Method and system for executing user request by AI agent

Abstract

The invention discloses a method and a system for executing a user request by an AI agent. The method comprises: in response to a received user request, determining a plurality of target tasks, an execution sequence among the plurality of target tasks, and a task executor corresponding to an execution field in which each of the plurality of target tasks is located, the task executor comprising at least one of a dedicated AI agent or an AI expert system; based on the execution sequence, calling each corresponding task executor in the plurality of target tasks, so that each corresponding task executor in the plurality of target tasks executes the corresponding target task and outputs a corresponding execution result; and integrating all the execution results to generate a final response result.

Description

Methods and systems for AI agents to execute user requests Technical Field

[0001] This invention relates to the field of computers, and more particularly to a method and system for an AI agent to execute user requests. Background Technology

[0002] With the deepening development of artificial intelligence technology, specialized AI tools are constantly emerging. They are deeply integrated into various vertical fields, such as legal text parsing, financial data analysis, creative content generation, and business process automation, driving productivity towards a more refined and professional direction. Summary of the Invention

[0003] According to an embodiment of the present invention, a method for executing a user request by an AI agent includes: in response to a received user request, determining a plurality of target tasks, an execution order among the plurality of target tasks, and a task executor corresponding to the execution domain of each of the plurality of target tasks, wherein the task executor includes at least one of a dedicated AI agent or an AI expert system; based on the execution order, invoking the corresponding task executor of each of the plurality of target tasks so that each corresponding task executor of the plurality of target tasks executes the corresponding target task and outputs a corresponding execution result; and integrating all execution results to generate a final response result.

[0004] A system for executing user requests by an AI agent according to an embodiment of the present invention includes: a task decomposition module configured to, in response to a received user request, determine multiple target tasks, the execution order among the multiple target tasks, and a task executor corresponding to the execution domain of each of the multiple target tasks, wherein the task executor includes at least one of a dedicated AI agent or an AI expert system; a task execution module configured to, based on the execution order, invoke the corresponding task executor of each of the multiple target tasks, so that each corresponding task executor of the multiple target tasks executes the corresponding target task and outputs a corresponding execution result; and a result integration module configured to integrate all execution results to generate a final response result.

[0005] According to an embodiment of the present invention, a computer-readable storage medium stores computer-executable instructions thereon, wherein, when executed by a processor, these computer-executable instructions cause the processor to perform the aforementioned method of the AI ​​agent executing a user request.

[0006] A computer program product according to an embodiment of the present invention includes computer-executable instructions, wherein, when executed by a processor, these computer-executable instructions cause the processor to perform the aforementioned method of the AI ​​agent executing a user request. Attached Figure Description

[0007] The present invention can be better understood from the following description of specific embodiments in conjunction with the accompanying drawings, wherein: Figure 1 shows a schematic flowchart of a method for an AI agent to execute a user request according to an embodiment of the present invention.

[0008] Figure 2 shows a schematic diagram of the relationship between multiple devices according to an embodiment of the present invention.

[0009] Figure 3 shows an example block diagram of a resource hierarchy model according to an embodiment of the present invention.

[0010] Figure 4 shows an example graph of loss monitoring according to an embodiment of the present invention.

[0011] Figure 5 shows a schematic block diagram of a system in which an AI agent executes a user request according to an embodiment of the present invention.

[0012] Figure 6 shows a schematic diagram of a computer system that can implement the method and apparatus for an AI agent to execute user requests according to embodiments of the present invention. Detailed Implementation

[0013] The features and exemplary embodiments of various aspects of the present invention will now be described in detail. Numerous specific details are set forth in the following detailed description to provide a thorough understanding of the invention. However, it will be apparent to those skilled in the art that the invention may be practiced without requiring some of these specific details. The following description of embodiments is merely intended to provide a better understanding of the invention by illustrating examples of the invention. The invention is by no means limited to any specific configurations and algorithms presented below, but covers any modifications, substitutions, and improvements to elements, components, and algorithms without departing from the spirit of the invention. Well-known structures and techniques are not shown in the drawings and the following description in order to avoid unnecessarily obscuring the invention.

[0014] With the deepening development of artificial intelligence technology, specialized AI tools are constantly emerging. These tools are deeply integrated into various vertical fields, such as legal text parsing, financial data analysis, creative content generation, and business process automation, driving productivity towards greater precision and specialization. However, for complex user requests, single specialized AI tools struggle to provide ideal user responses, resulting in low processing efficiency.

[0015] According to embodiments of the present invention, user requests are determined by analyzing user request text, multiple target tasks are determined based on user requests, and the execution effect of each target task can be improved by calling professional task executors in the execution domain of each target task. Calling different task executors in the execution order can improve the overall execution efficiency. Finally, the final response result obtained by integrating all execution results is unified, coherent and complete, which can quickly and efficiently meet user requests with complex content and strong comprehensiveness.

[0016] Figure 1 shows a schematic flowchart of a method for executing a user request by an AI agent according to an embodiment of the present invention. As shown in Figure 1, the method 100 for executing a user request by an AI agent includes S101-S103. S101: In response to a received user request, determine multiple target tasks, the execution order among the multiple target tasks, and the task executor corresponding to the execution domain of each of the multiple target tasks. The task executor includes at least one of a dedicated AI agent or an AI expert system. S102: Based on the execution order, invoke the corresponding task executor of each of the multiple target tasks so that each corresponding task executor of the multiple target tasks executes the corresponding target task and outputs the corresponding execution result. S103: Integrate all execution results to generate a final response result.

[0017] The method according to embodiments of the present invention is implemented by an AI agent. The AI ​​agent receives the text of a user request sent by the user. Here, the user request text sent directly by the user may have problems such as complex content with implicit information, vague and imprecise expression, and format that cannot be directly read. By performing semantic analysis and reorganization on the original user request text, a more insightful, clearer description, and machine-readable user request can be obtained. In some embodiments, the reorganized user request can also be displayed to the user to obtain the user's feedback on the user request. For example, if the user agrees with the user request, the next step is performed; if the user disagrees with the user request, the user request is adjusted again based on the user's feedback.

[0018] In the method according to embodiments of the present invention, a user request can be decomposed into multiple target tasks with different execution stages and different execution domains. The execution domain of each target task corresponds to the execution entity that performs that target task, i.e., the task executor. The execution stages of each target task are different, and for all target tasks, the complete execution order can be determined based on the execution stages of each target task. In the execution order, if two target tasks have a necessary sequential order, for example, if one target task, task_11, must be executed on the premise that another target task, task_10, is completed, then these two target tasks are dependent. In the execution order, two parallel target tasks, task_1 and task_2, have no association or dependency; they are executed asynchronously, resulting in low latency.

[0019] In the method according to embodiments of the present invention, the task executor can be a specialized AI agent or an AI specialist system. Both types of task executors can be deployed and updated according to actual needs. The AI ​​specialist system, with fixed and modular functions, solves specific problems in a defined domain. Domains that can be served by AI specialist systems include, but are not limited to: business and strategy, legal and compliance, financial analysis, market research, blockchain, program coding, program verification, visual and icon generation, enterprise data processing, and algorithm / quantitative analysis. The specialized AI agent, on the other hand, is a virtual role within an enterprise, personified in its role, assuming specific functions (e.g., CEO, CFO, COO, manager). The specialized AI agent integrates multiple capabilities around its role objectives, including analysis, decision-making, and coordination, and can flexibly utilize various resources to solve relatively broad and comprehensive problems (e.g., enterprise financial analysis, human resource allocation, strategic planning). Compared to AI specialist systems, specialized AI agents offer greater flexibility and autonomy.

[0020] In the method according to embodiments of the present invention, the corresponding task executors are invoked according to the execution order of each target task. The invoked task executors can execute the received target tasks and output the corresponding execution results. Executing the target tasks can include completion by the task executor itself, or completion through interaction between the task executor and a task object. Here, the task object mainly refers to an external application system of the task executor, such as a financial system, a messaging platform, a Human Resource Management System (HRMS), or an Enterprise Resource Planning (ERP) system within an enterprise integration platform. The task executor can interact with the application system through an intermediate interface.

[0021] Figure 2 illustrates a schematic diagram of the relationship between multiple devices according to an embodiment of the present invention. As shown in Figure 2, during the execution of the method of the embodiment of the present invention, the AI ​​agent 200 invokes one or more task executors 210. A task executor 210 can execute a target task on its own, or it can operate on the corresponding application system 220 based on the corresponding target task.

[0022] In the method according to embodiments of the present invention, the execution results output by each task executor are typically output in a machine-readable structured format. Generally, the final response result generated by integrating all execution results not only includes the content of all execution results, but also has more coherent content and a more easily readable format, facilitating user understanding.

[0023] The execution results output by each task executor may contain logical contradictions. For example, the execution result of one target task might include setting a specific parameter to 1, while the execution result of another target task might include setting that same parameter to 2. In this case, the two execution tasks are logically contradictory. Such contradictions need to be checked before integrating all execution results. In some embodiments, this method may further include: analyzing whether there are logical contradictions among all execution results before integrating them; if so, adjusting the logically contradictory execution results based on contradiction elimination rules. The checking methods may include checking for logical contradictions based on text semantic similarity and conflict pattern matching; mapping the execution results to a knowledge graph and then performing logical verification and contradiction checking; converting the execution results into logical expressions and then performing contradiction checking; or using a dedicated contradiction detection model for contradiction checking. Contradiction elimination rules may include confidence weight priority, domain-level priority (e.g., legal-level priority is higher than business-level priority), preset governance rules, and escalation mechanisms, etc. Adjusting an execution result that has a logical contradiction can be done by modifying the content of another execution result that has a logical contradiction with the first execution result, based on the execution result with higher priority. Alternatively, the task executor of the other execution result that has a logical contradiction with the first execution result can execute a new target task to output a new execution result that does not have a logical contradiction with the first execution result. The new target task can be the target task updated from the original target task based on the execution result with higher priority.

[0024] In some embodiments, the method may further include: after invoking each corresponding task executor among multiple target tasks, performing a compliance check on each execution result, and adjusting each execution result based on the compliance check results. A compliance check refers to a technical supervision process that systematically verifies execution results using preset rules and standards to ensure they comply with internal policies, industry norms, and laws and regulations. Adjusting the execution results may involve directly modifying execution results that fail the compliance check, or updating the target tasks based on the failed compliance check results, invoking the corresponding task executors to execute the updated target tasks, and outputting new execution results. Additionally, adjusting each execution result may also include: injecting warnings or disclaimers into the execution results.

[0025] For example, in one scenario, the user request is "Create a new product strategy and check compliance." Based on this request, several target tasks are identified: business strategy, legal compliance, and visual charts. After the corresponding task executor is invoked to perform the target tasks, a compliance check is performed on the execution results. The compliance check result indicates that a potential regulatory risk has been detected, and the adjustment measure is to inject a risk warning. The final response is a single report containing the risk warning and accompanying diagrams.

[0026] For example, in another instance, a user request might be "analyze the financial risks of new blockchain tokens." Based on this request, several target tasks are identified: financial analysis, blockchain verification, and algorithmic risk assessment. After executing these tasks and obtaining the results, a compliance check is performed. If the compliance check reveals that high-risk projects have been detected, the corresponding adjustment measure is to flag these high-risk projects. The analysis and recommendations are then combined into a single report as the final response output.

[0027] In some embodiments, while executing a target task and outputting the execution result, the AI ​​expert system can also output the confidence level of that result. At this point, the relationship between this confidence level and a confidence threshold can be used to determine whether to directly use the execution result or to replace the model and re-execute the target task to try and obtain a result with higher confidence. For example, the three AI expert systems Spec.1-3 in Table 1 below correspond to three execution domains: legal, financial, and blockchain. Each AI expert system includes a main AI model and a backup AI model. Table 1 also records the current version and confidence threshold of each AI expert system.

[0028] Table 1. Examples of information about AI expert systems In some embodiments, when an AI expert system is invoked, it may perform the following operations: execute the corresponding target task based on the main AI model and generate the corresponding main execution result and the confidence level of the main execution result; when the confidence level of the main execution result is not lower than the confidence level threshold, use the main execution result as the execution result of the AI ​​expert system and output it; when the confidence level of the main execution result is lower than the confidence level threshold, execute the target task again based on the backup AI model and generate the corresponding backup execution result and the confidence level of the backup execution result; use the one with the higher confidence level between the main execution result and the backup execution result as the execution result of the AI ​​expert system and output it.

[0029] In some embodiments, the method may further include: while invoking each corresponding task executor among multiple target tasks, allocating resources to the task executors based on the principle of minimum redundancy. Figure 3 shows an example block diagram of a hierarchical model of resources according to an embodiment of the present invention. As shown in Figure 3, the resources supporting the method of the embodiments of the present invention can be a hierarchical model, wherein the first layer is cache memory, the second layer is local storage, the third layer is a distributed cluster, and the fourth layer is an archived knowledge base. The resources of each layer can be allocated to the corresponding task executors according to the needs of different target tasks. The allocation objective is to minimize redundant computation, and the specific allocation form is task routing, similar to the means of recording the task executors corresponding to the target tasks.

[0030] In the method according to embodiments of the present invention, multiple target tasks determined based on user requests may have similar situations. To improve execution efficiency, previous contexts can be constructed by combining historical cached data, and incremental execution can be performed based on these previous contexts. This achieves partial reuse and adaptive optimization, reducing computational overhead and increasing efficiency. After each target task is executed, relevant data can be organized and archived as historical cached data, which can be accurately retrieved later by searching the global orchestration index. In some embodiments, determining multiple target tasks based on user requests may include: determining multiple tasks to be executed based on user requests; and updating the multiple tasks to be executed to multiple target tasks that require incremental execution based on historical cached knowledge. The historical cached knowledge can be stored in the aforementioned archived knowledge base. For example, when the task to be executed is the second quarter financial forecast, the cached first quarter model can be reused, and the task to be executed can be updated to: incrementally integrating new sales data from the second quarter based on the first quarter model. This phased incremental execution method preserves the previous context, avoids redundant recalculation, and generates an updated second quarter forecast with lower computational overhead.

[0031] In some embodiments, the method may further include: during the process of invoking each corresponding task executor among multiple target tasks based on the execution order, monitoring the cumulative loss of inaction and the risk of action loss for each of the multiple target tasks; and when the cumulative loss of inaction exceeds the risk of action loss, initiating a preset operation to reduce the cumulative loss of inaction. Here, the cumulative loss of inaction for each target task refers to the loss or cost that accumulates over time due to delayed decision-making or delayed action, and typically accelerates; the risk of action loss refers to the negative consequences that may arise from taking a specific action, the severity of which is within a predictable and acceptable limited range.

[0032] Figure 4 shows an example curve for loss monitoring according to an embodiment of the present invention. As shown in Figure 4, without taking any action, the curve L1, representing the cumulative loss due to inaction, will increase non-linearly over time. This is because delays exacerbate the negative impacts of inefficiency, missed opportunities, and may lead to compliance penalties. Examples of situations where no action is taken include: failure to reallocate budgets, failure to approve purchases, failure to rebalance the investment portfolio, failure to respond to compliance requests, and failure to execute marketing campaigns. In an enterprise system, this will manifest as delays, missed opportunities, operational inefficiency, and financial leaks, respectively. As shown in Figure 4, this method can also initiate preset operations when the detected cumulative loss due to inaction exceeds the trigger threshold L0.

[0033] As shown in Figure 4, the curve L2, representing the risk loss of an action, always remains within a finite range. This is because the negative impact of any single action is limited. Examples of negative impacts include: potential forecasting errors, inappropriate budget allocation, potential procurement overruns, and suboptimal personnel deployment. This type of action risk loss typically does not increase over time; it is often bounded because the action itself is bounded.

[0034] In some embodiments, the process of monitoring the cumulative loss of inaction may include: monitoring the cumulative loss of inaction based on at least one of an opportunity cost model, a nonlinear growth function, or a time-weighted penalty function. Specifically, for the opportunity cost model, its input parameters include: the expected benefit of timely action, the alternatives abandoned, and relevant market variables. For the nonlinear growth function, its input parameters include: the duration of the delay, the task sensitivity curve, and the growth parameter. For the time-weighted penalty function, its input parameters include: the task deadline, the time elapsed, and the penalty weight.

[0035] In some embodiments, the preset action may include at least one of: initiating a compliance check, reallocating resources, and issuing an alert. Issuing an alert may include issuing an alarm or notification, or escalating the task to a human supervisor.

[0036] In some embodiments, this method may also include immutable data logging during the execution of a user request for subsequent auditing or accountability. It can also be further configured to manage tasks across multiple enterprises in a multi-tenant environment, with task fingerprints and resource allocation isolated by enterprise.

[0037] Specifically, data recording may involve generating task fingerprints, including applying at least one of semantic hashing, Trie structures, Bloom filters, vector embeddings, or graph-based indexing to support variable-length and semantic matching of enterprise tasks. Another example is cryptographically binding task fingerprints to organizational identifiers to prevent task spoofing or misassignment. Furthermore, metadata about user requests or target tasks may also be recorded. Metadata for a user request can include user ID (user_id), timestamp, request ID (request_id), sensitivity level (sensitivity_level), and governance profile (governance_profile), as shown in the following example: {"user_id": "U123","timestamp": "2025-12-29T10:15:00Z","request_id": "REQ-987","sensitivity_level": "medium","governance_profile": "default"}. Similarly, metadata for a target task can include task ID (task_id), execution domain, dependencies (i.e., associated task IDs) determined by the execution order, expected output format (expected_output_format), and priority, as shown in the following example: {"task_id": "T001","domain": "legal","dependencies": ["finance_T002"],"expected_output_format": "json","priority": Correspondingly, each execution result can be returned in a unified structured JSON format. Specifically, each execution result can include a summary, risks, confidence, and metadata, as shown in the following example: {"summary": "...","risks": ["..."],"confidence": "high|medium|low","metadata": {...}} Figure 5 shows a schematic block diagram of a system for executing user requests by an AI agent according to an embodiment of the present invention. As shown in Figure 5, the system 500 for executing user requests by an AI agent includes a task decomposition module 501, a task execution module 502, and a result integration module 503.The task decomposition module 501 is configured to, in response to a received user request, determine multiple target tasks, the execution order among the target tasks, and the task executor corresponding to the execution domain of each of the target tasks. The task executor includes at least one from a dedicated AI agent or an AI expert system. The task execution module 502 is configured to, based on the execution order, invoke the corresponding task executor for each of the target tasks, so that each corresponding task executor executes its corresponding target task and outputs the corresponding execution result. The result integration module 503 is configured to integrate all execution results to generate a final response result.

[0038] In some embodiments, the result integration module 503 may be further configured to: analyze whether there are logical contradictions among all execution results before integrating all execution results; if so, adjust the execution results with logical contradictions based on the contradiction elimination rules.

[0039] In some embodiments, the task execution module 502 may be further configured to: perform a compliance check on each execution result after invoking each corresponding task executor among multiple target tasks, and adjust each execution result according to the compliance check result.

[0040] In some embodiments, when an AI expert system is invoked, it may perform the following operations: execute the corresponding target task based on the main AI model and generate the corresponding main execution result and the confidence level of the main execution result; when the confidence level of the main execution result is not lower than the confidence level threshold, use the main execution result as the execution result of the AI ​​expert system and output it; when the confidence level of the main execution result is lower than the confidence level threshold, execute the target task again based on the backup AI model and generate the corresponding backup execution result and the confidence level of the backup execution result; use the one with the higher confidence level between the main execution result and the backup execution result as the execution result of the AI ​​expert system and output it.

[0041] In some embodiments, the task execution module 502 may be further configured to: monitor the cumulative loss of inaction and the risk loss of action for each of the multiple target tasks during the process of calling each corresponding task executor in the multiple target tasks based on the execution order; and initiate a preset operation to reduce the cumulative loss of inaction when the cumulative loss of inaction is greater than the risk loss of action.

[0042] In some embodiments, the process of monitoring non-cumulative losses may include: monitoring non-cumulative losses based on at least one of an opportunity cost model, a nonlinear growth function, or a time-weighted penalty function.

[0043] In some embodiments, the preset operation may include at least one of: initiating a compliance check, reallocating resources, and issuing an alarm notification.

[0044] In some embodiments, the task decomposition module 501 may be further configured to: determine multiple tasks to be executed based on user requests; and update the multiple tasks to be executed into multiple target tasks that need to be incrementally executed based on historical cache knowledge.

[0045] In some embodiments, the task execution module 502 may be further configured to allocate resources to the task executor based on the principle of minimum redundancy while invoking each corresponding task executor among multiple target tasks.

[0046] In some embodiments, any task executor performing a corresponding target task may include: the task executor operating on the corresponding application system based on the corresponding target task.

[0047] Figure 6 illustrates a schematic diagram of a computer system that can implement the method and apparatus for executing user requests by an AI agent according to embodiments of the present invention. It should be understood that the computer system 600 shown in Figure 6 is merely an example and should not impose any limitation on the functionality and scope of use of the method and apparatus for executing user requests by an AI agent according to embodiments of the present invention.

[0048] As shown in Figure 6, the computer system 600 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage device 608 into a random access memory (RAM) 603. The RAM 603 also stores various programs and data required for the operation of the computer system 600. The processing device 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0049] Typically, the following devices can be connected to I / O interface 605: input devices 606 including, for example, touchscreens, touchpads, cameras, accelerometers, gyroscopes, sensors, etc.; output devices 607 including, for example, liquid crystal displays (LCDs), speakers, vibrators, motors, electronic speed controllers, etc.; storage devices 608 including, for example, flash cards; and communication devices 609. Communication device 609 allows computer system 600 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 6 illustrates a computer system 600 with various devices, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively. Each box shown in Figure 6 may represent one device, or multiple devices may be represented as needed.

[0050] In particular, according to some embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer programs. For example, a computer-readable medium is provided having a computer program stored thereon, the computer program containing program code for performing the method shown in FIG1 for executing a user request by an AI agent. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 609, or installed from a storage device 608, or installed from a ROM 602. When the computer program is executed by the processing device 601, the functional units defined in the apparatus for executing a user request by an AI agent according to embodiments of the present invention are implemented.

[0051] It should be noted that the computer-readable medium according to embodiments of the present invention may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. A computer-readable storage medium according to embodiments of the present invention may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. Additionally, a computer-readable signal medium according to embodiments of the present invention may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (Radio Frequency), etc., or any suitable combination thereof.

[0052] Computer program code for performing operations according to embodiments of the present invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0053] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0054] This invention can be implemented in other specific forms without departing from its spirit and essential characteristics. For example, the algorithm described in a particular embodiment can be modified without departing from the basic spirit of the invention. Therefore, the present embodiments are to be regarded as exemplary rather than limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and all changes falling within the meaning and scope of the claims and their equivalents are thus included within the scope of the invention.

Claims

1. A method for an AI agent to execute a user request, comprising: In response to a received user request, determine multiple target tasks, the execution order of the multiple target tasks, and the task executor corresponding to the execution domain of each of the multiple target tasks, wherein the task executor includes at least one of a dedicated AI agent or an AI expert system; Based on the execution order, each corresponding task executor among the plurality of target tasks is invoked, so that each corresponding task executor among the plurality of target tasks executes the corresponding target task and outputs the corresponding execution result; And integrate all the execution results to generate the final response result.

2. The method according to claim 1, wherein, It also includes: before integrating all the execution results, analyzing whether there are logical contradictions among all the execution results; if so, adjusting the execution results with logical contradictions based on the contradiction elimination rules.

3. The method according to claim 1, wherein, It also includes: after invoking each corresponding task executor among the plurality of target tasks, performing a compliance check on each of the execution results, and adjusting each of the execution results based on the compliance check results.

4. The method according to claim 1, wherein, When the AI ​​expert system is invoked, it performs the following operations: executes the corresponding target task based on the main AI model and generates the corresponding main execution result and the confidence level of the main execution result; If the confidence level of the main execution result is not lower than the confidence level threshold, the main execution result is used as the execution result of the AI ​​expert system and output. When the confidence level of the main execution result is lower than the confidence level threshold, the target task is executed again based on the backup AI model to generate a corresponding backup execution result and the confidence level of the backup execution result. The one with the higher confidence level between the main execution result and the backup execution result is taken as the execution result of the AI ​​expert system and output.

5. The method according to claim 1, wherein, Also includes: During the process of calling each corresponding task executor among the plurality of target tasks based on the execution order, the cumulative loss of inaction and the risk of action loss of each of the plurality of target tasks are monitored. When the cumulative loss of inaction is greater than the risk of action loss, a preset operation is initiated to reduce the cumulative loss of inaction.

6. The method according to claim 5, wherein, The process of monitoring the cumulative loss of the inaction includes: monitoring the cumulative loss of the inaction based on at least one of the opportunity cost model, the nonlinear growth function, or the time-weighted penalty function.

7. The method according to claim 5, wherein, The preset operations include at least one of the following: initiating a compliance check, reallocating resources, and issuing an alarm notification.

8. The method according to claim 1, wherein, Based on the user request, determining multiple target tasks includes: determining multiple tasks to be executed based on the user request; and updating the multiple tasks to be executed to the multiple target tasks that require incremental execution based on historical cache knowledge.

9. The method according to claim 1, wherein, Also includes: While invoking each corresponding task executor among the multiple target tasks, resources are allocated to the task executor based on the principle of minimum redundancy.

10. The method according to claim 1, wherein, The execution of a corresponding target task by any task executor includes: the task executor operating the corresponding application system based on the corresponding target task.

11. A system in which an AI agent executes a user request, comprising: The task decomposition module is configured to, in response to a received user request, determine multiple target tasks, the execution order among the multiple target tasks, and the task executor corresponding to the execution domain of each of the multiple target tasks, wherein the task executor includes at least one of a dedicated AI agent or an AI expert system. The task execution module is configured to call each corresponding task executor among the plurality of target tasks based on the execution order, so that each corresponding task executor among the plurality of target tasks executes the corresponding target task and outputs the corresponding execution result; The result integration module is configured to integrate all the execution results to generate the final response result.

12. The system according to claim 11, wherein, The result integration module is further configured to: before integrating all the execution results, analyze whether there are logical contradictions among all the execution results; if so, adjust the execution results with logical contradictions based on the contradiction elimination rules.

13. The system according to claim 11, wherein, The task execution module is further configured to: after invoking each corresponding task executor among the plurality of target tasks, perform a compliance check on each execution result, and adjust each execution result according to the compliance check result.

14. The system according to claim 11, wherein, When the AI ​​expert system is invoked, it performs the following operations: executes the corresponding target task based on the main AI model and generates the corresponding main execution result and the confidence level of the main execution result; If the confidence level of the main execution result is not lower than the confidence level threshold, the main execution result is used as the execution result of the AI ​​expert system and output. When the confidence level of the main execution result is lower than the confidence level threshold, the target task is executed again based on the backup AI model to generate a corresponding backup execution result and the confidence level of the backup execution result. The one with the higher confidence level between the main execution result and the backup execution result is taken as the execution result of the AI ​​expert system and output.

15. The system according to claim 11, wherein, The task execution module is further configured to: during the process of calling each corresponding task executor among the plurality of target tasks based on the execution order, monitor the cumulative loss of inaction and the risk loss of action for each of the plurality of target tasks; when the cumulative loss of inaction is greater than the risk loss of action, initiate a preset operation to reduce the cumulative loss of inaction.

16. The system according to claim 15, wherein, The process of monitoring the cumulative loss of the inaction includes: monitoring the cumulative loss of the inaction based on at least one of the opportunity cost model, the nonlinear growth function, or the time-weighted penalty function.

17. The system according to claim 15, wherein, The preset operations include at least one of the following: initiating a compliance check, reallocating resources, and issuing an alarm notification.

18. The system according to claim 11, wherein, The task decomposition module is further configured to: determine multiple tasks to be executed based on the user request; and update the multiple tasks to be executed to the multiple target tasks that need to be incrementally executed based on historical cache knowledge.

19. The system according to claim 11, wherein, The task execution module is further configured to allocate resources to each of the multiple target tasks based on the principle of minimum redundancy while invoking the corresponding task executor.

20. The system according to claim 11, wherein, The execution of a corresponding target task by any task executor includes: the task executor operating the corresponding application system based on the corresponding target task.

21. An apparatus for executing a user request by an AI agent, comprising: processor; And a memory having stored computer-executable instructions thereon, wherein, when executed by the processor, the computer-executable instructions cause the processor to perform the method of executing a user request by an AI agent as described in any one of claims 1 to 10.

22. A computer-readable storage medium having stored thereon computer-executable instructions, wherein, When executed by a processor, the computer-executable instructions cause the processor to perform the method of executing a user request by an AI agent as described in any one of claims 1 to 10.

23. A computer program product comprising computer-executable instructions, wherein, When executed by a processor, the computer-executable instructions cause the processor to perform the method of executing a user request by an AI agent as described in any one of claims 1 to 10.

Images