Generative artificial intelligence agent system and method for generating same

Abstract

The present invention relates to a generative artificial intelligence agent system and a method for generating same, and, more particularly, to a system and a method for generating and organizing a plurality of agents such that a generative artificial intelligence agent can constantly derive process performance results in response to changing conditions. The generative artificial intelligence agent system and the method for generating same, according to an embodiment of the present invention, comprise: an input unit for receiving an instruction prompt for a task to be performed by generative artificial intelligence; an agent programming unit, which uses the received instruction prompt so as to generate a meta-prompt in order to increase the performance capability of the generative artificial intelligence; a task plan inference unit for selecting a task plan on the basis of the generated meta-prompt; an agent generation unit for generating a task agent and a teamwork agent on the basis of the selected task plan; and a mapping unit for mapping memory agent interface to the generated task agent or omitting same therefrom, and mapping a skill agent interface to the generated task agent.

Description

Generative AI agent system and method for generating the same

[0001] The present invention relates to a generative artificial intelligence agent system using a meta prompt and a method for generating the same. More specifically, it relates to a system and method for generating and organizing a plurality of agents so that the generative artificial intelligence agent can consistently derive process execution results in response to changing conditions.

[0002] Robotic Process Automation (RPA) refers to the automation of repetitive and routine tasks through software.

[0003] This technology enhances work efficiency by reducing repetitive human labor through a predefined process for each step based on conditions and rules.

[0004] However, since robotic process automation technology performs tasks based on fixed conditions and rules, it is difficult to apply when exceptional situations or creative problems arise.

[0005] To solve these problems, applying a technology that integrates robotic process automation and generative artificial intelligence demonstrates a significant improvement in the ability to process cognitive processes.

[0006] In this context, Generative AI refers to Artificial Intelligence capable of generating new data, such as answers to questions, conversations, images, videos, music, and source code, using learned data.

[0007] Generative AI is being used in various industries such as finance, healthcare, automotive, manufacturing, media, entertainment, telecommunications, and energy, and generative AI technology is steadily developing across these sectors.

[0008] As prior art for processing processes based on such artificial intelligence technology, Registered Patent Publication No. 10-2658967, Published Patent Publication No. 10-2024-0063820, and Published Patent Publication No. 10-2024-0128509 are known.

[0009] Registered Patent Publication No. 10-2658967 relates to an electronic device providing a solution based on artificial intelligence, and presents a method in which a language model acquires an artificial intelligence task appropriate to a problem situation and provides an artificial intelligence-based solution API based on the artificial intelligence task.

[0010] Published Patent Application No. 10-2024-0063820 relates to a method for an artificial intelligence model to perform a task of a cleaning robot, and presents a method capable of determining information about surrounding objects of the cleaning robot and providing a suitable task for each object.

[0011] Published Patent Application No. 10-2024-0128509 presents a method for performing multi-task learning through an artificial neural network when a computer program is executed on one or more processors.

[0012] According to the prior art disclosed in Registered Patent Publication No. 10-2658967, Published Patent Publication No. 10-2024-0063820, and Published Patent Publication No. 10-2024-0128509, there is a problem in that, in order to complete the execution of a process, a person must intervene by directly manipulating each step or directly or indirectly utilize information obtained using separate software, and the execution results cannot be consistently maintained even in a diverse and changing work environment.

[0013] Therefore, it will be necessary to build a system that can improve the quality of the user's intended results by generating multiple AI agents so that performance outcomes can be maintained consistently even in a changing work environment.

[0014] [Patent Literature]

[0015] Registered Patent Publication No. 10-2658967 (Registration Date: April 16, 2024, Title: Method, Electronic Device, and System for Providing an Artificial Intelligence-Based Solution)

[0016] Published Patent Application No. 10-2024-0063820 (Publication Date: May 10, 2024, Title: Cleaning Robot and Method for Performing Tasks Thereof)

[0017] Published Patent Application No. 10-2024-0128509 (Publication Date: August 26, 2024, Title: Computer Program and Method for Multi-Task Learning)

[0018] Accordingly, the present invention is an invention conceived in accordance with the aforementioned necessity, and the technical problem of the present invention is to provide a generative artificial intelligence agent system that organizes a plurality of agents so that the generative artificial intelligence agent can consistently derive process execution results in response to changing conditions.

[0019] Furthermore, another objective of the present invention is to provide a method for generating a generative artificial intelligence agent that generates multiple agents to increase the scope of work of the generative artificial intelligence agent while simultaneously improving the quality of the execution results intended by the user.

[0020] A generative artificial intelligence agent system according to an embodiment of the present invention for solving such technical problems comprises: an input unit that receives a prompt for instructions regarding the tasks to be performed by the generative artificial intelligence; an agent programming unit that generates a meta prompt using the received prompt for instructions to enhance the performance of the generative artificial intelligence; a task plan inference unit that selects a task plan based on the generated meta prompt; an agent generation unit that generates a task agent and a teamwork agent based on the selected task plan; and a mapping unit that maps or omits a memory agent interface to the generated task agent and maps a skill agent interface to the generated task agent.

[0021] The above instruction prompt includes at least one of instructions describing a task background, task plan, task objective, and task query so that the generative AI can infer the task it performs.

[0022] The agent programming unit further includes generating a reference prompt to specify the format of the result to be output by the generative AI using the input instruction prompt, provide an example of the result to be output by the generative AI, or provide the source and origin of the content to be answered by the generative AI.

[0023] The above agent programming unit generates the meta prompt based on the instructions of the input instruction prompt using the first artificial intelligence model.

[0024] The above-described first artificial intelligence model is characterized by using the generated meta prompt to complete the instructions of the instruction prompt in order to enhance the performance of the above-described generative artificial intelligence.

[0025] The first artificial intelligence model uses the meta prompt to determine and remove instructions among the instructions that are not necessary for the execution of the generative artificial intelligence, uses the meta prompt to determine areas for improvement regarding details among the instructions necessary for the execution of the generative artificial intelligence and modifies the instructions, uses the meta prompt to determine the scope, length, and format of the instructions, and uses the meta prompt to provide additional context or examples for the instructions.

[0026] The task plan inference unit is characterized by selecting the task plan based on the generated meta prompt using a second artificial intelligence model.

[0027] The task plan inference unit includes a task creation unit that creates a task plan to be performed by the generative artificial intelligence based on the meta prompt, a task review unit that reviews the created task plan, a task execution unit that performs the reviewed task plan, and a final task selection unit that evaluates the task execution results to select a final task plan.

[0028] The task execution unit is characterized by sequentially executing each unit task of the task plan, subdividing the task plan into at least one task and executing it hierarchically, or subdividing the task plan into at least one task and executing each task, and selecting, merging, or eliminating the task execution results to derive a final task execution result.

[0029] The above agent generation unit is characterized by generating task agents that perform each unit task based on the final task plan selected by the task plan inference unit using the above second artificial intelligence model.

[0030] The above agent generation unit is characterized by generating a teamwork agent that communicates and controls each agent and each unit task using the above second artificial intelligence model.

[0031] The mapping unit maps the memory agent interface to the generated task agent using the second artificial intelligence model, wherein the mapping unit omits the memory agent interface if it is not necessary to individually recognize the interaction subject or object or specify the context.

[0032] The above memory agent interface includes at least one of short-term memory, long-term memory, entity memory, and interaction memory.

[0033] The above mapping unit is characterized by mapping the skill agent interface to the generated task agent using the above second artificial intelligence model.

[0034] The above mapping unit is characterized by mapping the task agent to the skill agent interface that is called according to the API (Application Programming Interface) calling convention.

[0035] Meanwhile, a method for generating a generative artificial intelligence agent according to another embodiment of the present invention is,

[0036] The method includes a step in which an input unit receives a prompt for instructions regarding the tasks to be performed by a generative artificial intelligence; a step in which an agent programming unit generates a meta prompt through a first artificial intelligence model using the received prompt for instructions; a step in which a task plan inference unit selects a final task plan through a second artificial intelligence model based on the generated meta prompt; a step in which an agent generation unit generates a task agent and a teamwork agent through the second artificial intelligence model based on the selected final task plan; a step in which a mapping unit maps or omits a memory agent interface to the generated task agent through the second artificial intelligence model, and a step in which a mapping unit maps a skill agent interface to the generated task agent using the second artificial intelligence model.

[0037] The above instruction prompt includes at least one of instructions describing a task background, task plan, task objective, and task query so that the generative artificial intelligence can infer the task performed by the above generative artificial intelligence.

[0038] The agent programming unit further includes generating a reference prompt to specify the format of the result to be output by the generative AI using the input instruction prompt, provide an example of the result to be output by the generative AI, or provide the source and origin of the content to be answered by the generative AI.

[0039] The first artificial intelligence model uses the meta prompt to determine and remove instructions among the instructions that are not necessary for the execution of the generative artificial intelligence, uses the meta prompt to determine areas for improvement regarding details among the instructions necessary for the execution of the generative artificial intelligence and modifies the instructions, uses the meta prompt to determine the scope, length, and format of the instructions, and uses the meta prompt to provide additional context or examples for the instructions.

[0040] The method for selecting a final task plan by the task plan inference unit described above includes the step of a task creation unit creating a task plan to be performed by the generative artificial intelligence based on the meta prompt, the step of a task review unit reviewing the created task plan, the step of a task execution unit performing the reviewed task plan, and the step of a final task selection unit evaluating the task execution results to select a final task plan.

[0041] The task execution unit is characterized by sequentially executing each unit task of the task plan, subdividing the task plan into at least one task and executing it hierarchically, or subdividing the task plan into at least one task and executing each task, and selecting, merging, or eliminating the task execution results to derive a final task execution result.

[0042] The above agent generation unit is characterized by generating task agents that perform each unit task based on the final task plan selected by the task plan inference unit using the above second artificial intelligence model.

[0043] The above agent generation unit is characterized by generating a teamwork agent that communicates and controls each agent and each unit task using the above second artificial intelligence model.

[0044] The mapping unit maps the memory agent interface to the generated task agent using the second artificial intelligence model, wherein the mapping unit omits the memory agent interface if it is not necessary to individually recognize the interacting subject or object or to specify the context.

[0045] The above memory agent interface includes at least one of short-term memory, long-term memory, entity memory, and interaction memory.

[0046] The above mapping unit is characterized by mapping the task agent to the skill agent interface that is called according to the API (Application Programming Interface) calling convention.

[0047] Specific details of other embodiments are included in the detailed description and drawings.

[0048] According to the means for solving the technical problem described above, the generative artificial intelligence agent system and the method for generating the same according to an embodiment of the present invention exhibit the following effects.

[0049] First, a generative artificial intelligence agent system and a method for generating it can consistently produce process execution results in response to changing conditions.

[0050] Second, the generative artificial intelligence agent system and the method for generating it can reduce process implementation costs because they produce consistent results even under changing conditions.

[0051] Third, the generative artificial intelligence agent system and the method for generating it can improve the quality of the performance results intended by the user by generating multiple agents that automate the process.

[0052] Figure 1 shows a generative artificial intelligence agent system according to an embodiment of the present invention.

[0053] Figure 2 shows a configuration diagram of a generative artificial intelligence agent system.

[0054] Figure 3 shows the instruction prompt of the input section.

[0055] Figure 4 shows the agent programming section.

[0056] Figure 5 shows an agent orchestration module.

[0057] Figure 6 shows the task plan inference unit.

[0058] Figure 7 shows the agent generation unit.

[0059] Figure 8 shows the mapping section.

[0060] Figure 9 shows a flowchart regarding a method for creating a generative artificial intelligence agent.

[0061] The following detailed description of the invention refers to the accompanying drawings, which illustrate specific embodiments in which the invention may be practiced, in order to clarify the objects, technical, solutions, and advantages of the invention. These embodiments are described in sufficient detail to enable a person skilled in the art to practice the invention. Furthermore, throughout the detailed description and claims of the invention, the word "comprising" and its variations are not intended to exclude other technical features, additions, components, or steps. Other objects, advantages, and characteristics of the invention will become apparent to a person skilled in the art, in part from this description and in part from the practice of the invention. The following examples and drawings are provided as examples and are not intended to limit the invention. Moreover, the invention encompasses all possible combinations of the embodiments shown herein. It should be understood that various embodiments of the invention are different but need not be mutually exclusive. It should also be understood that the location or arrangement of individual components within each disclosed embodiment may be changed without departing from the spirit and scope of the invention. Accordingly, the following detailed description is not intended to be taken in a limiting sense, and the scope of the invention is limited only by the appended claims, including all equivalents thereof, provided they are appropriately described. Furthermore, unless otherwise indicated in this specification or clearly contradictory to the context, items referred to in the singular encompass the plural unless otherwise required by the context. Additionally, in describing the invention, if it is determined that a detailed description of related known components or functions could obscure the essence of the invention, such detailed description will be omitted.

[0062] FIG. 1 shows a generative artificial intelligence agent system according to an embodiment of the present invention, and FIG. 2 shows a configuration diagram of a generative artificial intelligence agent system.

[0063] Before describing FIGS. 1 and FIGS. 2, the first artificial intelligence model (210), the second artificial intelligence model (220), and the third artificial intelligence model (230) of the generative artificial intelligence agent of the present invention can each perform tasks using different artificial intelligence.

[0064] Additionally, the first artificial intelligence model (210), the second artificial intelligence model (220), and the third artificial intelligence model (230) may include, but are not limited to, major algorithmic technologies such as a Generative Adversarial Network (GAN), a Variational Auto Encoder (VAE), a Transformer, and a Diffusion Model, and may include at least one of various algorithms.

[0065] Referring to FIGS. 1 and FIGS. 2, a generative artificial intelligence agent system (100) can generate multiple agents using a meta prompt.

[0066] The generative artificial intelligence agent system (100) may include an input unit (110), an agent programming unit (120), a task plan inference unit (130), an agent generation unit (140), a mapping unit (150), and a server (160).

[0067] The input section (110) of the generative artificial intelligence agent system (100) receives an instruction prompt (111) for the task to be performed by the generative artificial intelligence, and a detailed explanation of the instruction prompt (111) is to be described in FIG. 3.

[0068] The agent programming unit (120) of the generative artificial intelligence agent system (100) can generate a meta prompt (121) using an input instruction prompt (111) to increase the performance of the generative artificial intelligence, and a detailed description of the agent programming unit (120) will be described in FIG. 4.

[0069] The task plan inference unit (130) of the generative artificial intelligence agent system (100) can select a task plan based on a meta prompt (121) generated by the agent programming unit (120).

[0070] The agent generation unit (140) of the generative artificial intelligence agent system (100) can generate a task agent (141) and a teamwork agent (142) based on a task plan selected by the task plan inference unit (130).

[0071] The mapping unit (150) of the generative artificial intelligence agent system (100) may map or omit the memory agent interface (151) to the task agent (141) generated in the agent generation unit (140).

[0072] Additionally, the mapping unit (150) of the generative artificial intelligence agent system (100) can map a skill agent interface (152) to a task agent (141) generated in the agent generation unit (140).

[0073] Detailed descriptions of the task plan inference unit (130), agent generation unit (140), and mapping unit (150) will be described in FIG. 5.

[0074] The server (160) of the generative artificial intelligence agent system (100) can simultaneously process client requests from the input unit (110), agent programming unit (120), task plan inference unit (130), agent generation unit (140), and mapping unit (150) to respond to data that changes in real time.

[0075] Figure 3 shows the instruction prompt of the input section.

[0076] Referring to FIG. 3, the input unit (110) can receive an instruction prompt (111).

[0077] According to an embodiment of the present invention, the instruction prompt (111) received by the input unit (110) may include at least one of instructions describing a work background, a work plan, a work goal, and a work query so as to enable the generative artificial intelligence to infer the work performed by the artificial intelligence.

[0078] This is not limited to this, and multiple instructions such as clearly defining the tasks to be performed by generative AI, or the required memory, data, and skills may be added or omitted.

[0079] The entity creating the instruction prompt (111) may create it by the user directly inputting it through a user terminal or by a generative artificial intelligence determining the instruction sought by the user, but is not limited thereto and may include any electronic device capable of inputting and outputting text or inputting and outputting data corresponding to the text.

[0080] Figure 4 shows the agent programming section.

[0081] Referring to FIG. 4, the agent programming unit (120) can generate a meta prompt (121) using an instruction prompt (111) received from the input unit (110).

[0082] According to an embodiment of the present invention, a meta prompt (121) is generated to complement an instruction prompt (111) that directly instructs the execution of a generative artificial intelligence, and the instruction prompt (111) used to generate the meta prompt (121) may be a different prompt from the instruction prompt (111) that directly instructs the execution of a generative artificial intelligence.

[0083] The instruction prompt (111) used to generate such a meta prompt (121) may also be commonly referred to as a system prompt.

[0084] The agent programming unit (120) can increase the performance of the generative artificial intelligence through a meta prompt (121) generated based on the instructions of the instruction prompt (111) received from the input unit (110) using the first artificial intelligence model (210).

[0085] According to an embodiment of the present invention, the first artificial intelligence model (210) can complete the instructions of the instruction prompt (111) in the following way using a meta prompt (121) to increase the performance of the generative artificial intelligence.

[0086] The first artificial intelligence model (210) can use a meta prompt (121) to determine and remove instructions that are not necessary for the execution of the generative artificial intelligence.

[0087] Additionally, the first artificial intelligence model (210) can use the meta prompt (121) to determine the necessary details for performing the generative artificial intelligence in the instructions and modify the instructions.

[0088] Additionally, the first artificial intelligence model (210) can determine the range, length, and format of the instructions using the meta prompt (121).

[0089] Additionally, the first artificial intelligence model (210) can use the meta prompt (121) to provide additional context or examples for the instructions.

[0090] The agent programming unit (120) can generate a reference prompt (122) using the first artificial intelligence model (210), and the reference prompt (122) according to an embodiment of the present invention can use the input instruction prompt (111) to specify the format of the result to be output by the generative artificial intelligence, provide an example of the result to be output by the generative artificial intelligence, or provide the source and origin of the content to be answered by the generative artificial intelligence.

[0091] As such, the reference prompt (122) may represent a part of the instruction prompt (111) that is finally input to the generative artificial intelligence through the generative artificial intelligence agent system (100).

[0092] According to an embodiment of the present invention, a generative artificial intelligence agent system (100) may generate a meta prompt (121) or a reference prompt (122) based on an instruction prompt (111) using a single artificial intelligence model, or complete the instructions of the instruction prompt (111) using the meta prompt (121), but is not limited thereto, and may use different artificial intelligence models for generating the meta prompt or the reference prompt or completing the instructions of the instruction prompt.

[0093] Figure 5 shows an agent orchestration module, Figure 6 shows a task plan inference unit, Figure 7 shows an agent generation unit, and Figure 8 shows a mapping unit.

[0094] Referring to FIGS. 5 through 8, the agent orchestration module may include a task plan inference unit (130), an agent generation unit (140), a mapping unit (150), a reflection unit (160), a memory unit (170), a database (180), and a skill factory (190).

[0095] As can be seen in FIG. 5, the reflection section (160) of the agent orchestration module can review and correct logical errors or contradictions in relation to the prompt received from the prompt and input section (110) finally generated in the generative artificial intelligence agent system (100) and the output of the agent programming section (120), task plan inference section (130), agent generation section (140), and mapping section (150).

[0096] The memory section (170) of the agent orchestration module may store short-term memory, long-term memory, entity memory, and interaction memory, but is not limited thereto, and may store data, firmware, software, and process code, etc.

[0097] According to an embodiment of the present invention, the memory unit (170) may be used as a main memory device of a generative artificial intelligence agent system (100), and the memory may be configured in various ways such as DRAM (dynamic random access memory), SRAM (static random access memory), PRAM (phase-change random access memory), MRAM (magnetic random access memory), FERAM (ferroelectric random access memory), RRAM (resistive random access memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), and PROM (Programmable Read-Only Memory), and the number of memories may include at least one.

[0098] According to one embodiment of the present invention, since the artificial intelligence has length limits for the input prompt and output text, it may be subject to limitations even when retrieving context from memory and inputting it into the prompt, so it can be classified into short-term memory and long-term memory similar to humans.

[0099] In this context, short-term memory refers to volatile context used only for the current session, while long-term memory can represent context stored for reuse in the next conversation.

[0100] Furthermore, entity memory refers to the memory that stores object information within the total memory, while interaction memory can refer to memory that stores only the history of entity memory usage.

[0101] Thus, in order to recommend context to individuals or to effectively search for necessary context, the memory unit (170) can configure entity memory and interaction memory separately from short-term memory and long-term memory.

[0102] The database (180) of the agent orchestration module may include a knowledge graph database and a vector database.

[0103] The skill factory (190) of the agent orchestration module is an external skill for linking with an external service or system. According to an embodiment of the present invention, when the skill factory (190) receives data from an external service and external system as input to a generative artificial intelligence, or when the data of the generative artificial intelligence is output to an external service and external system, an implementation that links with an API (Application Programming Interface) may stay in the skill factory (190).

[0104] As can be seen in FIG. 6, the task plan inference unit (130) can select a task plan based on a meta prompt (121) generated by the agent programming unit (120) using the second artificial intelligence model (220).

[0105] At this time, the task plan inference unit (130) may include a task creation unit (131), a task review unit (132), a task execution unit (133), and a final task selection unit (134).

[0106] The task creation unit (131) of the task plan inference unit (130) can create a task plan to be performed by the generative artificial intelligence based on the meta prompt (121).

[0107] The task review unit (132) of the task plan inference unit (130) can review the created task plan.

[0108] The task execution unit (133) of the task plan inference unit (13) can execute the reviewed task plan.

[0109] According to an embodiment of the present invention, the task execution unit (133) may perform task sequencing (133-1), task hierarchy (133-2), or task consensus (133-3) separately to characterize the task.

[0110] The task execution unit (133) can perform task sequencing (133-1) by sequentially executing each unit task of the task plan.

[0111] The task execution unit (133) can perform task hierarchy (133-2) by subdividing the task plan into at least one or more tasks and performing them hierarchically.

[0112] In the task hierarchy described above (133-2), the preceding task executes the subsequent task and returns the result of the subsequent task to the preceding task so that the preceding task can continue execution.

[0113] The task execution unit (133) can perform task consensus (133-3) by subdividing the task plan into at least one task and performing each task, and by selecting, merging, or eliminating the task execution results so that the task execution unit (133) derives the final task execution result.

[0114] To explain the task consensus (133-3) described above in more detail, it can be explained by combining and summarizing the different results produced by the second artificial intelligence model (220), or by selecting a specific result from among the results produced by the second artificial intelligence model.

[0115] According to one embodiment of the present invention, procedural, hierarchical, and logical execution can be carried out without a person having to program separately through task sequencing (133-1), task hierarchy (133-2), and task consensus (133-3) of the task execution unit (133).

[0116] The final task selection unit (134) of the task plan inference unit (130) can select a final task plan (130-1) by evaluating the results of the task execution.

[0117] According to an embodiment of the present invention, the task of the task planning inference unit (130) may include any task that can be solved using artificial intelligence, such as image classification, image segmentation, image conversion, image classification, object detection, text generation, speech recognition, question, and answer.

[0118] As can be seen in FIG. 7, the agent generation unit (140) can generate a task agent (141) that performs each unit task based on the final task plan (130-1) selected by the task plan inference unit (130) using the second artificial intelligence model (220).

[0119] Additionally, the agent generation unit (140) can generate a teamwork agent (142) that communicates and controls each agent and each unit task using the second artificial intelligence model (220).

[0120] As can be seen in FIG. 8, the mapping unit (150) maps the memory agent interface (151) to the task agent (141) generated by the agent generation unit (140) using the second artificial intelligence model (220), but the mapping unit (150) may omit mapping the memory agent interface (151) if it is not necessary to individually recognize the interaction subject or object or specify the context.

[0121] According to an embodiment of the present invention, since the result may vary depending on the user even with the same instruction prompt, the memory agent interface (151) may receive past memories of interaction through the memory agent interface to specify the context of interaction between the user and the artificial intelligence or between the artificial intelligence and the artificial intelligence.

[0122] Therefore, the memory agent interface (151) may include at least one of short-term memory, long-term memory, entity memory, and interaction memory.

[0123] The mapping unit (150) can map a skill agent interface (152) to a task agent (141) generated in the agent generation unit (140) using the second artificial intelligence model (220).

[0124] According to an embodiment of the present invention, the mapping unit (150) can map a task agent (141) to a skill agent interface (152) that queries and calls in conjunction with an API according to an API (Application Programming Interface) calling protocol for the context requested by the user.

[0125] Figure 9 shows a flowchart regarding a method for creating a generative artificial intelligence agent.

[0126] Referring to FIG. 9, the method for generating a generative artificial intelligence agent of a generative artificial intelligence agent system (100) is as follows.

[0127] The input unit (110) proceeds to the step of receiving a prompt (111) for instructions regarding the task to be performed by the generative artificial intelligence (S100).

[0128] Next, the step of generating a meta prompt (121) through the first artificial intelligence model (210) of the agent programming unit (120) using the input instruction prompt (111) is carried out (S110).

[0129] Next, based on the meta prompt (121) generated by the agent programming unit (120), the task plan inference unit (130) proceeds to the step of selecting the final task plan (130-1) through the second artificial intelligence model (220) (S120).

[0130] Next, based on the final task plan (130-1) selected by the task plan inference unit (130), the agent generation unit (140) proceeds with the step of generating a task agent (141) and a teamwork agent (142) through the second artificial intelligence model (220) (S130).

[0131] Next, the mapping unit (150) maps or omits the memory agent interface (151) to the task agent (141) generated in the agent generation unit (140) through the second artificial intelligence model (220), and

[0132] The mapping unit (150) proceeds with the step of mapping the task agent (141) generated in the agent generation unit (140) to the skill agent interface through the second artificial intelligence model (220) (S140), and the generative artificial intelligence agent system (100) completes the method of generating the generative artificial intelligence agent.

[0133] Although the present invention has been described above with reference to the embodiments illustrated in the drawings, this is merely illustrative, and those skilled in the art will understand that various modifications and equivalent alternative embodiments are possible therefrom. For example, in the embodiments of the present invention, the invention has been described assuming a generative artificial intelligence agent, but it can be equally applied to all cases where a method is presented to consistently derive process execution results using artificial intelligence without any significant modification. Therefore, the true technical scope of protection of the present invention should be determined only by the appended claims.

[0134] [Explanation of the symbol]

[0135] 100 : Generative AI Agent System

[0136] 110 : Input section

[0137] 120 : Agent Programming Department

[0138] 130 : Task Planning Inference Unit

[0139] 140 : Agent creation section

[0140] 150 : Mapping section

[0141] 160 : Server

Claims

1. An input unit that receives a prompt for instructions regarding the tasks to be performed by the generative artificial intelligence; An agent programming unit that generates a meta prompt using the input instruction prompt to enhance the performance of the above generative artificial intelligence; A task plan inference unit that selects a task plan based on the meta prompt generated above; An agent generation unit that generates task agents and teamwork agents based on the above-mentioned selected task plan; and A generative artificial intelligence agent system comprising: a mapping unit that maps or omits a memory agent interface to the generated task agent and maps a skill agent interface to the generated task agent.

2. In Paragraph 1, The above instruction prompt is, A generative artificial intelligence agent system comprising at least one of instructions describing a task background, task plan, task goal, and task query to enable the generative artificial intelligence to infer the task performed by the generative artificial intelligence.

3. In Paragraph 2, The above agent programming unit uses the input instruction prompt, Specify the format for the result to be output by the above generative AI, or Provide an example of the result to be output by the above generative AI, or A generative AI agent system further comprising generating a reference prompt to provide the source and origin of the content to be answered by the generative AI.

4. In Paragraph 2, The above agent programming unit is a generative artificial intelligence agent system that generates the meta prompt based on the instructions of the input instruction prompt using a first artificial intelligence model.

5. In Paragraph 4, A generative artificial intelligence agent system characterized by the above-mentioned first artificial intelligence model completing the instructions of the instruction prompt using the generated meta prompt to enhance the performance of the generative artificial intelligence.

6. In Paragraph 4, The above-mentioned first artificial intelligence model is, Using the above meta prompt, instructions among the above instructions that are not necessary for the execution of the generative artificial intelligence are determined and removed, and Using the above meta prompt, determine areas for improvement regarding the details required for the execution of the above generative artificial intelligence among the above instructions, and modify the above instructions, and Determine the range, length, and format of the above instructions using the above meta prompt, and A generative artificial intelligence agent system that provides additional context or examples for the above instructions using the above meta prompt.

7. In Paragraph 1, A generative artificial intelligence agent system characterized by the task plan inference unit selecting the task plan based on the generated meta prompt using a second artificial intelligence model.

8. In Paragraph 7, The above task plan inference unit, A task creation unit that creates a task plan to be performed by the generative artificial intelligence based on the above meta prompt; A task review unit that reviews the task plan written above; A task execution unit that performs the task plan reviewed above; and A generative artificial intelligence agent system comprising: a final task selection unit that evaluates the results of the above-mentioned task execution and selects a final task plan.

9. In Paragraph 8, The above task execution unit is, Perform each unit task of the above task plan sequentially, or Perform the above task plan hierarchically by subdividing it into at least one or more of the above tasks, or A generative artificial intelligence agent system characterized by subdividing the above task plan into at least one or more of the above tasks, performing each task, and selecting, merging, or eliminating the results of the task execution to derive a final task execution result.

10. In Paragraph 9, A generative artificial intelligence agent system characterized in that the agent generation unit generates task agents that perform each unit task based on the final task plan selected by the task plan inference unit using the second artificial intelligence model.

11. In Paragraph 10, A generative artificial intelligence agent system characterized by the above agent generation unit generating a teamwork agent that communicates and controls each agent and each unit task using the above second artificial intelligence model.

12. In Paragraph 1, The above mapping unit maps the memory agent interface to the generated task agent using the above second artificial intelligence model, wherein A generative artificial intelligence agent system characterized by the above mapping unit omitting the memory agent interface if it is not necessary to individually recognize the interacting subject or object or to specify the context.

13. In Paragraph 12, The above memory agent interface is a generative artificial intelligence agent system comprising at least one of short-term memory, long-term memory, entity memory, and interaction memory.

14. In Paragraph 10, A generative artificial intelligence agent system characterized by the mapping unit mapping the skill agent interface to the generated task agent using the second artificial intelligence model.

15. In Paragraph 14, A generative artificial intelligence agent system characterized by the above mapping unit mapping the task agent to the skill agent interface called according to the API (Application Programming Interface) calling convention.

16. A step in which an input unit receives a prompt for instructions regarding the tasks to be performed by the generative artificial intelligence; A step in which the agent programming unit generates a meta prompt through the first artificial intelligence model using the input instruction prompt received above; A step in which a task plan inference unit selects a final task plan through a second artificial intelligence model based on the meta prompt generated above; A step in which an agent generation unit generates a task agent and a teamwork agent through the second artificial intelligence model based on the above-mentioned final task plan; and The mapping unit maps or omits the memory agent interface to the task agent generated above through the second artificial intelligence model, and A method for generating a generative artificial intelligence agent comprising the step of a mapping unit mapping a skill agent interface to the generated task agent using the second artificial intelligence model.

17. In Paragraph 16, The above instruction prompt is, A method for generating a generative AI agent comprising at least one of instructions describing a task background, task plan, task goal, and task query so as to enable the generative AI to infer the task performed by the generative AI.

18. In Paragraph 17, The above agent programming unit uses the input instruction prompt, Specify the format for the result to be output by the above generative AI, or Provide an example of the result to be output by the above generative AI, or A method for generating a generative AI agent that further includes generating a reference prompt to provide the source and origin of the content to be answered by the generative AI.

19. In Paragraph 17, The above-mentioned first artificial intelligence model Using the above meta prompt, instructions among the above instructions that are not necessary for the execution of the generative artificial intelligence are determined and removed, and Using the above meta prompt, determine areas for improvement regarding the details required for the execution of the above generative artificial intelligence among the above instructions, and modify the above instructions, and Determine the range, length, and format of the above instructions using the above meta prompt, and A method for generating a generative artificial intelligence agent that uses the above meta prompt to provide additional context or examples for the above instructions.

20. In Paragraph 16, The final task plan selection method of the above task plan inference unit is, A step in which a task creation unit creates a task plan to be performed by the generative artificial intelligence based on the above meta prompt; A step in which the task review department reviews the task plan prepared above; A step in which the task execution unit performs the above-reviewed task plan; and A method for generating a generative artificial intelligence agent comprising the step of a final task selection unit evaluating the results of the task execution and selecting a final task plan.

21. In Paragraph 16, The above task execution unit is, Perform each unit task of the above task plan sequentially, or Perform the above task plan hierarchically by subdividing it into at least one or more of the above tasks, or A method for generating a generative artificial intelligence agent characterized by subdividing the above task plan into at least one or more of the above tasks, performing each task, and selecting, merging, or eliminating the results of the task performance to derive a final task performance result.

22. In Paragraph 21, A method for generating a generative artificial intelligence agent, characterized in that the agent generation unit generates a task agent that performs each unit task based on the final task plan selected by the task plan inference unit using the second artificial intelligence model.

23. In Paragraph 22, A method for generating a generative artificial intelligence agent, characterized in that the agent generation unit generates a teamwork agent that controls each agent and each unit task by communicating using the second artificial intelligence model.

24. In Paragraph 16, The above mapping unit maps the memory agent interface to the generated task agent using the above second artificial intelligence model, wherein A method for generating a generative artificial intelligence agent characterized by the above mapping unit omitting the memory agent interface if it is not necessary to individually recognize the interacting subject or object or to specify the context.

25. In Paragraph 24, The above memory agent interface is a method for generating a generative artificial intelligence agent comprising at least one of short-term memory, long-term memory, entity memory, and interaction memory.

26. In Paragraph 16, A method for generating a generative artificial intelligence agent, characterized in that the mapping unit maps the task agent to the skill agent interface called according to the API (Application Programming Interface) calling convention.

Images