a relatively powerful data semantic understanding capability, and the execution capability boundary range of the agent under test or the sub-agent under test can be accurately described by generating the capability description information, thereby improving the capability description accuracy of the agent under test, and improving the scheduling efficiency, the task execution efficiency and the execution quality of calling the agent to execute the task for different types of tasks.

[0085] In one embodiment, the intermediate detection result is obtained by processing the intermediate execution result by using a fourth sub-agent of the main agent, and an intermediate task requirement condition for the intermediate task example. The intermediate task requirement condition can be output by the third sub-agent. The intermediate task requirement condition can be used as an execution condition of the intermediate task example to control the agent under test or the sub-agent under test to execute the intermediate task example based on the intermediate task requirement condition.

[0086] In one embodiment, the capability description method of the agent can further include iteratively processing the currently generated at least one intermediate task example and the intermediate detection result as the currently generated task example by using the first sub-agent of the main agent to process the example detection result of the currently generated task example, to update the currently generated task example in each iteration round to obtain a new intermediate task example, so that the agent under test or the sub-agent under test can be controlled to execute the intermediate task examples generated in multiple rounds by the cooperation of multiple sub-agents in the main agent to determine the intermediate execution result and the intermediate detection result in multiple rounds. The second sub-agent is used to process the intermediate task examples, the intermediate execution result and the intermediate detection result in multiple rounds as the currently generated task example, the example execution result and the example detection result to understand the task execution capability of the agent under test, and to obtain the capability description information which more accurately describes the capability boundary range of the agent under test.

[0087] In some embodiments, based on the example detection result, updating the task example by using the first sub-agent of the main agent to obtain the intermediate task example includes: based on the example detection result, updating the initial capability description information for the agent under test by using the first sub-agent to obtain intermediate capability description information; and processing the intermediate capability description information by using the third sub-agent of the main agent to obtain the intermediate task example.

[0088] In an embodiment, the updating, based on the example detection result, the initial capability description information of the to-be-tested intelligent agent by the first sub-intelligent agent can include processing, by the first sub-intelligent agent, the example detection result of the current generated task example, so that the first sub-intelligent agent understands the execution capability of the to-be-tested intelligent agent for the diversified task example through the example detection result of the current generated task example, and thus the intermediate capability description information can more accurately represent the execution capability boundary range of the to-be-tested intelligent agent.

[0089] In an embodiment, the capability description method of the intelligent agent can further include iteratively processing, by the first sub-intelligent agent of the main intelligent agent, the example detection result of the current generated at least one intermediate task example and intermediate detection result as the current generated task example, to generate new intermediate capability description information in each iteration round. Processing, by the third sub-intelligent agent, the intermediate capability description information generated in each round to obtain the intermediate task example of the current round.

[0090] In an embodiment, the intermediate task example generated by the main intelligent agent in at least one round can be a plurality of intermediate task examples suitable for execution by different sub-to-be-tested intelligent agents in the to-be-tested intelligent agent, so that the execution capability of the different sub-to-be-tested intelligent agents for cooperatively executing complex multi-task can be detected by generating a plurality of intermediate task examples with dependency relationship for a plurality of sub-to-be-tested intelligent agents executing the logic arrangement, to improve the accuracy of the capability description information for the complex multi-intelligent agent cooperation capability of the execution logic, and improve the scheduling efficiency and accuracy of the intelligent agent.

[0091] According to an embodiment of the present disclosure, the task examples can include a plurality of task examples, and the to-be-tested intelligent agent includes a plurality of sub-to-be-tested intelligent agents with dependency relationship, and the plurality of sub-to-be-tested intelligent agents execute the plurality of task examples based on the arranged dependency relationship.

[0092] Processing, by the third sub-intelligent agent of the main intelligent agent, the intermediate capability description information to obtain the intermediate task example can include processing, by the third sub-intelligent agent, the defect description information in the intermediate capability description information to obtain a first intermediate example.

[0093] According to an embodiment of the present disclosure, the defect description information characterizes a defect factor of the first example execution result, and the first example execution result is determined by the sub-to-be-tested intelligent agent executing a first task example in the plurality of task examples. The defect factor can represent a defect type of the first example execution result, a defect score, and other defect attributes of the example execution result, for example, the defect factor can be represented as a code error type of the generated code.

[0094] In one example, the defect description information represents a degree of mismatch between the first example execution result and the corresponding task requirement condition, or the defect description information can also describe a difference degree between the first example execution result and a benchmark execution result represented by the corresponding task requirement condition. For example, the defect factor can be represented as a difference value between a logical coherence score of the generated script and a benchmark script logical coherence score represented by the task requirement condition.

[0095] In one embodiment, the defect description information can describe that the sub-agent-under-test fails to successfully execute the corresponding task example or the intermediate task example.

[0096] By utilizing the third sub-agent to process the defect description information in the intermediate capability description information, the defect factor indicated by the defect description information can be used as a prompt word to control the third sub-agent to adjust the task requirement condition for the currently generated task example or the intermediate task example according to the defect description information, so as to adjust the currently generated task example or the intermediate task example, so that the generated first intermediate example can be successfully executed by the sub-agent-under-test, or the example execution result obtained by the sub-agent-under-test executing the first intermediate example can be closer to the task requirement condition for the first intermediate example. Thus, the first intermediate example can be iteratively generated through multiple rounds to enable the sub-agent-under-test to eventually successfully execute the currently generated first intermediate example, so that the second sub-agent can understand the execution capability of the sub-agent-under-test for multiple task examples corresponding to changing task requirement conditions by understanding the intermediate execution results and the intermediate detection results of multiple rounds, thereby more accurately describing the task execution capability boundary of the sub-agent-under-test and improving the description accuracy of the capability boundary of the agent.

[0097] In one embodiment, based on the example detection result, the initial capability description information for the agent-under-test is updated by the first sub-agent to obtain the intermediate capability description information, including: processing the defect detection result in the example detection result by the first sub-agent to obtain defect description information for the sub-agent-under-test.

[0098] The defect detection result represents that the first example execution result does not match the task requirement condition for the first task example. For example, the defect detection result can represent that the generated image is different from the vehicle type required by the task requirement condition, or the vehicle color required by the task requirement condition, etc. Or the defect detection result can also represent that one or more steps in the problem solving step data have problem solving errors.

[0099] By utilizing the first sub-agent to process the defect detection result in the example detection result, it can be prompted that the first sub-agent sub-test intelligent agent is difficult to perform according to the task demand condition in the case of executing a specific task example, so that the defect detection result can be used as a prompt to control the first sub-agent to adjust the ability description of the sub-test intelligent agent, so that the third sub-agent can generate a new intermediate task example by using the defect description information which more accurately describes the defect factors of the sub-test intelligent agent, so as to detect the ability boundary range of the sub-test intelligent agent according to the new intermediate task example.

[0100] In some embodiments, the task examples further include a second task example, and processing the intermediate ability description information by the third sub-agent of the main agent to obtain the intermediate task example can further include: processing the completion description information in the intermediate ability description information by the third sub-agent to obtain a second intermediate example.

[0101] Wherein, the completion description information represents that the second example execution result matches the task demand condition for the second task example, and the sub-test intelligent agent has successfully executed the second example based on the completion description information. The intermediate task demand condition for the second intermediate example is different from the task demand condition for the second task example.

[0102] In one embodiment, the intermediate task demand condition for the second intermediate example can represent a completely different demand type from the demand type represented by the task demand condition for the second task example, for example, the task demand condition for the second task example is a word number demand condition "at least 500 words", and the intermediate task demand condition for the second intermediate example can be represented as "the copywriting has a summary section, an execution step section and a summary section". By generating a second intermediate example with a different demand type, the multi-dimensional ability boundary detection of the sub-test intelligent agent is performed to improve the multi-dimensional ability description accuracy of the ability description information for multiple sub-test intelligent agents of the test intelligent agent.

[0103] In one embodiment, the intermediate task demand condition for the second intermediate example can represent the same demand type as the demand type represented by the task demand condition for the second task example, but the demand attribute information of the demand type is different. For example, the task demand condition for the second task example is a word number demand condition "at least 500 words", and the intermediate task demand condition for the second intermediate example can be represented as "at least 5000 words". By adjusting the demand attribute of the task demand condition of the same demand type to generate a new intermediate task example, the multiple sub-agents in the main agent can be used to explore the execution ability of the sub-test intelligent agent for the task example of the same demand type of the demand condition, and the ability description information output by the main agent can more accurately represent the task execution ability boundary and the ability range of the test intelligent agent, thereby improving the scheduling accuracy of the intelligent agent.

[0104] Figure 3 An application scenario diagram of the capability description method of the intelligent agent is schematically shown according to an embodiment of the present disclosure.

[0105] As shown in Figure 3 The main intelligent agent can include a first sub-intelligent agent 311, a second sub-intelligent agent 312, a third sub-intelligent agent 313, and a fourth sub-intelligent agent 314. The to-be-tested intelligent agent 320 can include a plurality of sub-to-be-tested intelligent agents based on logical orchestration. The sub-to-be-tested intelligent agents can be represented based on solid circle graph elements, and the dependency relationship can be represented as edges between different solid circle graph elements.

[0106] In the 0th round (or initial round), the intelligent agent attribute is input into the first sub-intelligent agent 311, the intelligent agent attribute of the to-be-tested intelligent agent 320 is processed by using the first sub-intelligent agent 311, and initial capability description information is obtained. The initial capability description information is processed by using the third sub-intelligent agent 313 to generate a plurality of task examples of the respective sub-to-be-tested intelligent agents. The plurality of task examples are executed by the plurality of sub-to-be-tested intelligent agents of the to-be-tested intelligent agent 320 through an interface call, and the example execution results are sent to the fourth sub-intelligent agent 314. The plurality of task examples, the task requirement conditions for the task examples, and the plurality of example execution results are processed by using the fourth sub-intelligent agent 314 to obtain a plurality of example detection results.

[0107] In the 1st round, the plurality of example detection results are processed by using the first sub-intelligent agent 311 to obtain 1st intermediate capability description information representing the collaboration capability between the plurality of sub-to-be-tested intelligent agents and the respective task execution capability of the plurality of sub-to-be-tested intelligent agents. The 1st intermediate capability description information is processed by using the third sub-intelligent agent 313 to generate a plurality of 1st intermediate task examples of the respective sub-to-be-tested intelligent agents. The plurality of 1st intermediate task examples are executed by the plurality of sub-to-be-tested intelligent agents of the to-be-tested intelligent agent 320 through an interface call, and 1st intermediate execution results are sent to the fourth sub-intelligent agent 314. The plurality of 1st intermediate task examples, the 1st intermediate task requirement conditions for the 1st intermediate task examples, and the plurality of 1st intermediate execution results are processed by using the fourth sub-intelligent agent 314 to obtain a plurality of 1st intermediate detection results.

[0108] In the second round, the first intermediate detection result and the example detection result of the 0th round are taken as the currently generated example detection result, and the first sub-agent 311 is used to process the currently generated example detection result to obtain the second intermediate capability description information of the to-be-tested agent. The third sub-agent 313 is used to process the second intermediate capability description information to generate the second intermediate task examples of the plurality of sub-to-be-tested agents. The plurality of sub-to-be-tested agents of the to-be-tested agent 320 are controlled to execute the second intermediate task examples by calling the interface, and the fourth sub-agent 314 is sent the second intermediate execution results. The fourth sub-agent 314 is used to process the plurality of second intermediate task examples, the second intermediate task requirement conditions for the second intermediate task examples, and the plurality of second intermediate execution results to obtain the plurality of second intermediate detection results.

[0109] It should be understood that in the nth round, the (n-1)th intermediate detection result to the example detection result of the 0th round are taken as the currently generated example detection result, and the first sub-agent 311 is used to process the currently generated example detection result to obtain the nth intermediate capability description information of the to-be-tested agent. The third sub-agent 313 is used to process the nth intermediate capability description information to generate the nth intermediate task examples of the plurality of sub-to-be-tested agents. The plurality of sub-to-be-tested agents of the to-be-tested agent 320 are controlled to execute the nth intermediate task examples by calling the interface, and the fourth sub-agent 314 is sent the nth intermediate execution results. The fourth sub-agent 314 is used to process the plurality of nth intermediate task examples, the nth intermediate task requirement conditions for the nth intermediate task examples, and the plurality of nth intermediate execution results to obtain the plurality of nth intermediate detection results. Wherein, n is an integer greater than 1.

[0110] In the case of N=n, the intermediate task examples, intermediate execution results and intermediate detection results of n rounds, and the task examples, example detection results and example execution results of the 0th round can be input to the second sub-agent 312 as the currently generated task examples, example detection results and example execution results. The second sub-agent 312 is used to describe the task execution capability of the to-be-tested agent 320 to obtain the capability description information.

[0111] In some embodiments, the capability description information includes at least one of the following: a dependency relationship between the plurality of sub-to-be-tested agents in the to-be-tested agent; data attribute conditions of input data and output data of the sub-to-be-tested agent, and defect examples and complete examples.

[0112] In one embodiment, the data attribute conditions of the input data and the output data of the sub-to-be-tested agent can represent constraint conditions of data format, data type, numerical range, etc. data attribute information of the input data or the output data.

[0113] The first example execution result obtained by the sub-agent under test executing the first task example in the task examples does not match the task requirement condition for the first task example. The defect example can represent a boundary-outside task example related to the sub-agent under test, which is difficult to generate an execution result matching the task requirement condition according to the task requirement condition of the defect example.

[0114] The second example execution result obtained by the sub-agent under test executing the second task example in the task examples matches the task requirement condition for the second task example. The complete task example can represent a boundary-inside task example related to the sub-agent under test. The sub-agent under test can efficiently execute the complete task example and obtain an example execution result matching the task requirement condition.

[0115] It should be noted that the complete example or the defect example can be the task example in the above embodiments, or can also be an intermediate task example. For example, the complete example or the defect example is a task example and an intermediate task example generated by the master agent at each round based on multiple rounds of cooperation.

[0116] In some embodiments, the capability description information can further include a boundary task example. The boundary task example can represent that the sub-agent under test can successfully execute under specific conditions and obtain an example execution result satisfying the task requirement condition. However, the quality of the example execution result is not stable, or there is a result error. The boundary task example can more accurately represent the task execution capability boundary of the sub-agent under test, so as to more accurately represent the task execution capability of the agent under test through the boundary task example and the description text for describing the boundary task example.

[0117] In some embodiments, the capability description information can further include example execution results of the task examples and the intermediate task examples, and a dependency relationship between multiple sub-agents under test in the agent under test. The dependency relationship can include a communication order, a data interface of input data and output data, a logic control unit for controlling cooperation of the multiple sub-agents under test, and the like. Thus, the target task can be accurately invoked by the agent through the capability description information.

[0118] Figure 4 An application scenario diagram of a capability description method of an agent according to an embodiment of the present disclosure is schematically shown.

[0119] As Figure 4As shown, the agent attribute of the to-be-tested agent 410 can be input to the agent capability evaluation system 401. The agent capability evaluation system 401 can be constructed based on a plurality of sub-agents in the master agent provided in the embodiments of the present disclosure. The agent capability evaluation system 401 can generate capability description information 420 by performing the method for describing an agent provided in the embodiments of the present disclosure. The capability description information 420 can include data attribute conditions of input data and output data for the to-be-tested agent 410, dependency relationship description text for describing the dependency relationship between a plurality of to-be-tested agents 410 in the to-be-tested agent 410, defect examples, complete examples, defect example execution results, complete example execution results, and capability description text. The capability description information can be stored based on structured information, so as to call the to-be-tested agent 410 through the structured capability description information.

[0120] By performing the method for describing an agent provided in the embodiments of the present disclosure, the capability description information of the to-be-tested agent is determined, which can accurately describe the task execution capability of a complex to-be-tested agent constructed based on a plurality of sub-agent cooperation processes, and accurately and flexibly schedule a sub-agent cooperation process matched with a target task based on the capability description information, so as to improve the scheduling accuracy and deployment flexibility of the agent cooperation mechanism in diversified application scenarios, and improve the execution efficiency and execution accuracy of complex tasks.

[0121] Figure 5 A flowchart of a method for executing a task based on an agent according to an embodiment of the present disclosure is schematically shown.

[0122] As shown, Figure 5 The method for executing a task based on an agent includes operations S510-S540.

[0123] In operation S510, task requirement information is obtained.

[0124] In operation S520, task intention detection is performed on the task requirement information to obtain task description information.

[0125] In operation S530, a target agent is determined from the execution agent based on the task description information and the capability description information of the preset execution agent.

[0126] According to the embodiments of the present disclosure, the capability description information is determined according to the method for describing an agent provided in the embodiments of the present disclosure.

[0127] In operation S540, the target agent is controlled to execute a target task related to the task requirement information to obtain a task execution result.

[0128] According to an embodiment of the present disclosure, the task demand information can be demand information for indicating a target task to be executed. The task demand information can be represented based on data types in any format such as voice or text.

[0129] In some embodiments, the task intention detection on the task demand information can include processing the task demand information by using a trained large language model to obtain task description information. The task description information can represent text or characters and other data that describe target task demand conditions, target task attribute parameters, and other task attributes of the target task to be executed.

[0130] In some embodiments, the matching result can be obtained by matching the task description information with the capability description information of the execution agent. The target agent suitable for executing the target task can be determined from the plurality of execution agents according to the matching result.

[0131] It should be noted that the above operations S510-S540 can be performed based on an electronic device such as a server.

[0132] In some embodiments, the above operations S510-S540 can be performed based on a scheduling agent arranged in a server or a server cluster. The scheduling agent can be used to generate a target task according to the demand information of a user, and to call a target agent to execute the target task, so as to obtain a high-precision task execution result of the target task.

[0133] In some embodiments, the execution agent includes a plurality of sub-execution agents having a dependency relationship. The plurality of sub-execution agents can be used to execute the target task according to an execution order or execution logic of the plurality of target tasks, so as to obtain an execution result of the target task.

[0134] In some embodiments, the scheduling agent can include a plurality of sub-scheduling agents. The plurality of sub-scheduling agents can be used to perform different operation steps in the above operations S510-S540.

[0135] In some embodiments, determining the target agent from the execution agent includes determining the target agent from the plurality of execution agents by using a second sub-scheduling agent based on the capability description information and the task description information.

[0136] For example, the second sub-scheduling agent can be used to process the capability description information and the task description information of each of the plurality of execution agents, so as to determine target capability description information matched with the task description information. Then, the execution agent corresponding to the target capability description information can be taken as the target agent. Thus, in the case where the execution agent includes a plurality of sub-execution agents, the target agent suitable for executing the target task can be quickly determined by the capability description information and the task description information, and the execution efficiency of the target task can be improved.

[0137] In an embodiment, determining the target agent from the execution agents comprises: processing the task description information and the capability description information of the sub-execution agents by the first sub-scheduling agent to obtain sub-target agents in the plurality of sub-execution agents; and logically arranging the plurality of sub-target agents by the first sub-scheduling agent to obtain the target agent.

[0138] According to an embodiment of the present disclosure, the sub-target agents are used to execute associated target tasks in the plurality of target tasks. By logically arranging the plurality of sub-target agents in the target agent by the first sub-scheduling agent, the dependency relationship between the plurality of sub-target agents is obtained, so as to implement the target agent capable of executing the complex execution process in a fine-grained manner to execute the plurality of target tasks. The respective associated target tasks are executed by the dependency relationship to obtain the task execution results of the plurality of associated target tasks, so as to improve the execution efficiency and accuracy of the target tasks.

[0139] In an embodiment, the execution agents comprise a plurality of sub-target agents, and at least two of the plurality of sub-target agents are determined from different execution agents. By determining the sub-execution agents respectively matched with the plurality of target tasks from the plurality of execution agents, and logically arranging the plurality of sub-execution agents to obtain the target agent, the complex execution process for executing the target tasks can be obtained by finely decomposing and arranging the sub-execution agents under the condition that the current execution agent is difficult to meet the task requirements, so as to improve the execution efficiency and accuracy of the target tasks.

[0140] Figure 6 An application scenario diagram of the intelligent agent-based task execution method according to an embodiment of the present disclosure is schematically shown.

[0141] As shown in Figure 6 The scheduling agent can comprise a plurality of sub-scheduling agents, and the plurality of sub-scheduling agents can respectively be a dialogue understanding sub-agent, a demand understanding sub-agent, a logical arrangement sub-agent, and an interaction sub-agent. The scheduling agent can be arranged in a distributed server cluster, and the plurality of sub-scheduling agents can transmit information through the communication link between the distributed server cluster. The scheduling agent cooperates with the plurality of sub-scheduling agents to make decisions and control the task execution demand of the user.

[0142] Specifically, the user sends unstructured dialogue information to the scheduling agent through a smart terminal such as a smart phone. The dialogue understanding sub-agent performs task intention detection by processing the dialogue information. For example, operation S601 can be performed by the dialogue understanding sub-agent to determine whether the user is in a casual intention by processing the dialogue information. If the result of operation S601 is yes, it can be determined that the user's dialogue information can be general dialogue content or simple question and answer content, and there is no need to call an intelligent agent. The reply content can be generated by processing the dialogue information by using the interaction sub-agent, and the real-time dialogue demand of the user can be met by sending the reply content to the user.

[0143] If the result of operation S601 is no, it can be determined that the user needs to call an intelligent agent to perform a task to obtain a task execution result. The dialogue understanding sub-agent processes the demand information to perform task intention detection, and obtains structured task description information. The structured task description information is beneficial for the demand understanding sub-agent to perform in-depth analysis and accurate understanding by representing the task semantics and task execution logic of the task description information. The demand understanding sub-agent performs operation S602 to determine whether there is a matching target intelligent agent in the intelligent agent library. Specifically, the demand understanding sub-agent processes the task description information and the capability description information of the execution intelligent agent to determine whether the capability description information matches the task description information. If the result of operation S602 is yes, the demand understanding sub-agent can determine a target intelligent agent that matches the task description information from at least one execution intelligent agent. The target intelligent agent is suitable for executing a target task corresponding to the task description information. The target task can be obtained by updating the task attribute parameters of the task example of the target intelligent agent through the task description information, or can also be determined by processing the task description information by using the demand understanding sub-agent. Then, the demand understanding sub-agent performs operation S603 to control the target intelligent agent to execute the target task, and obtains the task execution result of the target task.

[0144] In a case where the result of the operation S602 is yes, a logic arrangement sub-agent is used to undertake the responsibility of dynamically constructing a plurality of sub-agents in cooperation. The logic arrangement sub-agent is used to process the task description information and the capability description information of the sub-execution agent, to determine a plurality of sub-target agents matching the task description information from the respective sub-execution agents of the plurality of execution agents. The logic arrangement sub-agent is used to process the task description information to logically arrange the plurality of sub-target agents, to obtain the dependency relationship among the plurality of sub-target agents, thereby constructing a target agent capable of meeting the new target task requirement of the current user, and realizing the construction of the cooperation process and mode of the plurality of sub-target agents. The logic arrangement sub-agent infers and filters out a plurality of sub-target agents suitable for executing the target task through the task requirement information and the target task, and constructs the cooperation mode of the plurality of sub-target agents for the target task through logical arrangement of the plurality of sub-target agents, thereby improving the scheduling flexibility and accuracy of the target agent. Therefore, the execution agent reuse and dynamic agent cooperation mode construction can be realized through the cooperation of the plurality of sub-scheduling agents based on the scheduling agent, thereby realizing efficient and robust task response capability in the case of task diversity, strong uncertainty of intelligent cooperation, and heterogeneous distribution of agent capability.

[0145] The interaction sub-agent converts one or more task execution results into task reply content understandable by the user and expressed in natural language based on the processing of the task execution result of the target task. For example, the interaction sub-agent can generate task reply content understandable by the user by summarizing and refining the task execution result.

[0146] Alternatively, the natural language reply content expressed based on the configured reply mode can also be generated based on the reply mode configured by the user. For example, the user can be fed back voice data expressing the task reply content based on the tone of the specified object. The interaction sub-agent can also actively feed back relevant task reply information to the user based on the context content of the multi-round dialogue.

[0147] In addition, the scheduling agent can also have a target task execution state tracking, a target task execution failure rollback mechanism, and the like, to improve the stability and robustness of the target task execution.

[0148] It should be understood that, Figure 6 In the embodiment shown, the first scheduling sub-agent is a logic arrangement sub-agent, and the second scheduling sub-agent is a requirement understanding sub-agent.

[0149] Figure 7 A block diagram of a capability description device of an agent according to an embodiment of the present disclosure is schematically shown.

[0150] As Figure 7As shown, the capability description apparatus 700 of the agent includes a first obtaining module 710, an example detection result obtaining module 720, and a capability description information obtaining module 730.

[0151] The first obtaining module 710 is configured to obtain a task example for a to-be-tested agent.

[0152] The example detection result obtaining module 720 is configured to detect, by using a master agent, an example execution result to obtain an example detection result, the example execution result being determined based on the to-be-tested agent performing the task example, and the example detection result representing a matching degree between the example execution result and a task requirement condition for the task example.

[0153] The capability description information obtaining module 730 is configured to perform, by using the master agent, a task execution capability description on the to-be-tested agent based on the example detection result and the task example, to obtain capability description information for the to-be-tested agent.

[0154] According to an embodiment of the present disclosure, the capability description information obtaining module 730 includes a first obtaining sub-module and a second obtaining sub-module.

[0155] The first obtaining sub-module is configured to update, by using a first sub-agent of the master agent, the task example based on the example detection result, to obtain an intermediate task example.

[0156] The second obtaining sub-module is configured to process, by using a second sub-agent of the master agent, the intermediate task example and an intermediate detection result to obtain the capability description information, wherein the intermediate detection result is determined based on an intermediate execution result, and the intermediate execution result is determined based on the to-be-tested agent performing the intermediate task example.

[0157] According to an embodiment of the present disclosure, the first obtaining sub-module includes an intermediate capability description information obtaining unit and an intermediate task example obtaining unit.

[0158] The intermediate capability description information obtaining unit is configured to update, by using the first sub-agent, initial capability description information for the to-be-tested agent based on the example detection result, to obtain intermediate capability description information.

[0159] The intermediate task example obtaining unit is configured to process, by using a third sub-agent of the master agent, the intermediate capability description information to obtain the intermediate task example.

[0160] According to an embodiment of the present disclosure, the task example includes a plurality of task examples, and the to-be-tested agent includes a plurality of sub-to-be-tested agents having a dependency relationship.

[0161] The intermediate task example obtaining unit includes a first obtaining sub-unit.

[0162] The first obtaining subunit is configured to obtain a first intermediate example by processing defect description information in the intermediate capability description information by using the third sub-agent, wherein the defect description information represents a defect factor of a first example execution result, and the first example execution result is determined by the sub-agent under test executing a first task example in the plurality of task examples.

[0163] According to an embodiment of the present disclosure, the intermediate capability description information obtaining unit comprises a second obtaining subunit.

[0164] The second obtaining subunit is configured to obtain defect description information for the sub-agent under test by processing defect detection results in the example detection results by using the first sub-agent, wherein the defect detection results represent that the first example execution result does not match the task requirement condition for the first task example.

[0165] According to an embodiment of the present disclosure, the intermediate task example obtaining unit comprises a third obtaining subunit.

[0166] The third obtaining subunit is configured to obtain a second intermediate example by processing completion description information in the intermediate capability description information by using a third sub-agent of the main agent, wherein the completion description information represents that a second example execution result matches a task requirement condition for a second task example, the task examples further comprise the second task example, and an intermediate task requirement condition for the second intermediate example is different from the task requirement condition for the second task example.

[0167] According to an embodiment of the present disclosure, the intermediate detection result is obtained by processing the intermediate execution result by using a fourth sub-agent of the main agent and an intermediate task requirement condition for the intermediate task example.

[0168] According to an embodiment of the present disclosure, the capability description information comprises at least one of the following: a dependency relationship between a plurality of sub-agents under test in the agent under test; a data attribute condition of input data and output data for the sub-agent under test; a defect example representing that a first example execution result obtained by the sub-agent under test executing a first task example in the task examples does not match a task requirement condition for the first task example; and a completion example representing that a second example execution result obtained by the sub-agent under test executing a second task example in the task examples matches a task requirement condition for the second task example.

[0169] According to an embodiment of the present disclosure, the example detection result obtaining module 720 comprises an example detection result obtaining sub-module.

[0170] The example detection result obtaining sub-module is configured to obtain a plurality of example detection results by processing a plurality of example execution results by using a fourth sub-agent of the main agent and a task requirement condition for each of the plurality of example execution results.

[0171] According to an embodiment of the present disclosure, the first obtaining module 710 comprises a task example obtaining sub-module.

[0172] The task example obtaining sub-module is configured to obtain a task example by processing the agent attribute of the to-be-tested agent by using the main agent.

[0173] According to an embodiment of the present disclosure, the task example obtaining sub-module comprises an initial capability description information obtaining unit and a task example obtaining unit.

[0174] The initial capability description information obtaining unit is configured to obtain initial capability description information related to the to-be-tested agent by processing the agent attribute of the to-be-tested agent by using a first sub-agent of the main agent.

[0175] The task example obtaining unit is configured to obtain the task example by processing the initial capability description information by using a third sub-agent of the main agent.

[0176] According to an embodiment of the present disclosure, the task example comprises at least one of a video generation task example, an image editing task example, and a copywriting generation task example.

[0177] Figure 8 A block diagram of an agent-based task execution apparatus according to an embodiment of the present disclosure is schematically shown.

[0178] As shown in Figure 8 The agent-based task execution apparatus 800 comprises a second obtaining module 810, a task description information obtaining module 820, a target agent determining module 830, and a task execution result obtaining module 840.

[0179] The second obtaining module 810 is configured to obtain task demand information.

[0180] The task description information obtaining module 820 is configured to perform task intent detection on the task demand information to obtain task description information.

[0181] The target agent determining module 830 is configured to determine a target agent from execution agents based on the task description information and capability description information of the execution agents, wherein the capability description information is determined according to an agent capability description method provided by an embodiment of the present disclosure.

[0182] The task execution result obtaining module 840 is configured to control the target agent to execute a target task related to the task demand information to obtain a task execution result.

[0183] According to an embodiment of the present disclosure, the execution agent comprises a plurality of sub-execution agents having a dependency relationship; and the target agent determining module 830 comprises:

[0184] The sub-target intelligent agent obtaining sub-module is configured to obtain, by using the first sub-scheduling intelligent agent, the sub-target intelligent agent from the plurality of sub-execution intelligent agents based on the task description information and the capability description information.

[0185] The first obtaining sub-module is configured to logically arrange, by using the first sub-scheduling intelligent agent, the plurality of sub-target intelligent agents to obtain the target intelligent agent.

[0186] According to an embodiment of the present disclosure, the plurality of execution intelligent agents includes a plurality of sub-target intelligent agents, and at least two of the plurality of sub-target intelligent agents are determined from different execution intelligent agents.

[0187] According to an embodiment of the present disclosure, the target intelligent agent determining module 830 includes a second obtaining sub-module.

[0188] The second obtaining sub-module is configured to determine, by using the second sub-scheduling intelligent agent, the target intelligent agent from the plurality of execution intelligent agents based on the capability description information and the task description information.

[0189] Figure 9 A structural block diagram of an intelligent agent of an artificial intelligence according to an embodiment of the present disclosure is schematically shown.

[0190] In an embodiment of the present disclosure, as shown in Figure 9 The AI intelligent agent 900 can include an input module 910, a processing module 920, and an output module 930.

[0191] The input module 910 is configured to receive input information.

[0192] The processing module 920 is configured to execute the capability description method of the intelligent agent or the task execution method based on the intelligent agent based on the input information received by the input module to obtain output information.

[0193] The output module 930 is configured to output the output information obtained by the processing module.

[0194] According to an embodiment of the present disclosure, the input module 910 is responsible for receiving or perceiving information such as queries, requests, instructions, signals, or data from the outside world (for example, a user or an external environment), and converting the information into a format that can be understood and processed by the AI intelligent agent 900. The input module 910 is the first link for the AI intelligent agent 900 to interact with the outside world, and it enables the AI intelligent agent 900 to efficiently and accurately obtain necessary "sensory" information from the outside world and respond to the information.

[0195] In an example, the input module 910 can input the task examples or requirement information described in the foregoing, etc.

[0196] In an example, the processing module 920 is the core support of the AI agent 900 to process the complex task capability. The processing module 920 can execute the capability description method or the task execution method of the agent described above.

[0197] In an example, the performance of the processing module 920 can be closely related to the large model based on which the AI agent 900 is based. In order to fully exert the capability of the large model, the internal structure of the processing module 920 can be designed to be highly configurable and scalable in order to cope with various different types of tasks and requirements in real scenarios.

[0198] In an example, after obtaining the demand voice, the processing module 920 can utilize the main agent to detect the example execution result to obtain an example detection result; based on the example detection result and the task example, the main agent is utilized to describe the task execution capability of the to-be-tested agent to obtain the capability description information for the to-be-tested agent, and the capability description information is transmitted to the output module 930.

[0199] It can be understood that although the large language model has excellent language understanding and generation capability, it, like a person, can solve a limited number of tasks without the aid of any tool. When the AI agent 900 is endowed with the capability of tool invocation, it can implement tasks such as completing mathematical operations with the aid of a calculator, completing data analysis with the aid of python, and completing weather forecasting with the aid of a search engine.

[0200] In an example, the output module 930 can output the capability description information or the task execution result described above.

[0201] The AI agent 900 according to the embodiments of the present disclosure can simply and effectively improve the intelligent degree and improve the flexibility and versatility.

[0202] According to the embodiments of the present disclosure, the present disclosure further provides an electronic device, a readable storage medium, and a computer program product.

[0203] According to the embodiments of the present disclosure, an electronic device comprises at least one processor and a memory connected with the at least one processor in communication; the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method described above.

[0204] According to the embodiments of the present disclosure, a non-transitory computer readable storage medium stores computer instructions, wherein the computer instructions are used to enable a computer to execute the method described above.

[0205] According to an embodiment of this disclosure, a computer program product includes a computer program that, when executed by a processor, implements the method described above.

[0206] Figure 10 A schematic block diagram of an electronic device 1000, illustrating a capability description method for an intelligent agent and an example of an agent-based task execution method that can be used to implement embodiments of the present disclosure, is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0207] like Figure 10 As shown, device 1000 includes a computing unit 1001, which can perform various appropriate actions and processes according to a computer program stored in read-only memory (ROM) 1002 or a computer program loaded into random access memory (RAM) 1003 from storage unit 1008. The RAM 1003 may also store various programs and data required for the operation of device 1000. The computing unit 1001, ROM 1002, and RAM 1003 are interconnected via bus 1004. Input / output (I / O) interface 1005 is also connected to bus 1004.

[0208] Multiple components in device 1000 are connected to I / O interface 1005, including: input unit 1006, such as keyboard, mouse, etc.; output unit 1007, such as various types of monitors, speakers, etc.; storage unit 1008, such as disk, optical disk, etc.; and communication unit 1009, such as network card, modem, wireless transceiver, etc. Communication unit 1009 allows device 1000 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0209] The computing unit 1001 can be various general and / or special purpose processing components having processing and computing capabilities. Some examples of the computing unit 1001 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, and the like. The computing unit 1001 performs various methods and processes described above, such as the agent capability description method or the agent-based task execution method. For example, in some embodiments, the agent capability description method or the agent-based task execution method can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 1008. In some embodiments, part or all of the computer program can be loaded and / or installed onto the device 1000 via the ROM 1002 and / or the communication unit 1009. When the computer program is loaded onto the RAM 1003 and executed by the computing unit 1001, one or more steps of the agent capability description method or the agent-based task execution method described above can be performed. Alternatively, in other embodiments, the computing unit 1001 can be configured to perform the agent capability description method or the agent-based task execution method by any other suitable means, such as by means of firmware.

[0210] Various implementations of the systems and techniques described above can be realized in digital electronic circuitry, integrated circuitry, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), a system on a chip (SOC), a programmable logic device (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include implementation in one or more computer programs that are executable and / or interpretable on a programmable system including at least one programmable processor, which can be special or general purpose, coupled to receive data and instructions from, and to transmit data and instructions to, a storage system, at least one input device, and at least one output device.

[0211] Program code for carrying out methods of the present disclosure can be written in any combination of one or more programming languages. The program code can be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the program code, when executed by the processor or controller, produces the functions / operations specified in the flowcharts and / or block diagrams. The program code can be executed entirely on a machine, partially on a machine, partially on a machine as a standalone software package, or entirely on a remote machine or server.

[0212] In the context of the present disclosure, a machine-readable medium can be a tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include but is not limited to an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the machine-readable storage medium will include one or more lines of electrical connections, portable computer disks, hard disk drives, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or Flash memory), optical fibers, portable compact disc read-only memories (CD-ROMs), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0213] To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can be used to provide for interaction with a user as well; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form, including acoustic, speech, or tactile input.

[0214] The systems and techniques described here can be implemented in a computing system that includes a back end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front end component (e.g., a user computer having a graphical user interface or a Web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back end, middleware, or front end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0215] The computer system can include clients and servers. This relationship can be. The servers are typically remote from the clients with the interactions between them occurring over a communication network. The relationship between a client and a server is one of client-server. The server can be a cloud server, a server of a distributed system, or a server incorporating a blockchain.

[0216] It should be understood that the various forms of flow shown above can be re-ordered, added to, or deleted from without departing from the scope of the present disclosure. For example, the steps recited in the present disclosure can be executed in parallel, executed in series, or executed in different orders, as long as the desired results of the technical solutions of the present disclosure are achieved, and the present disclosure is not limited herein.

[0217] The specific implementation described above does not constitute a limitation on the protection scope of the present disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent replacements, and improvements within the spirit and principles of the present disclosure should be included in the protection scope of the present disclosure.

Claims

1. A method for describing capability of an agent, comprising: obtaining a task example for a to-be-tested agent; detecting, by a master agent, an example execution result to obtain an example detection result, the example execution result being determined based on the to-be-tested agent executing the task example, and the example detection result representing a matching degree between the example execution result and a task requirement condition for the task example; updating, by a first sub-agent of the master agent, initial capability description information for the to-be-tested agent based on the example detection result to obtain intermediate capability description information; processing, by a third sub-agent of the master agent, the intermediate capability description information to obtain an intermediate task example; processing, by a second sub-agent of the master agent, the intermediate task example and an intermediate detection result to obtain capability description information, the intermediate detection result being determined based on an intermediate execution result, the to-be-tested agent executing the intermediate task example to determine the intermediate execution result.

2. The method of claim 1, wherein, The task example comprises a plurality of task examples, and the to-be-tested agent comprises a plurality of sub-to-be-tested agents having a dependency relationship. The processing, by the third sub-agent of the master agent, of the intermediate capability description information to obtain the intermediate task example comprises: processing, by the third sub-agent, defect description information in the intermediate capability description information to obtain a first intermediate example, wherein the defect description information represents a defect factor of a first example execution result, and the first example execution result is determined by the sub-to-be-tested agent executing a first task example in the plurality of task examples.

3. The method of claim 2, wherein, The updating, by the first sub-agent of the master agent, of the initial capability description information for the to-be-tested agent based on the example detection result to obtain the intermediate capability description information comprises: processing, by the first sub-agent, defect detection result in the example detection result to obtain defect description information for the sub-to-be-tested agent, wherein the defect detection result represents that the first example execution result does not match a task requirement condition for the first task example.

4. The method of claim 1, wherein, The processing, by the third sub-agent of the master agent, of the intermediate capability description information to obtain the intermediate task example comprises: processing, by the third sub-agent, completion description information in the intermediate capability description information to obtain a second intermediate example, wherein the completion description information represents that a second example execution result matches a task requirement condition for a second task example, the task example further comprising the second task example, and an intermediate task requirement condition for the second intermediate example being different from the task requirement condition for the second task example.

5. The method of claim 1, wherein, The intermediate detection result is obtained by processing, by a fourth sub-agent of the master agent, the intermediate execution result and an intermediate task requirement condition for the intermediate task example.

6. The method of claim 1, wherein, The capability description information comprises at least one of: a dependency relationship between the plurality of sub-to-be-tested agents in the to-be-tested agent; data attribute conditions of input data and output data for the sub-to-be-tested agent; and a data attribute condition of input data and output data for the sub-to-be-tested agent. a defect example, representing a first example execution result obtained by the sub-agent-under-test executing a first task example in the task examples, does not match a task requirement condition for the first task example; a completion example, representing a second example execution result obtained by the sub-agent-under-test executing a second task example in the task examples, matches a task requirement condition for the second task example.

7. The method of any one of claims 1 to 6, wherein, The detecting of the example execution result by the main agent includes: processing, by a fourth sub-agent of the main agent, a plurality of the example execution results and a task requirement condition for each of the plurality of the example execution results, to obtain a plurality of the example detection results.

8. The method of claim 1, wherein, The obtaining of the task examples for the agent-under-test includes: processing, by the main agent, an agent attribute for the agent-under-test, to obtain the task examples.

9. The method of claim 8, wherein, The processing of the agent attribute for the agent-under-test by the main agent to obtain the task examples includes: processing, by a first sub-agent of the main agent, the agent attribute for the agent-under-test, to obtain initial capability description information related to the agent-under-test; processing, by a third sub-agent of the main agent, the initial capability description information, to obtain the task examples.

10. The method of any one of claims 1 to 6, wherein, The task examples include at least one of: a video generation task example, an image editing task example, and a script generation task example.

11. An agent-based task execution method, comprising: obtaining task requirement information; performing task intent detection on the task requirement information to obtain task description information; determining a target agent from a plurality of execution agents based on the task description information and capability description information for the execution agents, wherein the capability description information is determined according to the method of any one of claims 1 to 10; and controlling the target agent to execute a target task related to the task requirement information to obtain a task execution result. The execution agents include a plurality of sub-execution agents having a dependency relationship; and the determining of the target agent from the execution agents includes:

12. The method of claim 11, wherein, processing, by a first sub-scheduling agent, the task description information and capability description information for the sub-execution agents to obtain a sub-target agent in the plurality of the sub-execution agents, the sub-target agent being used to execute an associated target task in the plurality of the target tasks; and logically arranging, by the first sub-scheduling agent, the plurality of the sub-target agents to obtain the target agent. The execution agents include a plurality of, and at least two of the plurality of the sub-target agents are determined from different execution agents.

13. The method of claim 12, wherein, The determining of the target agent from the execution agents includes:

14. The method of any one of claims 11 to 13, wherein, determining, by a second sub-scheduling agent, the target agent from the plurality of the execution agents based on the capability description information and the task description information.

15. An agent capability description apparatus, comprising: a first obtaining module, configured to obtain task examples for an agent-under-test; ​ An example detection result obtaining module is configured to detect an example execution result by using a master agent to obtain an example detection result, the example execution result being determined based on the agent under test executing the task example, and the example detection result representing a matching degree between the example execution result and a task requirement condition for the task example; An ability description information obtaining module is configured to perform task execution ability description on the agent under test by using the master agent based on the example detection result and the task example to obtain ability description information for the agent under test; The ability description information obtaining module comprises: An intermediate ability description information obtaining unit is configured to update initial ability description information for the agent under test based on the example detection result by using a first sub-agent to obtain intermediate ability description information; An intermediate task example obtaining unit is configured to process the intermediate ability description information by using a third sub-agent of the master agent to obtain an intermediate task example; A second obtaining sub-module is configured to process the intermediate task example and intermediate detection result by using a second sub-agent of the master agent to obtain the ability description information, the intermediate detection result being determined based on intermediate execution result, the agent under test executing the intermediate task example determining the intermediate execution result.

16. An agent-based task execution apparatus, comprising: A second obtaining module is configured to obtain task requirement information; A task description information obtaining module is configured to perform task intention detection on the task requirement information to obtain task description information; A target agent determining module is configured to determine a target agent from the execution agent based on the task description information and ability description information for a preset execution agent, the ability description information being determined according to the method in any one of claims 1 to 10; and A task execution result obtaining module is configured to control the target agent to execute a target task related to the task requirement information to obtain a task execution result.

17. An artificial intelligence agent configured to perform the method in any one of claims 1 to 14.

18. An electronic device, comprising: at least one processor; and a memory connected to the at least one processor in communication; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method in any one of claims 1 to 14. The computer instructions are used to enable the computer to perform the method in any one of claims 1 to 14.

19. A non-transitory computer readable storage medium having stored thereon computer instructions, wherein, 20. A computer program product comprising a computer program which, when executed by a processor, implements the method in any one of claims 1 to 14.

20. A computer program product comprising a computer program which, when executed by a processor, implements the method in any one of claims 1 to 14.

Images