Supervision system and method based on artificial intelligence model information output

Abstract

The invention discloses a supervision system and method based on artificial intelligence model information output, and relates to the technical field of artificial intelligence, and the system comprises a first artificial intelligence model and a second artificial intelligence model which are logically isolated; the first artificial intelligence model performs analysis and retrieval according to the input information of the user, generates output information and feeds back the output information to the second artificial intelligence model; receiving risk information fed back by the second artificial intelligence model, and regenerating output information corresponding to the risk information until no-risk information is received; the second artificial intelligence model constructs a graph according to laws and regulations, ethics and morality, social risks and copyright information, a risk knowledge graph is obtained, and compliance verification is conducted on output information in combination with the risk knowledge graph. According to the method, through one-way dialogue of the two logically isolated artificial intelligence models, a closed-loop system of automatic compliance supervision and main model autonomous optimization is realized, and potential interference of the supervision model on main model decision logic is eliminated.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a supervision system and method based on artificial intelligence model information output. Background Art

[0002] With the rapid development of artificial intelligence (AI) technology, AI models have been widely used in natural language processing and machine learning tasks. However, existing technologies primarily rely on the model's own calibration mechanisms for output information, lacking external supervision and verification. Furthermore, some systems employing cascade models experience parameter coupling, and the supervisory model can negatively impact the judgment logic of the primary model. Traditional feedback mechanisms, which include specific explanations of the causes, can easily lead to path dependencies in the primary model, inhibiting its autonomous optimization capabilities. Summary of the Invention

[0003] The purpose of this application is to provide a supervision system and method based on artificial intelligence model information output, which can prevent artificial intelligence self-inspection cheating, and use two artificial intelligence models (supervision model and main model) to realize a closed-loop system of automated compliance supervision and autonomous optimization of the main model, eliminating the potential interference of the supervision model on the decision-making logic of the main model.

[0004] To achieve the above objectives, this application provides the following solutions:

[0005] In a first aspect, the present application provides a supervision system based on artificial intelligence model information output, comprising: a first artificial intelligence model and a second artificial intelligence model; the first artificial intelligence model and the second artificial intelligence model are two logically isolated artificial intelligence models;

[0006] The first artificial intelligence model includes: a first information analysis and retrieval module and a first iterative training module;

[0007] The first information analysis and retrieval module is used to analyze and retrieve the user's input information, generate output information based on the requirements and content of the input information, and feed the output information back to the second artificial intelligence model;

[0008] The first iterative training module is configured to receive risk information fed back by the second artificial intelligence model and cause the first information analysis and retrieval module to regenerate output information corresponding to the risk information until risk-free information fed back by the second artificial intelligence model is received, at which point the first information analysis and retrieval module outputs output information corresponding to the risk-free information;

[0009] The second artificial intelligence model includes: a knowledge graph module, a second information processing module and an information feedback module;

[0010] The knowledge graph module is used to construct a graph based on laws and regulations, ethics, social risks, and copyright information to obtain a risk knowledge graph;

[0011] A second information processing module is configured to receive output information fed back by the first artificial intelligence model, and perform compliance verification on the output information in combination with the risk knowledge graph to obtain a verification result; the verification result may include: risk-free information or risk-positive information;

[0012] The information feedback module is used to feed back the verification result to the first artificial intelligence model.

[0013] Optionally, the second artificial intelligence model further includes:

[0014] The manual review module is used to obtain manual review information; the manual review information is the result of manual risk judgment on the verification result; the judgment result includes: no risk information or risk information.

[0015] Optionally, the first artificial intelligence model further includes:

[0016] The first input module is used to receive user input information.

[0017] Optionally, the first artificial intelligence model further includes:

[0018] Self-test module, used to:

[0019] The output information is self-checked, and when there is a risk in the self-check, the first iterative training module is caused to regenerate the output information.

[0020] Optionally, the self-checking the output information specifically includes:

[0021] Build a graph based on laws and regulations, ethics, social risks, and copyright information to obtain a risk knowledge graph;

[0022] The output information is verified for compliance in combination with the risk knowledge graph to obtain a verification result; the verification result includes: no risk or risk.

[0023] Optionally, the first artificial intelligence model further includes:

[0024] The first information output module is used to receive the risk-free information fed back by the second artificial intelligence model and output the output information corresponding to the risk-free information.

[0025] In a second aspect, this application provides a supervision method based on artificial intelligence model information output, including:

[0026] Utilizing a first artificial intelligence model to analyze and retrieve user input information, generate output information based on the requirements and content of the input information, and feed the output information back to a second artificial intelligence model; the first artificial intelligence model and the second artificial intelligence model are two logically isolated artificial intelligence models;

[0027] Using a second artificial intelligence model, combined with a risk knowledge graph, the output information is subjected to compliance verification to obtain a verification result; the verification result includes: risk-free information or risky information; the risk knowledge graph is a knowledge graph constructed based on laws and regulations, ethics, social risks, and copyright information;

[0028] Feeding the verification result back to the first artificial intelligence model using the second artificial intelligence model;

[0029] Utilize the first artificial intelligence model to receive the verification result fed back by the second artificial intelligence model; when the verification result is risky information, regenerate the output information corresponding to the risky information, and feed the regenerated output information back to the second artificial intelligence model, and at the same time return to the step of "using the second artificial intelligence model and combining the risk knowledge graph to verify the compliance of the output information and obtain the verification result"; until the risk-free information fed back by the second artificial intelligence model is received, output the output information corresponding to the risk-free information.

[0030] Optionally, before using the first artificial intelligence model to analyze and retrieve the user's input information, generating output information according to the requirements and content in the input information, and feeding the output information back to the second artificial intelligence model, the method further includes:

[0031] The first artificial intelligence model is used to receive user input information.

[0032] Optionally, the first artificial intelligence model is used to receive a verification result fed back by the second artificial intelligence model; when the verification result indicates risk information, the output information corresponding to the risk information is regenerated and the regenerated output information is fed back to the second artificial intelligence model, and the process returns to the step of "using the second artificial intelligence model to verify the compliance of the output information in combination with the risk knowledge graph to obtain a verification result"; until risk-free information is received from the second artificial intelligence model, the output information corresponding to the risk-free information is output, specifically including:

[0033] Utilize the first artificial intelligence model to receive the verification result fed back by the second artificial intelligence model; when the verification result is risky information, regenerate the output information corresponding to the risky information, perform self-inspection on the regenerated output information, and regenerate the output information when the self-inspection shows risk, until the self-inspection shows no risk, feed back the regenerated output information to the second artificial intelligence model, and simultaneously return to the step of "utilizing the second artificial intelligence model and combining the risk knowledge graph to perform compliance verification on the output information to obtain the verification result"; until risk-free information is received from the second artificial intelligence model, output the output information corresponding to the risk-free information.

[0034] Optionally, after using the first artificial intelligence model to receive the verification result fed back by the second artificial intelligence model; when the verification result is risk-free information, using the first artificial intelligence model, output information corresponding to the risk-free information is output.

[0035] According to the specific embodiments provided in this application, this application discloses the following technical effects:

[0036] The present application provides a supervision system and method based on artificial intelligence model information output, the system comprising: a first artificial intelligence model and a second artificial intelligence model; the first artificial intelligence model and the second artificial intelligence model are two logically isolated artificial intelligence models; the first artificial intelligence model comprises: a first information analysis and retrieval module and a first iterative training module; the first information analysis and retrieval module is used to analyze and retrieve according to the user's input information, and generate output information according to the requirements and content in the input information, and feed the output information back to the second artificial intelligence model; the first iterative training module is used to receive risky information fed back by the second artificial intelligence model, and enable the first information analysis and retrieval module to The output information is regenerated until the risk-free information fed back by the second artificial intelligence model is received, and the output information corresponding to the risk-free information is output; the second artificial intelligence model includes: a knowledge graph module, a second information processing module and an information feedback module; the knowledge graph module is used to construct a graph according to laws and regulations, ethics, social risks and copyright information to obtain a risk knowledge graph; the second information processing module is used to receive the output information fed back by the first artificial intelligence model, and perform compliance verification on the output information in combination with the risk knowledge graph to obtain a verification result; the verification result includes: risk-free information or risky information; the information feedback module is used to feed back the verification result to the first artificial intelligence model. In the related technology, supervision is carried out in an unisolated artificial intelligence model, and its supervision mechanism and generation mechanism are in the same logic, which makes collusion more likely to occur. In the present application, two logically isolated artificial intelligence models have a one-way dialogue, realizing a closed-loop system of automated compliance supervision and autonomous optimization of the main model, eliminating the potential interference of the supervision model on the decision-making logic of the main model. In addition, the traditional feedback mechanism includes specific reason descriptions, which can easily cause the main model to generate path dependence and inhibit its autonomous optimization capability. In the present application, the second artificial intelligence model only informs the first artificial intelligence model that there is a problem with the output information, without explaining the specific reasons, thereby effectively improving the autonomous optimization capability of the first artificial intelligence model. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0038] Figure 1 A flowchart of a supervision method based on artificial intelligence model information output provided in one embodiment of the present application. DETAILED DESCRIPTION

[0039] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0040] With the rapid development of artificial intelligence (AI) technology, AI models have been widely used in natural language processing and machine learning tasks. However, existing technologies primarily rely on the model's own calibration mechanism for output information, lacking external supervision and verification. Furthermore, some systems employing cascade models experience parameter coupling, and the supervisory model may adversely affect the judgment logic of the main model. Traditional feedback mechanisms, which include specific explanations of the causes, can easily lead to path dependency in the main model, inhibiting its autonomous optimization capabilities. Furthermore, existing supervisory mechanisms, relying solely on AI, do not fully address certain risks. This verification approach may result in output information that does not conform to social requirements, such as ethics, laws, and regulations, thereby posing potential risks.

[0041] This application uses an independent second artificial intelligence model to effectively supervise the output information of the first artificial intelligence model, thereby realizing an effective closed loop of automated compliance supervision and autonomous optimization of the main model.

[0042] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0043] In an exemplary embodiment, a supervision system based on artificial intelligence model information output is provided, including: a first artificial intelligence model and a second artificial intelligence model; the first artificial intelligence model and the second artificial intelligence model are two logically isolated artificial intelligence models.

[0044] The first artificial intelligence model includes: a first input module, a first information analysis and retrieval module, a self-check module and a first iterative training module.

[0045] The first input module is used to receive user input information.

[0046] The first information analysis and retrieval module is used to analyze and retrieve based on the user's input information, generate output information based on the requirements and content in the input information, and feed the output information back to the second artificial intelligence model.

[0047] The first iterative training module is used to receive risky information fed back by the second artificial intelligence model, and enable the first information analysis and retrieval module to regenerate the output information corresponding to the risky information until risk-free information fed back by the second artificial intelligence model is received, and then the output information corresponding to the risk-free information is output.

[0048] The first iterative training module in this embodiment receives risk information fed back by the second artificial intelligence model, iteratively trains the first artificial intelligence model, processes the input information and controls the risk of the output information until it is mature, and outputs compliance output information.

[0049] The self-test module is used to:

[0050] The output information is self-checked, and when there is a risk in the self-check, the first iterative training module is caused to regenerate the output information.

[0051] The self-checking of the output information specifically includes:

[0052] Build a graph based on laws and regulations, ethics, social risks, and copyright information to obtain a risk knowledge graph;

[0053] The output information is verified for compliance in combination with the risk knowledge graph to obtain a verification result; the verification result includes: no risk or risk.

[0054] In addition, the first artificial intelligence model also includes:

[0055] The first information output module is used to receive the risk-free information fed back by the second artificial intelligence model and output the output information corresponding to the risk-free information; the function of this module is that if the output information generated for the first time is fed back as risk-free by the second artificial intelligence model, then it will be directly output using this module.

[0056] The second artificial intelligence model includes: a knowledge graph module, a second information processing module and an information feedback module.

[0057] The knowledge graph module is used to construct and save a graph based on laws and regulations, ethics, social risks, and copyright information to obtain a risk knowledge graph.

[0058] The second information processing module is used to receive the output information fed back by the first artificial intelligence model, and perform compliance verification on the output information in combination with the risk knowledge graph to obtain a verification result; the verification result includes: risk-free information or risky information.

[0059] The second information processing module in this embodiment receives the output information fed back by the first information output module, processes the output information in combination with the graph constructed by the knowledge graph module, performs compliance verification on the output information, and classifies the verification results as risk-free or risky.

[0060] The information feedback module is used to feed back the verification result to the first artificial intelligence model.

[0061] If the verification result is risk-free, the verification result is directly fed back to the first artificial intelligence model, and the first artificial intelligence model outputs the output information;

[0062] If the verification result is risky, the risk information will be fed back to the first artificial intelligence model. The first artificial intelligence model will self-check the output information, regenerate the output information, and repeat the verification until the output is compliant.

[0063] The second artificial intelligence model also includes:

[0064] The manual review module is used to obtain manual review information; the manual review information is the result of manual risk judgment on the verification result; the judgment result includes: no risk information or risk information.

[0065] In this embodiment, the verification results are fed back to a manual review module for manual review and risk assessment, which is then fed back to the first AI model. Based on the risk assessment results, the first AI model determines whether to directly output the information or perform a new search. The corresponding first AI model also includes an iterative training module. Based on the risk feedback results, the first AI model undergoes iterative training until it has matured its ability to process the input information and control the risk of the output information, ultimately outputting compliance-compliant information.

[0066] This embodiment has the following advantages over the prior art:

[0067] This embodiment effectively solves the problem of lack of verification of artificial intelligence model output information in the existing technology by introducing an external supervision mechanism, and combines it with manual review methods to ensure that the output information complies with ethical and legal requirements, reduce social risks and infringement risks, improve the security and reliability of artificial intelligence models, and promote the healthy and sustainable development of artificial intelligence technology.

[0068] This embodiment implements a closed-loop system of automated compliance supervision and autonomous optimization of the master model.

[0069] This embodiment eliminates potential interference of the supervisory model on the decision logic of the main model.

[0070] This embodiment improves model iteration efficiency while ensuring information security.

[0071] This embodiment avoids information leakage and encourages the first artificial intelligence model to self-adjust. The first artificial intelligence model is retrained after each feedback, gradually improving compliance.

[0072] Based on the same inventive concept, Figure 1 As shown, the embodiment of the present application also provides a supervision method based on artificial intelligence model information output, comprising the following steps:

[0073] S1. Utilize the first artificial intelligence model to analyze and retrieve the user's input information, generate output information based on the requirements and content in the input information, and feed the output information back to the second artificial intelligence model; the first artificial intelligence model and the second artificial intelligence model are two independent artificial intelligence models that are logically isolated.

[0074] In this embodiment, the user inputs information, the first artificial intelligence model receives the input information, searches and processes the input information, and generates output information.

[0075] S2. Use the second artificial intelligence model in combination with the risk knowledge graph to verify the compliance of the output information to obtain a verification result; the verification result includes: risk-free information or risky information; the risk knowledge graph is a knowledge graph constructed based on laws and regulations, ethics, social risks and copyright information.

[0076] In this embodiment, a second AI model is independently deployed to perform multi-dimensional risk oversight, with output information fed into the second AI model. The first and second AI models rely on the existing AI model and are completely independent. The second AI model verifies the compliance of the output information and feeds back the verification results to the first AI model.

[0077] The second artificial intelligence model conducts compliance supervision on the output information in terms of laws and regulations, ethics, social risks, infringement risks, etc. This process does not involve analysis and retrieval of input information.

[0078] The output information review step also includes a manual review method, which is combined with the second artificial intelligence model review method to conduct a comprehensive review of the output information.

[0079] S3. Use the second artificial intelligence model to feed back the verification result to the first artificial intelligence model.

[0080] S4. Use the first artificial intelligence model to receive the verification result fed back by the second artificial intelligence model; when the verification result is risky information, regenerate the output information corresponding to the risky information, and feed the regenerated output information back to the second artificial intelligence model, and at the same time return to the step of "using the second artificial intelligence model, combined with the risk knowledge graph, to verify the compliance of the output information and obtain the verification result"; until the risk-free information fed back by the second artificial intelligence model is received, the output information corresponding to the risk-free information is output.

[0081] Specifically, the first artificial intelligence model is used to receive the verification results fed back by the second artificial intelligence model; when the verification result is risky information, the output information corresponding to the risky information is regenerated, the regenerated output information is self-checked, and the output information is regenerated when the self-check is risky. After the self-check is risk-free, the regenerated output information is fed back to the second artificial intelligence model, and the step of "using the second artificial intelligence model and combining the risk knowledge graph to verify the compliance of the output information and obtain the verification result" is returned; until the risk-free information fed back by the second artificial intelligence model is received, the output information corresponding to the risk-free information is output. Among them, during the self-check, the output information corresponding to the risky information can also be compared with the regenerated output information. The output information corresponding to the risky information can also be compared with the output information corresponding to the risk-free information, so as to evolve on its own.

[0082] In this embodiment, a limited feedback mechanism is established, and the second artificial intelligence model combines manual review to classify the preliminary output information compliance verification results into risky and non-risky.

[0083] If the verification result is risk-free, that is, the output information meets the output requirements, the risk-free verification result will be fed back to the first artificial intelligence model, and the artificial intelligence model will output the output information as the final information.

[0084] If the verification result is risky, that is, the output information involving the above supervision information does not comply with the output regulations, the second artificial intelligence model will feedback the risk verification result to the first artificial intelligence model, and the feedback information is only the result information and does not include the reason. At this time, the first artificial intelligence model will re-search and process according to the feedback information.

[0085] Based on the risk marker triggering iterative training of the main model, the first artificial intelligence model re-searches and processes according to the feedback information, trains its own retrieval mechanism, and regenerates the output information until the output information supervision result is risk-free and the information is output.

[0086] The first AI model undergoes iterative training based on risk markers, triggering the main model through cyclical verification until the output is compliant. Input information is processed and the risk of output information is controlled until mature, and compliance output information is output.

[0087] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0088] Those skilled in the art will understand that all or part of the processes in the above-mentioned embodiment methods can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM may be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0089] The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may include, but are not limited to, general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic units, data processing logic units based on quantum computing, and the like.

[0090] The technical features of the above embodiments can be combined arbitrarily. In order to make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0091] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A supervision system based on artificial intelligence model information output, characterized in that: include: a first artificial intelligence model and a second artificial intelligence model; The first artificial intelligence model and the second artificial intelligence model are two logically isolated artificial intelligence models; The first artificial intelligence model includes: a first information analysis and retrieval module and a first iterative training module; The first information analysis and retrieval module is used to analyze and retrieve the user's input information, generate output information based on the requirements and content of the input information, and feed the output information back to the second artificial intelligence model; The first iterative training module is configured to receive risk information fed back by the second artificial intelligence model and cause the first information analysis and retrieval module to regenerate output information corresponding to the risk information until risk-free information fed back by the second artificial intelligence model is received, at which point the first information analysis and retrieval module outputs output information corresponding to the risk-free information; The second artificial intelligence model includes: a knowledge graph module, a second information processing module and an information feedback module; The knowledge graph module is used to construct a graph based on laws and regulations, ethics, social risks, and copyright information to obtain a risk knowledge graph; A second information processing module is configured to receive output information fed back by the first artificial intelligence model, and perform compliance verification on the output information in combination with the risk knowledge graph to obtain a verification result; the verification result may include: risk-free information or risk-positive information; The information feedback module is used to feed back the verification result to the first artificial intelligence model.

2. The supervision system based on artificial intelligence model information output according to claim 1 is characterized in that: The second artificial intelligence model also includes: The manual review module is used to obtain manual review information; the manual review information is the result of manual risk judgment on the verification result; the judgment result includes: no risk information or risk information.

3. The supervision system based on artificial intelligence model information output according to claim 1 is characterized in that: The first artificial intelligence model also includes: The first input module is used to receive user input information.

4. The supervision system based on artificial intelligence model information output according to claim 1 is characterized in that: The first artificial intelligence model also includes: Self-test module, used to: The output information is self-checked, and when there is a risk in the self-check, the first iterative training module is caused to regenerate the output information.

5. The supervision system based on artificial intelligence model information output according to claim 4 is characterized in that: The self-checking of the output information specifically includes: Build a graph based on laws and regulations, ethics, social risks, and copyright information to obtain a risk knowledge graph; The output information is verified for compliance in combination with the risk knowledge graph to obtain a verification result; the verification result includes: no risk or risk.

6. The supervision system based on artificial intelligence model information output according to claim 1 is characterized in that: The first artificial intelligence model also includes: The first information output module is used to receive the risk-free information fed back by the second artificial intelligence model and output the output information corresponding to the risk-free information.

7. A supervision method based on information output of an artificial intelligence model, characterized in that: include: Utilizing a first artificial intelligence model to analyze and retrieve user input information, generate output information based on the requirements and content of the input information, and feed the output information back to a second artificial intelligence model; the first artificial intelligence model and the second artificial intelligence model are two logically isolated artificial intelligence models; Using the second artificial intelligence model and combining it with the risk knowledge graph, the output information is verified for compliance to obtain a verification result; The verification result includes: no risk information or risk information; the risk knowledge graph is a knowledge graph constructed based on laws and regulations, ethics, social risks and copyright information; Feeding the verification result back to the first artificial intelligence model using the second artificial intelligence model; Utilize the first artificial intelligence model to receive the verification result fed back by the second artificial intelligence model; when the verification result is risky information, regenerate the output information corresponding to the risky information, and feed the regenerated output information back to the second artificial intelligence model, and at the same time return to the step of "using the second artificial intelligence model and combining the risk knowledge graph to verify the compliance of the output information and obtain the verification result"; until the risk-free information fed back by the second artificial intelligence model is received, output the output information corresponding to the risk-free information.

8. The supervision method based on artificial intelligence model information output according to claim 7, characterized in that: Before using the first artificial intelligence model to analyze and retrieve the user's input information, generating output information based on the requirements and content in the input information, and feeding the output information back to the second artificial intelligence model, the method further includes: The first artificial intelligence model is used to receive user input information.

9. The supervision method based on artificial intelligence model information output according to claim 7, characterized in that: Utilize the first artificial intelligence model to receive the verification result fed back by the second artificial intelligence model; when the verification result indicates risk information, regenerate the output information corresponding to the risk information, and feed the regenerated output information back to the second artificial intelligence model, while returning to the step of "using the second artificial intelligence model to verify the compliance of the output information in combination with the risk knowledge graph to obtain a verification result"; until receiving risk-free information fed back by the second artificial intelligence model, output the output information corresponding to the risk-free information, specifically including: Utilize the first artificial intelligence model to receive the verification result fed back by the second artificial intelligence model; when the verification result is risky information, regenerate the output information corresponding to the risky information, perform self-inspection on the regenerated output information, and regenerate the output information when the self-inspection shows risk, until the self-inspection shows no risk, feed back the regenerated output information to the second artificial intelligence model, and simultaneously return to the step of "using the second artificial intelligence model and combining the risk knowledge graph to perform compliance verification on the output information to obtain a verification result"; until risk-free information is received from the second artificial intelligence model, output the output information corresponding to the risk-free information.

10. The supervision method based on artificial intelligence model information output according to claim 7, characterized in that: After receiving the verification result fed back by the second artificial intelligence model using the first artificial intelligence model; when the verification result is risk-free information, using the first artificial intelligence model, output information corresponding to the risk-free information is output.

Images