Artificial intelligence behavior pre-judgment method for human-computer interaction

Abstract

The application discloses an AI behavior pre-judgment method for human-computer interaction, and is applied to an electronic terminal with AI capability, and the core scheme is as follows: after the terminal is started, all human-computer interaction entrances of the electronic terminal are continuously taken over, a unique necessary instruction preprocessing channel for all human-computer interaction instructions to enter a native operating system is constructed; for each human-computer interaction instruction, interception is completed before a whole process of receiving, analyzing, scheduling and executing of the instruction by the native operating system, and AI behavior attribute pre-judgment is executed; the instruction determined as an AI interaction behavior is introduced into a special AI control system for processing, and the instruction determined as a non-AI interaction behavior is released to the native operating system for processing. The application solves the problems of after-control, easy bypassing and missed or mistaken judgment in the prior art, realizes AI behavior whole-link source control, the scheme is non-invasive deployment, is compatible with all kinds of electronic terminals, and has strong industrial practicability.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence terminal management technology, specifically to a method for pre-judgment of AI behavior for human-computer interaction. It is applicable to all electronic terminals equipped with AI capabilities, such as smartphones, in-vehicle terminals, smart hardware, edge computing devices, and AI robots. It is fully compatible with the terminal's native operating system and supports non-intrusive deployment. Background Technology

[0002] Existing AI behavior recognition and control solutions for smart terminals mainly suffer from the following core technological deficiencies: First, the control sequence is severely lagging. Existing solutions generally adopt a post-event identification mode, only identifying and judging AI behavior after the AI ​​application is started, computing power is called, and data interaction is completed. It is impossible to achieve source control before the AI ​​behavior is executed, making it difficult to prevent security risks caused by unauthorized AI calls and unauthorized AI behavior. Once a violation of AI operation occurs, the loss has already occurred, and the control has completely lost its practical meaning.

[0003] Secondly, the coverage of judgments is incomplete, and there are technical vulnerabilities that can be bypassed. Existing solutions do not achieve unified and mandatory end-to-end control over all human-computer interaction entry points on the terminal. There are many bypass channels. For example, hardware-triggered AI behavior can bypass software layer control, system-level AI services can bypass application layer judgment, and AI behavior triggered by cross-device peripherals can bypass local terminal control. This makes it easy for missed judgments and false judgments to occur, and the control link cannot form a complete closed loop. Some existing solutions have implemented input content filtering for specific AI applications, but they still cannot break away from the lagging logic of processing by the native system before control, and they have not built a single control channel covering all human-computer interaction entry points, so they cannot achieve source control of AI behavior in all scenarios.

[0004] Third, the judgment logic has blind spots and cannot adapt to AI interaction in all scenarios. Existing solutions are mostly designed for specific types of AI applications or hardware-triggered scenarios, and do not cover all scenarios such as system-level AI services, background linkage AI calls, multimodal AI interaction, offline AI inference, and hybrid triggered AI behavior. With the rapid iteration of AI technology, new AI interaction forms are very easy to get out of control, and the solution has extremely poor universality and scalability.

[0005] Fourth, the solutions lack compatibility and feasibility. Existing technologies are mostly tied to specific hardware and software environments, specific AI application scenarios, or specific system architectures, making them unable to adapt to full-scenario AI behavior control across all types of terminals. Moreover, most solutions require intrusive modifications to the native operating system, disrupting the original system ecosystem and resulting in extremely high costs for large-scale deployment and adaptation. Summary of the Invention

[0006] (a) Purpose of the invention In view of the shortcomings of the existing technology, the purpose of this invention is: 1. Construct a full-scale mandatory pre-judgment mechanism to ensure that all human-computer interaction commands received by the terminal are judged for AI behavior attributes before being processed by the native operating system. This fundamentally solves the core technical problems of post-event control, bypass risks, and missed or misjudged judgments in existing technologies, and achieves full-chain source control of AI behavior.

[0007] 2. Clearly define the protection boundaries of the core technical solution, with the pre-judgment and diversion control of full human-computer interaction as the core invention point. The independent claims only record the necessary technical features for solving the core technical problem, taking into account both maximizing the scope of protection and the complete feasibility of the technical solution, which meets the patent examination standards.

[0008] 3. Achieve full-scenario coverage of the judgment logic without blind spots, fill in the gaps in existing AI interaction scenarios and judgment methods, ensure that all types and all triggering sources of AI interaction behavior can be accurately identified without judgment blind spots, and adapt to all existing and future possible AI interaction forms.

[0009] 4. An authorized deployment mechanism is adopted, clearly defining the boundaries of rule initialization permissions. Compliant deployment permissions are granted to terminal original equipment manufacturers, which facilitates large-scale deployment and commercial operation by manufacturers. At the same time, the tampering permissions of unauthorized entities are strictly limited, taking into account both industry practicality and control security.

[0010] 5. Achieve ultimate compatibility and non-intrusive deployment of the solution, without modifying or reconstructing the underlying architecture of the native operating system, without interfering with the normal operation of the native system, adapting to all electronic terminals equipped with AI capabilities, and reducing the cost of large-scale deployment and adaptation.

[0011] (II) Core Technology Solution To achieve the above-mentioned objectives, the core technical solution provided by this invention is as follows: A method for pre-judgment of AI behavior in human-computer interaction is applied to electronic terminals equipped with AI capabilities. After the terminal is powered on, it continuously takes over all human-computer interaction entry points of the electronic terminal and constructs a unique instruction preprocessing channel. This instruction preprocessing channel is the only necessary path for all human-computer interaction instructions to enter the terminal's native operating system. The native operating system can only receive non-AI interaction instructions that have been diverted through the instruction preprocessing channel and has no right to directly receive human-computer interaction instructions that have not been pre-judged. For each human-computer interaction instruction received by the electronic terminal, the instruction is intercepted before the native operating system of the electronic terminal performs the entire process of receiving, parsing, scheduling, and executing the instruction. The intercepted human-computer interaction instructions are pre-judged to determine their AI behavior attributes, distinguishing whether the human-computer interaction instruction is an AI interaction behavior or a non-AI interaction behavior. Based on the pre-judgment result, a diversion operation is performed: if it is determined to be an AI interaction behavior, the instruction is imported into a dedicated AI control system for processing; if it is determined to be a non-AI interaction behavior, the instruction is allowed to be processed by the native operating system.

[0012] In this solution, the human-computer interaction instructions include all operation instructions pointing to the terminal that are triggered directly by the user, triggered by the application on behalf of the user, triggered by the peripheral device on behalf of the user, and triggered by cross-device linkage; the AI ​​interaction behavior refers to all interactive behaviors that trigger the terminal's AI computing power call, AI model inference, AI content generation, AI intelligent response, and AI data processing, regardless of their triggering source, running carrier, or online or offline status.

[0013] (III) Optimization of Technical Solutions As a supplement to the core technical solution, the preferred technical solution of the present invention is as follows: 1. The full range of human-computer interaction entry points includes, but is not limited to, terminal software interaction entry points, hardware trigger entry points, peripheral device interaction entry points, wired communication interaction entry points, wireless communication interaction entry points, and system-level service interaction entry points, covering all channels of the terminal that can receive human-computer interaction commands, with no blind spots in control.

[0014] 2. In the aforementioned pre-judgment stage, the triggering source of the intercepted human-computer interaction commands is first identified and strictly classified into four categories: software-triggered interaction, hardware-triggered interaction, system-level service-triggered interaction, and cross-device peripheral-triggered interaction. Then, independent AI behavior judgments are performed on each type of triggering interaction. The judgment logic of each type is completely independent, does not overlap, and does not interfere with each other. For interaction commands triggered by multiple sources, parallel judgment of all sources is performed. If any source meets the AI ​​interaction behavior judgment criteria, it is judged as an AI interaction behavior.

[0015] 3. The software-triggered interaction refers to human-computer interaction behavior triggered by an upper-layer software interface or a third-party application; it is determined to be a software-side AI interaction behavior if any of the following conditions are met: (1) Interactive behaviors are handled by AI applications, large model clients, and intelligent assistant software; (2) The interaction scenarios are multimodal intelligent question answering, command request, content generation, and recognition processing in the software, including text / voice / image / video. (3) The system marks this interaction as a software-side AI intelligent interaction scenario, distinguishing it from ordinary social, office, and call behaviors, etc., which are not AI software behaviors; (4) AI reasoning and data processing behaviors triggered in the background by the application and directly related to the user's previous human-computer interaction.

[0016] 4. The hardware-triggered interaction refers to a human-computer interaction behavior that is triggered independently by hardware without any upper-level software interface initiation or third-party application active acceptance; it is determined to be a hardware-side AI interaction behavior if any of the following conditions are met: (1) The interaction is directly triggered by the terminal's dedicated AI physical button and hardware wake-up module, without any software application actively undertaking it; (2) AI inference, calculation, recognition, and noise reduction processing behaviors independently initiated by local hardware modules (neural network processor, sensor, AI chip, image signal processor, baseband chip); (3) The system marks this interaction as a pure hardware-level AI scheduling and hardware intelligent perception scenario, without any active participation from upper-level software interaction; (4) Offline AI reasoning and recognition behavior executed independently by the local hardware module when there is no network connection.

[0017] 5. The system-level service-triggered interaction refers to human-computer interaction behavior triggered by system-level AI services built into the terminal's native operating system and not provided by third-party applications; it is determined to be a system-level AI interaction behavior if any of the following conditions are met: (1) The interaction is handled by the native system’s built-in AI input method, AI album, AI call noise reduction, AI subtitles, and AI voice assistant system services; (2) The interaction scenarios are system-level AI function calls, system-level AI computing power scheduling, and system service AI data processing; (3) The system marks this interaction as a system-level AI intelligent interaction scenario, which is different from ordinary system service behavior.

[0018] 6. The cross-device peripheral-triggered interaction refers to human-computer interaction behavior triggered by an external device that has completed a legal connection with the terminal and across terminal devices; it is determined to be a cross-device AI interaction behavior if any of the following conditions are met: (1) The interaction is directly triggered by Bluetooth peripherals, wired peripherals, smart home devices, vehicle terminals and wearable devices connected to the terminal, involving the invocation of AI capabilities; (2) Interactions include AI command requests, AI content generation, and AI recognition processing in cross-device screen projection, interconnection, and collaboration scenarios; (3) The system marks this interaction as a cross-device AI intelligent interaction scenario, which is different from ordinary cross-device data transmission and control behavior.

[0019] 7. When an AI interaction behavior is first identified as belonging to the corresponding category, an AI-specific directory for the corresponding category is independently created and fixed in the corresponding storage area of ​​the terminal according to the terminal's underlying preset rules. Among them, a dedicated AI software directory is created for software-side AI interaction behaviors, a dedicated AI hardware directory is created for hardware-side AI interaction behaviors, a dedicated AI system service directory is created for system-level AI interaction behaviors, and a dedicated AI cross-device directory is created for cross-device AI interaction behaviors.

[0020] 8. The directory creation rules and AI behavior judgment rules are globally unified preset rules at the terminal level. Only the original equipment manufacturers (OEMs) of the terminal are authorized to initialize the rules and write them to the underlying layer for compliant deployment. Apart from the authorized OEMs, third-party applications, hardware drivers, ordinary users and unauthorized partners have no right to modify, delete, reconstruct or bypass the execution of the rules.

[0021] 9. The root structure of various AI-specific directories is permanently locked immediately after creation. Only the dedicated AI management system has regular read, write, and management permissions. Authorized original equipment manufacturers can perform initial configuration within the preset rule framework. Other entities have no permission to modify, delete, or reconstruct. When AI interaction behavior is subsequently determined to be of the corresponding category again, the fixed corresponding dedicated directory will be reused directly without being recreated.

[0022] 10. This method is deployed non-intrusively throughout the entire process, without modifying, reconstructing, or interfering with the underlying architecture and normal operation of the terminal's native operating system. It is fully compatible with the terminal's native operating system and does not require customized adaptation for different system versions or hardware platforms.

[0023] 11. The pre-judgment unconditionally executes every human-computer interaction instruction received by the terminal, without any skipping or exceptions, ensuring that all human-computer interaction instructions are judged first and then the corresponding diversion operation is executed, without any post-judgment scenarios.

[0024] (iv) Beneficial effects 1. This invention constructs a closed-loop AI behavior source control system by continuously taking over all human-computer interaction entry points, building a unique instruction preprocessing channel, and intercepting and forcibly pre-judging instructions in advance. It fundamentally solves the core technical problems of post-event control, bypass risk, and missed or misjudged judgment in existing technologies, and realizes full-process, all-round control of all AI interaction behaviors.

[0025] 2. The core technical solution of this invention is not bound to specific judgment criteria, software and hardware environments, or application scenarios. The independent claims only record the necessary technical features for solving the core technical problem, and the scope of protection is clear and maximized. At the same time, through the layered optimization of technical solutions, a complete technical protection system is formed, which takes into account both the universality of the solution and the rigor of protection, and meets the patent examination standards.

[0026] 3. The judgment logic of this invention covers four major categories of triggering scenarios: software side, hardware side, system-level services, and cross-device peripherals. It supplements the judgment scenarios that existing technologies have omitted, such as multimodal AI interaction, offline AI inference, background linkage AI invocation, and hybrid triggering AI behavior. It has no judgment blind spots and can adapt to all existing and future AI interaction forms, with extremely strong versatility and scalability.

[0027] 4. This invention adopts an authorized deployment design, granting the original equipment manufacturer (OEM) the permission to write core rules, which facilitates the manufacturer's large-scale deployment and commercial operation. At the same time, it strictly restricts the tampering permissions of unauthorized entities, taking into account both industrial practicality and control security, and has extremely strong value for large-scale deployment.

[0028] 5. This invention is deployed non-intrusively throughout the entire process, without modifying the underlying architecture of the native operating system, without interfering with the normal operation of the native system and normal use by users, and is compatible with all electronic terminals equipped with AI capabilities. No customization is required, and the implementation cost is extremely low.

[0029] 6. This invention ensures the stability and security of the control system through the underlying fixed unified rules and the permanent locking mechanism of the directory structure, avoiding the risk of core rules being tampered with and control logic being bypassed, and improving the system's security protection capabilities. Detailed Implementation

[0031] The technical solution of the present invention will be further clearly and completely described below with reference to specific embodiments. The described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] Example 1 System Initialization Deployment The execution subject of this embodiment is a smartphone equipped with the native Android operating system. The phone has a built-in neural network processor chip, and the native system comes with AI input method, AI album system-level AI service, and a dedicated AI physical wake-up button. It supports Bluetooth peripherals and cross-device collaboration with vehicle-machine interconnection.

[0033] After the phone is powered on, the AI ​​control system corresponding to this method simultaneously completes a lightweight startup and enters a background resident running state. The terminal original equipment manufacturer completes the underlying initialization writing of the core judgment rules and directory creation rules as authorized. The control system continuously takes over all human-computer interaction entry points of the phone, including software interaction entry points, hardware trigger entry points, system-level service interaction entry points, and peripheral and cross-device interaction entry points, and builds a unique instruction preprocessing channel. This channel is the only necessary path for all human-computer interaction instructions to enter the Android native operating system. The native operating system can only receive non-AI interaction instructions after being diverted through this channel, and has no right to directly receive human-computer interaction instructions that have not been pre-judged. There are no bypass channels to bypass this.

[0034] The management system does not modify or reconstruct the underlying architecture of the Android native operating system during startup and operation, does not interfere with the normal startup and operation of the native system, occupies only a very small amount of computing power and memory resources of the phone, and does not affect the normal performance of the terminal.

[0035] Example 2: Implementation process of software-side AI interaction behavior Users open a third-party large model client application on their mobile phones, input a multimodal AI generation request with mixed text and images via the touchscreen, triggering human-computer interaction on the software side, and the mobile phone receives the interaction command.

[0036] Before the Android native operating system completes the entire process of receiving, parsing, scheduling, and executing the instruction, the control system intercepts the instruction and immediately performs a pre-judgment.

[0037] In the preliminary judgment stage, the control system first identifies that the triggering source of the instruction is software-side triggering, and then performs independent judgment of software-side AI behavior: the interaction behavior is handled by a third-party large model client application, belongs to the multimodal AI content generation scenario within the software, meets the judgment criteria of software-side AI interaction behavior, and is finally judged as AI interaction behavior.

[0038] The control system directly imports the instruction into the dedicated AI control system for subsequent full-link control; at the same time, it confirms that the instruction is the first software-side AI interaction behavior judged after this boot, and according to the underlying fixed and unmodifiable rules, it creates a dedicated AI software directory in the software-specific area stored inside the phone, and permanently locks the directory root structure, allowing only the dedicated AI control system to perform compliant read and write operations.

[0039] When users trigger AI interaction behaviors again through this application or other third-party software, the control system will perform a preliminary judgment and directly reuse the fixed AI software-specific directory without creating it again.

[0040] Example 3: Implementation process of hardware-side AI interaction behavior When a user is offline without a network connection, pressing the dedicated AI physical wake-up button on the phone and issuing an offline voice recognition command triggers hardware-side human-computer interaction without any software interface actively launching or third-party applications actively responding. The phone then receives the interaction command.

[0041] Before the Android native operating system completes the entire process of receiving, parsing, scheduling, and executing the instruction, the control system intercepts the instruction and immediately performs a pre-judgment.

[0042] In the preliminary judgment stage, the control system first identifies that the triggering source of the instruction is hardware-side triggering, and then performs independent judgment of hardware-side AI behavior: the interaction is directly triggered by the dedicated AI physical button, without any software application actively undertaking it, which is an offline AI voice recognition behavior independently executed by the local neural network processor chip, which meets the judgment criteria of hardware-side AI interaction behavior, and is finally judged as AI interaction behavior.

[0043] The control system directly imports the instruction into the dedicated AI control system for subsequent full-link control; at the same time, it confirms that the instruction is the first hardware-side AI interaction behavior determined after this boot, and according to the underlying fixed and unmodifiable rules, it creates a dedicated AI hardware directory in the dedicated area of ​​the phone's hardware storage control, and permanently locks the directory root structure, allowing only the dedicated AI control system to perform compliant read and write operations.

[0044] When users trigger AI interaction behaviors again through hardware modules such as AI physical buttons and local inference by built-in neural network processors, the management system will directly reuse the fixed AI hardware-specific catalog after completing the preliminary judgment, without creating it again.

[0045] Example 4: Implementation Process of System-Level AI Interaction Behavior When a user opens the phone's native built-in photo album application, the AI ​​album's image classification and scene recognition functions are triggered. Without the involvement of any third-party applications, a system-level service human-computer interaction is triggered, and the phone receives the interaction command.

[0046] Before the Android native operating system completes the entire process of receiving, parsing, scheduling, and executing the instruction, the control system intercepts the instruction and immediately performs a pre-judgment.

[0047] In the preliminary judgment stage, the control system first identifies that the triggering source of the instruction is a system-level service trigger, and then performs an independent judgment of system-level AI behavior: the interaction is handled by the native system's built-in AI album system service, which belongs to the system-level AI image recognition scenario and meets the judgment criteria of system-level AI interaction behavior, and is finally judged as an AI interaction behavior.

[0048] The control system directly imports the instruction into the dedicated AI control system for subsequent full-link control; at the same time, it confirms that the instruction is the first system-level AI interaction behavior judged after this boot, and according to the underlying fixed and unmodifiable rules, it creates a dedicated AI system service directory in the dedicated area of ​​the phone system service storage, and permanently locks the directory root structure, allowing only the dedicated AI control system to perform compliant read and write operations.

[0049] When users trigger system-level AI functions such as native system AI input method and AI call noise reduction again, the management system will perform a preliminary judgment and directly reuse the fixed AI system service exclusive directory without creating it again.

[0050] Example 5: Implementation process of cross-device peripheral AI interaction behavior Users can trigger the built-in AI voice assistant function of wireless smart earbuds by connecting to their mobile phones via Bluetooth, and initiate an AI question-and-answer request. The mobile phone software interface does not need to be actively launched; the mobile phone receives the interaction command through the wireless communication interface.

[0051] Before the Android native operating system completes the entire process of receiving, parsing, scheduling, and executing the instruction, the control system intercepts the instruction and immediately performs a pre-judgment.

[0052] In the preliminary judgment stage, the control system first identifies that the triggering source of the instruction is a cross-device peripheral trigger, and then performs an independent judgment of cross-device AI behavior: the interaction is directly triggered by a Bluetooth peripheral connected to the mobile phone, involves the call of AI voice question answering capabilities, meets the judgment criteria of cross-device AI interaction behavior, and is finally judged as AI interaction behavior.

[0053] The control system directly imports the instruction into the dedicated AI control system for subsequent full-link control; at the same time, it confirms that the instruction is the first cross-device AI interaction behavior judged after this boot, and according to the underlying fixed and unmodifiable rules, it creates an AI cross-device exclusive directory in the dedicated area of ​​the mobile phone's cross-device collaborative storage, and permanently locks the directory root structure, allowing only the dedicated AI control system to perform compliant read and write operations.

[0054] When users trigger AI interaction behaviors again through cross-device scenarios such as vehicle-machine interconnection and screen projection collaboration, the management system will directly reuse the fixed AI cross-device exclusive directory after completing the preliminary judgment, and will not create it again.

[0055] Example 6: Implementation process of hybrid-triggered AI interaction behavior The user first presses the phone's dedicated AI physical wake-up button, then enters a text AI request in the native system's smart assistant interface, simultaneously triggering human-computer interaction on both the hardware and software sides. The phone then receives this hybrid interactive command.

[0056] Before the Android native operating system completes the entire process of receiving, parsing, scheduling, and executing the instruction, the control system intercepts the instruction and immediately performs a pre-judgment.

[0057] In the preliminary judgment stage, the control system identifies the instruction as being triggered by multiple sources and simultaneously performs parallel and independent judgments on both the software and hardware sides. If both judgments meet the AI ​​interaction behavior judgment criteria, the instruction is ultimately judged as an AI interaction behavior.

[0058] The control system directly imports the instruction into the dedicated AI control system for subsequent end-to-end control, while reusing the already fixed dedicated AI software and AI hardware directories, eliminating the need for repeated creation.

[0059] Example 7: Implementation process of non-AI interactive behavior When users perform non-AI interactive behaviors such as making regular voice calls, chatting on social media, viewing local photos, browsing web pages, recording audio, and transferring files across devices, the mobile phone receives the corresponding human-computer interaction commands.

[0060] Before the Android native operating system completes the entire process of receiving, parsing, scheduling, and executing the instruction, the control system intercepts the instruction and immediately performs a pre-judgment.

[0061] Based on the preliminary assessment, the instruction did not meet the criteria for any AI interaction behavior and was ultimately determined to be a non-AI interaction behavior.

[0062] The control system immediately allows the instruction to be sent directly to the Android native operating system, which then processes it normally according to the original process. The control system does not make any additional intervention, and the entire process does not affect the user's normal use.

Claims

1. A method for pre-judgment of AI behavior for human-computer interaction, applied to electronic terminals equipped with AI capabilities, characterized in that, After the terminal is powered on, it continuously takes over all human-computer interaction entry points of the electronic terminal, constructing a unique instruction preprocessing channel. This instruction preprocessing channel is the only necessary path for all human-computer interaction instructions to enter the terminal's native operating system. The native operating system can only receive non-AI interaction instructions after being diverted through the instruction preprocessing channel and has no right to directly receive human-computer interaction instructions that have not undergone pre-judgment. For each human-computer interaction instruction received by the electronic terminal, the instruction is intercepted before the electronic terminal's native operating system performs the entire process of receiving, parsing, scheduling, and executing the instruction. The intercepted human-computer interaction instructions undergo pre-judgment of AI behavior attributes to distinguish whether the human-computer interaction instruction is an AI interaction behavior or a non-AI interaction behavior. Based on the pre-judgment result, a diversion operation is performed: if it is determined to be an AI interaction behavior, the instruction is imported into a dedicated AI control system for processing. If the behavior is determined to be non-AI interaction, the instruction will be allowed to be processed by the native operating system.

2. The method according to claim 1, characterized in that, The comprehensive human-computer interaction entry points include software interaction entry points, hardware trigger entry points, peripheral device interaction entry points, wired communication interaction entry points, wireless communication interaction entry points, and system-level service interaction entry points, covering all channels on the terminal that can receive human-computer interaction commands. The human-computer interaction commands include all operation commands pointing to the terminal that are triggered directly by the user, triggered by the application on behalf of the user, triggered by the peripheral device on behalf of the user, and triggered by cross-device linkage. The AI ​​interaction behavior refers to all interactive behaviors that trigger the terminal's AI computing power call, AI model inference, AI content generation, AI intelligent response, and AI data processing, regardless of their triggering source, operating platform, or online or offline status.

3. The method according to claim 1, characterized in that, The pre-judgment process first identifies the triggering source of the intercepted human-computer interaction commands, strictly classifying them into four categories: software-triggered interaction, hardware-triggered interaction, system-level service-triggered interaction, and cross-device peripheral-triggered interaction. Then, independent AI behavior judgments are performed on each type of triggering interaction. The judgment logic of each type is completely independent, does not overlap, and does not interfere with each other. For interaction commands triggered by multiple sources, parallel judgment of all sources is performed. If any source meets the AI ​​interaction behavior judgment criteria, it is judged as an AI interaction behavior.

4. The method according to claim 3, characterized in that, The software-triggered interaction refers to human-computer interaction behavior triggered by an upper-layer software interface or a third-party application; it is determined to be a software-side AI interaction behavior if any of the following conditions are met: (1) Interactive behaviors are handled by AI applications, large model clients, and intelligent assistant software; (2) The interaction scenarios are multimodal intelligent question answering, command request, content generation, and recognition processing in the software, including text / voice / image / video. (3) The system marks this interaction as a software-side AI intelligent interaction scenario, distinguishing it from ordinary non-AI software behavior; (4) AI reasoning and data processing behaviors triggered in the background by the application and directly related to the user's previous human-computer interaction.

5. The method according to claim 3, characterized in that, The hardware-triggered interaction is a human-computer interaction behavior that is triggered independently by hardware without any upper-level software interface or third-party application. It is determined to be a hardware-side AI interaction behavior if it meets any of the following conditions: (1) The interaction is directly triggered by the terminal's dedicated AI physical button and hardware wake-up module, without any software application actively undertaking it; (2) AI inference, computation, recognition, and noise reduction processing behaviors initiated independently by the local hardware module; (3) The system marks this interaction as a pure hardware-level AI scheduling and hardware intelligent perception scenario, without any active participation from upper-level software interaction; (4) Offline AI reasoning and recognition behavior executed independently by the local hardware module when there is no network connection.

6. The method according to claim 3, characterized in that, The system-level service-triggered interaction refers to the human-computer interaction behavior triggered by the system-level AI service built into the terminal's native operating system. The cross-device peripheral-triggered interaction refers to the human-computer interaction behavior triggered by external devices that have completed a legal connection with the terminal and cross-terminal devices. If any of the judgment conditions of the corresponding category are met, it is judged as the AI ​​interaction behavior of the corresponding category.

7. The method according to claim 3, characterized in that, When an AI interaction behavior is first identified as belonging to the corresponding category, an AI-specific directory for the corresponding category is independently created and fixed in the corresponding storage area of ​​the terminal according to the terminal's underlying preset rules. Among them, a dedicated AI software directory is created for software-side AI interaction behaviors, a dedicated AI hardware directory is created for hardware-side AI interaction behaviors, a dedicated AI system service directory is created for system-level AI interaction behaviors, and a dedicated AI cross-device directory is created for cross-device AI interaction behaviors.

8. The method according to claim 7, characterized in that, The directory creation rules and AI behavior judgment rules are globally unified preset rules at the terminal level. Only authorized terminal original equipment manufacturers (OEMs) have the compliant deployment permissions for rule initialization and underlying writing. Other unauthorized entities have no permission to modify, delete, reconstruct, or bypass execution. Once the root structure of each type of AI-specific directory is created, it is permanently locked immediately. Only the dedicated AI management system has regular read, write, and management permissions. Authorized OEMs can perform initialization configuration within the preset rule framework. Other entities have no permission to change it. When the AI ​​interaction behavior is subsequently judged to be of the corresponding category, the fixed corresponding dedicated directory is directly reused and no longer created.

9. The method according to claim 1, characterized in that, This method involves non-intrusive deployment throughout the entire process, without modifying, refactoring, or interfering with the underlying architecture and normal operation of the terminal's native operating system, and is fully compatible with the terminal's native operating system.

10. The method according to claim 1, characterized in that, The pre-judgment process executes every human-computer interaction command received by the terminal unconditionally, without any skipping or exceptions, ensuring that all human-computer interaction commands are judged before the corresponding routing operation is executed.