Human work assisting method and system based on artificial intelligence

Abstract

The invention relates to a human work assisting method and system based on artificial intelligence. The method comprises the following steps that multi-modal data input by a user is obtained, preliminary task disassembling is carried out after preprocessing, then logic dependence intensity, granularity adaptation degree and disassembling priority are calculated, task disassembling optimization is carried out, and a sub-task list is generated; generating a human capability vector and an artificial intelligence capability vector based on the capability portrait, generating a sub-task skill demand vector based on the sub-task list, calculating an allocation utility function constructed by a skill adaptation degree, a load penalty item and a risk award item, and determining the skill adaptation degree of the sub-task list under the constraint of satisfying an allocation decision rule. Performing man-machine task allocation on each sub-task in the sub-task list; in the execution process of the auxiliary work task, the execution progress and the execution quality of the task are monitored in real time, and man-machine task allocation is dynamically adjusted. Compared with the prior art, the method has the advantages that effective disassembly and allocation of complex tasks can be realized, and the artificial intelligence auxiliary efficiency is further improved.

Description

Technical Field

[0001] This invention relates to the field of AI-assisted application technology, and in particular to a method and system for assisting human work based on artificial intelligence. Background Technology

[0002] Currently, various AI-assisted tools exist both domestically and internationally, covering multiple fields such as office work, healthcare, education, and industrial manufacturing. For example, intelligent document processing tools in office scenarios (such as Adobe Acrobat's AI text recognition and editing functions) can automatically extract key information from documents; AI-assisted diagnostic systems in the medical field (such as Baidu Medical Brain) can preliminarily identify lesions based on medical images; AI quality inspection equipment in industrial manufacturing can detect surface defects in products through image recognition; and AI tutoring software in the education field (such as Yuanfudao's AI Q&A system) can provide students with question analysis and knowledge point explanations. In addition, there are general-purpose AI assistants (such as ChatGPT and iFlytek Xinghuo) that can provide basic auxiliary services such as text generation, information retrieval, and logical reasoning.

[0003] However, existing tools can only handle single-stage, low-logically-connected tasks, and cannot cope with multi-stage, highly logically complex workflows. For example, in project management scenarios, existing AI tools can remind users of task deadlines, but cannot automatically adjust task allocation based on team members' workloads and task dependencies (e.g., task B cannot start if task A is not completed). In the legal field, AI text tools can retrieve legal provisions, but cannot combine them with the evidence chain and key points of contention in a specific case to generate a targeted defense strategy framework. This is essentially because AI struggles to effectively break down complex tasks, and therefore struggles to effectively process them.

[0004] Chinese patent CN120996543A discloses a human-machine co-creation enterprise business process management system to address the problems of unreasonable task allocation, insufficient process monitoring, and difficulty in knowledge reuse in traditional systems. The system comprises six modules: task triggering and parsing, task decomposition and planning, workflow generation and optimization, etc. The task triggering and parsing module acquires and parses tasks, decomposes tasks using various strategies, determines the workflow through multi-objective optimization, coordinates human and machine execution of sub-tasks, monitors the process in real time and dynamically adjusts it, and simultaneously collects business data to optimize AI capabilities and accumulate enterprise knowledge. However, this method's task decomposition is mechanical and lacks quantitative logic, allocating tasks solely based on single-dimensional efficiency indicators, making it unable to accurately adapt to tasks with high logical complexity. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a human work assistance method and system based on artificial intelligence.

[0006] The objective of this invention can be achieved through the following technical solutions: According to a first aspect of the present invention, an artificial intelligence-based method for assisting human work is provided, the method comprising the following steps: Acquire multimodal data input by the user, preprocess it, and generate structured task data; The structured task data is initially decomposed to generate an initial set of subtasks. The logical dependency strength, granularity adaptability and decomposition priority of each task in the initial set of subtasks are calculated. The task decomposition is optimized to generate a list of subtasks. Human-machine capability profiles are quantified, generating human capability vectors and artificial intelligence capability vectors respectively. Subtask skill requirement vectors are generated based on the subtask list. Skill fit is calculated based on the matching degree between the capability vectors and the subtask skill requirement vectors. An allocation utility function is constructed by combining the skill fit, load penalty term, and risk reward term. Based on the allocation utility function, human-machine task allocation is performed on each subtask in the subtask list under the constraints of the allocation decision rules. During the execution of auxiliary tasks, the progress and quality of task execution are monitored in real time, and the allocation of human and machine tasks is dynamically adjusted.

[0007] The calculation of the logical dependency strength includes the following steps: For any subtask in the initial subtask set Extract its direct predecessor subtasks from the initial subtask set and determine the number of predecessor subtasks. ; Construct a Bayesian network, and calculate subtasks based on the dependency execution data of similar historical tasks and the preceding subtasks. Dependency trigger probability ; Based on the number of preceding subtasks and the probability of triggering dependencies Calculate the strength of logical dependencies: , in, For subtasks The strength of logical dependencies This represents the total number of subtasks in the initial subtask set.

[0008] The calculation of the particle size fit includes the following steps: Based on historical task data, K-means clustering is used to obtain the time intervals that a single role can efficiently complete, and the median of these time intervals is taken as the optimal time threshold. ; Calculate the estimated execution time of subtasks based on historical execution time data of similar tasks. ; Granularity fit is calculated based on the estimated time consumption of subtasks and the optimal time consumption threshold: , in, For subtasks Granularity fit.

[0009] The calculation method for the disassembly priority is as follows: , in, , As weight, For subtasks The strength of logical dependencies For subtasks Granularity fit For subtasks Disassembly priority.

[0010] The task decomposition and optimization specifically includes the following steps: When the decomposition priority is greater than or equal to the preset priority threshold, the subtask corresponding to that decomposition priority will not be adjusted. When the decomposition priority is less than the preset priority threshold, the relationship between the corresponding thresholds of logical dependency strength and granularity fit is further determined: If the logical dependency strength is less than the first preset threshold but the granularity fit is greater than or equal to the second preset threshold, merge similar subtasks, recalculate the granularity fit and decomposition priority, or adjust the task dependency relationship and recalculate the logical dependency strength and decomposition priority. If the logical dependency strength is greater than or equal to the first preset threshold but the granularity fit is less than the second preset threshold, granularity calibration is performed. When the granularity is too coarse, the subtask is split a second time. When the granularity is too fine, similar subtasks are merged, and the logical dependency strength, granularity fit, and splitting priority are recalculated. If the logical dependency strength is less than the first preset threshold and the granularity fit is less than the second preset threshold, reconstruct the subtask, delete redundant subtasks, and recalculate the logical dependency strength, granularity fit, and decomposition priority. If the logical dependency strength is greater than or equal to the first preset threshold and the granularity fit is greater than or equal to the second preset threshold, the corresponding subtask will not be adjusted.

[0011] The calculation of skill suitability includes the following steps: Human-machine capability profiles are quantified, and human capability vectors and artificial intelligence capability vectors are generated respectively. Each element in the human capability vector and artificial intelligence capability vector represents the skill value of the human and artificial intelligence for a certain skill, respectively. For each subtask in the subtask list, mark the core skill requirements and generate a subtask skill requirement vector. Each element in the subtask skill requirement vector is a skill requirement indicator value. A skill requirement indicator value of 1 indicates that the subtask requires this type of skill, and a skill requirement indicator value of 0 indicates that the subtask does not require this type of skill. After normalizing the human ability vector and the artificial intelligence ability vector based on skill type, the skill probability corresponding to each skill type is calculated. Based on the skill probabilities, the skill importance weights are calculated using information entropy. The normalized skill importance weight is obtained by multiplying the skill importance weight by the skill requirement indicator value of the corresponding skill in the subtask skill requirement vector. The normalized skill importance weights are multiplied and summed with the corresponding skills in the human ability vector and the artificial intelligence ability vector, respectively, to obtain the human skill fit and the artificial intelligence skill fit.

[0012] The allocation utility function is expressed as follows: , in, Indicates whether the task is performed by a human or artificial intelligence; Subtasks The implementing body is Skill suitability at the time; , Indicates the adjustment coefficient; Indicates the implementing entity The load, For load penalty items; For subtasks The implementing body is The risk factor at any given time is determined based on the corresponding risk level. This is a risk-based reward item.

[0013] The allocation decision rules include: Basic rule: Select the entity with the largest allocation utility function value as the execution role; Constraint rule: When the risk level is greater than the preset level, if the subject with the largest allocation utility function value is artificial intelligence, it will be forcibly converted to a human.

[0014] The aforementioned real-time monitoring of task execution progress and quality, and dynamic adjustment of human-machine task allocation, specifically includes the following steps: During task execution, progress and quality indicators are calculated in real time, and the progress-quality coupling coefficient is calculated based on the progress and quality indicators. When the progress-quality coupling coefficient is greater than or equal to the adjustment threshold, the human ability vector and the artificial intelligence ability vector are updated, and the human-machine task allocation is redistributed. The adjustment threshold is determined based on the risk level.

[0015] According to a second aspect of the present invention, an artificial intelligence-based human work assistance system is provided, the system comprising: Data acquisition and preprocessing module: used to acquire multimodal data input by the user, and generate structured task data after preprocessing; Task decomposition module: used to perform preliminary task decomposition on structured task data, generate an initial subtask set, calculate the logical dependency strength, granularity adaptability and decomposition priority of each task in the initial subtask set, optimize task decomposition, and generate a subtask list. Task allocation module: used to quantify human-machine ability profiles, generate human ability vectors and artificial intelligence ability vectors respectively, generate sub-task skill requirement vectors based on the sub-task list, calculate skill fit based on the matching degree between the ability vector and the sub-task skill requirement vector, construct allocation utility function by combining skill fit, load penalty term and risk reward term, and allocate human-machine tasks to each sub-task in the sub-task list based on the allocation utility function and under the constraints of allocation decision rules; Real-time adjustment module: Used to monitor the progress and quality of tasks in real time during the execution of auxiliary tasks, and dynamically adjust the human-machine task allocation.

[0016] According to a third aspect of the present invention, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the program to implement the method described thereon.

[0017] According to a fourth aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the method described thereon.

[0018] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention achieves logical closed loop and granular controllable task decomposition by quantitatively calculating the logical dependency strength and granularity adaptability, which can improve the decomposition accuracy and directly adapt to multi-stage, high logical complexity tasks such as financial auditing and R&D projects, avoiding logical breakage and granularity imbalance.

[0019] (2) This invention introduces the entropy weight method to objectively calculate skill weights, and integrates skill suitability, load penalty and risk reward through the allocation utility function, thereby improving the accuracy of task allocation, effectively controlling human load and AI resource utilization, and avoiding advantage mismatch and resource waste.

[0020] (3) This invention determines the indicator value for triggering adjustment by real-time monitoring of progress and quality, and dynamically determines the trigger threshold according to the risk level. It can take into account risks in practical applications, make task adjustments in advance, reduce task delay rate and quality failure rate, cope with dynamic changes of highly complex tasks (such as sub-task quality abnormalities and dependency delays), and ensure the stability of the process loop.

[0021] (4) The present invention realizes the full-process data-driven iteration of disassembly-allocation-adjustment-optimization. The system adaptability is continuously improved with the execution of tasks, and the processing efficiency of similar tasks is iteratively improved. It can adapt to changes in business scenarios without manual intervention. Attached Figure Description

[0022] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a flowchart illustrating the task breakdown of the present invention; Figure 3 This is a flowchart of the task allocation process of the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0024] Unless otherwise defined, the technical or scientific terms used in this application shall have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms “a,” “an,” “an,” “the,” and similar words used in this application do not indicate quantity limitation and may indicate singular or plural. The terms “comprising,” “including,” “having,” and any variations thereof used in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or modules (units) is not limited to the listed steps or units, but may also include steps or units not listed, or may include other steps or units inherent to these processes, methods, products, or devices. The terms “connected,” “linked,” “coupled,” and similar words used in this application are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. “Multiple” used in this application refers to two or more. “And / or” describes the relationship between related objects, indicating that three relationships may exist; for example, “A and / or B” can represent: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following objects are in an "or" relationship. The terms "first," "second," and "third" used in this application are merely to distinguish similar objects and do not represent a specific ordering of the objects.

[0025] This embodiment provides a human work assistance method based on artificial intelligence, such as... Figure 1 As shown, the method includes the following steps: S1: Acquire multimodal data input by the user, preprocess it, and generate structured task data.

[0026] In this embodiment, the multimodal data input by the user can be multimodal data such as text, speech, image and video. After being encoded and processed by existing text, speech, image or video processing models, the semantics are aligned, and the core elements related to the task are extracted, including the task objective T, constraints C and output results O, etc., and structured task data is output.

[0027] S2 performs preliminary task decomposition on the structured task data, generates an initial subtask set, calculates the logical dependency strength, granularity adaptability, and decomposition priority of each task in the initial subtask set, optimizes the task decomposition, and generates a subtask list.

[0028] like Figure 2 As shown, it includes the following steps: S21, Initial subtask set generation.

[0029] Using structured task data as input, based on a large model Prompt: "with <t>For the goal, according to <c>Constraints are broken down into executable subtasks, each of which must explicitly output a result, to generate an initial set of subtasks. , This represents the total number of subtasks in the initial subtask set.

[0030] Taking the assistance provided during the audit process as an example, Table 1 shows an example of the initial set of subtasks.

[0031] Table 1 S22, Calculation of logical dependency strength.

[0032] S221, for any subtask in the initial subtask set Extract its direct predecessor subtasks from the initial subtask set and determine the number of predecessor subtasks. ; In this context, a preceding subtask refers to a subtask that must be completed before the current subtask can be started, as shown in Table 1. The preceding subtask is , , The preceding subtask is and , .

[0033] S222, Construct a Bayesian network, with subtasks as nodes and edges as dependencies. Build a conditional probability table based on historical dependency execution data of similar tasks, and calculate the subtask's dependencies based on the preceding subtasks. Dependency trigger probability ; For example, The preceding subtask is and The completion rates are 97% and 95% respectively, then .

[0034] S223, based on the number of preceding subtasks and the probability of triggering dependencies Calculate the strength of logical dependencies: , in, For subtasks The logical dependency strength is indicated by a higher value, signifying a stronger logical connection, and thus requiring priority to ensure dependency closure. For example, .

[0035] S23, Particle size fit calculation.

[0036] S231, Based on historical task data, K-means clustering is used to obtain the time intervals that a single role can efficiently complete, and the median of the time intervals is taken as the optimal time threshold. ; S232, Calculate the estimated time consumption of subtasks based on historical execution time data of similar tasks. The specific estimation method is a common practice in this field and will not be described in detail in this embodiment. The estimation result in one embodiment is shown in the "Estimated Time" column in Table 1.

[0037] S233, Calculate granularity fit based on subtask estimated time and optimal time threshold: , in, For subtasks Granularity fit.

[0038] For example, the one determined in this embodiment Hour, .

[0039] S24, Disassembly priority calculation: , in, , The weights can be dynamically adjusted based on the task type. , For subtasks Disassembly priority.

[0040] In this embodiment, since the application scenario is financial auditing, the focus is on high logical dependencies, therefore , ,but .

[0041] S25, task breakdown optimization.

[0042] 1. When disassembly priority When the priority is greater than or equal to the preset priority threshold (set to 0.25 in this embodiment), the subtasks corresponding to the decomposition priority will not be adjusted; 2. When disassembly priority When the priority level is less than the preset threshold, it indicates that the corresponding subtask is prone to logical breaks or low execution efficiency, requiring further assessment of the logical dependency strength. and particle size fit The relationship between the corresponding thresholds: 21. If the logical dependency strength is less than the first preset threshold but the granularity fit is greater than or equal to the second preset threshold, it indicates that the subtask has a weak logical correlation with the overall task, but its own granularity is appropriate (it can be executed independently). There is no need to adjust the granularity, only to enhance the correlation or merge: Adjustment scheme 1: Merge similar subtasks, recalculate the granularity fit and decomposition priority; Adjustment scheme 2: Reanalyze the potential logical correlation of subtasks, adjust the task dependency relationship, update the number of preceding subtasks and the dependency trigger probability, and recalculate the logical dependency strength and decomposition priority.

[0043] 22. If the logical dependency strength is greater than or equal to the first preset threshold but the granularity fit is less than the second preset threshold, it indicates that the subtask is closely related to the overall logic (cannot be deleted), but the granularity is too coarse or too fine. Therefore, granularity calibration is performed. When the granularity is too coarse, the subtask is split a second time. When the granularity is too fine, similar subtasks are merged, and the logical dependency strength, granularity fit, and splitting priority are recalculated.

[0044] 23. If the logical dependency strength is less than the first preset threshold and the granularity fit is less than the second preset threshold, it indicates that the subtasks have no strong logical relationship and the granularity is not suitable. They are mostly redundant or have unclear goals. In this case, the subtasks should be reconstructed. The reconstruction methods include, but are not limited to, merging and splitting subtasks, adjusting task dependencies, etc., and deleting redundant subtasks. The logical dependency strength, granularity fit and decomposition priority should be recalculated.

[0045] 24. If the logical dependency strength is greater than or equal to the first preset threshold and the granularity fit is greater than or equal to the second preset threshold, no adjustment will be made to the corresponding subtask.

[0046] S3. Perform human-machine capability profiling and quantification, generating human capability vectors and artificial intelligence capability vectors respectively. Generate sub-task skill requirement vectors based on the sub-task list. Calculate skill fit based on the matching degree between the capability vectors and the sub-task skill requirement vectors. Construct an allocation utility function by combining skill fit, load penalty term, and risk reward term. Based on the allocation utility function, and under the constraints of the allocation decision rules, allocate human-machine tasks to each sub-task in the sub-task list.

[0047] like Figure 3 As shown, it includes the following steps: S31, Skill Adaptability Calculation.

[0048] S311, perform human-machine capability profiling and quantification, and generate human capability vectors respectively. and artificial intelligence capability vector ,in, For the number of skill types, For humanity's understanding of the first Skill value of a skill type For artificial intelligence to the first Skill value for a skill type; S312, for each subtask in the subtask list Label core skill requirements and generate sub-task skill requirement vectors. ,in, This is the skill requirement indicator value. =1 indicates a subtask The first one is needed Skills =0 indicates a subtask No need for the first Skills; S313: After normalizing the human ability vector and the artificial intelligence ability vector based on skill type, calculate the skill probability corresponding to each skill type. ,in, Indicates the first The first subject (human / artificial intelligence) Classified normalized skill values; S314, based on the skill probability Skill importance weights are calculated based on information entropy. : , , in, This refers to information entropy.

[0049] S315, assign the importance weight of the skills Corresponding to the first subtask skill requirement vector Skill requirement indicator value for a skill class Multiply by this to obtain the normalized skill importance weights. ; S316, based on the normalized skill importance weights and the human ability vector and artificial intelligence capability vector Multiplying the elements in the table, we obtain the human skill fit and the artificial intelligence skill fit: , Represents human ability vector or artificial intelligence capability vector Corresponding skills The element value.

[0050] S32, construct the allocation utility function.

[0051] , in, Indicates whether the task is performed by a human or artificial intelligence; Subtasks The implementing body is Skill suitability at the time; , Indicates the adjustment coefficient; Indicates the implementing entity The load is defined by the ratio of the estimated total time of the currently allocated subtasks to the maximum load time of the main task. For load penalty items; For subtasks The implementing body is The risk factor at any given time is determined based on the corresponding risk level. This is a risk-based reward system. In this embodiment, the risk level is divided into 5 levels, defined as low, low-medium, medium, medium-high, and extremely high risk. Humans and AI have corresponding pre-set risk coefficients for different risk levels, ensuring that higher rewards are given to humans participating in high-risk tasks.

[0052] S33 performs human-machine task allocation.

[0053] In this embodiment, the allocation decision rules include: Basic rule: Select the entity with the largest allocation utility function value as the execution role; Constraint rule: When the risk level is greater than the preset level, if the subject with the largest allocation utility function value is artificial intelligence, it will be forcibly converted to a human.

[0054] Based on the above allocation decision rules, human and machine tasks are allocated.

[0055] S4 monitors the progress and quality of tasks in real time during the execution of auxiliary tasks, and dynamically adjusts the allocation of human and machine tasks.

[0056] S41, During the task execution process, the progress indicators and quality indicators are calculated in real time, and the progress-quality coupling coefficient is calculated based on the progress indicators and quality indicators.

[0057] The schedule metric is defined as the ratio of time elapsed to the estimated total time. The quality metric is defined as follows: when the executor is AI, it is defined as the execution accuracy rate; when the executor is human, it is defined as the review pass rate. Both the schedule metric and the quality metric have values ​​between 0 and 1. The schedule metric is divided by the sum of the quality metric and the performance metric (to prevent the denominator from being 0) to obtain the schedule-quality coupling coefficient.

[0058] S42, when the progress-quality coupling coefficient is greater than or equal to the adjustment threshold, update the human ability vector and the artificial intelligence ability vector, and reassign human and machine tasks. The adjustment threshold is determined based on the risk level. If the risk level is high, the adjustment threshold is correspondingly smaller to better cope with the risk.

[0059] This embodiment also provides an artificial intelligence-based human work assistance system, which includes: Data acquisition and preprocessing module: used to acquire multimodal data input by the user, and generate structured task data after preprocessing; Task decomposition module: used to perform preliminary task decomposition on structured task data, generate an initial subtask set, calculate the logical dependency strength, granularity adaptability and decomposition priority of each task in the initial subtask set, optimize task decomposition, and generate a subtask list. Task allocation module: used to quantify human-machine ability profiles, generate human ability vectors and artificial intelligence ability vectors respectively, generate sub-task skill requirement vectors based on the sub-task list, calculate skill fit based on the matching degree between the ability vector and the sub-task skill requirement vector, construct allocation utility function by combining skill fit, load penalty term and risk reward term, and allocate human-machine tasks to each sub-task in the sub-task list based on the allocation utility function and under the constraints of allocation decision rules; Real-time adjustment module: Used to monitor the progress and quality of tasks in real time during the execution of auxiliary tasks, and dynamically adjust the human-machine task allocation.

[0060] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the described module can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0061] In one embodiment, the system may consist of a perception and interaction layer, a core processing layer, and an application adaptation layer, wherein: The perception and interaction layer consists of a multimodal information acquisition module (camera, microphone, keyboard / mouse operation capture unit, external device data interface) and a personalized interaction module (voice interaction unit, text interaction unit, operation habit learning unit).

[0062] Function 1: The multimodal information acquisition module can collect visual information (such as actions of operating a computer screen and viewing document content), voice information (such as work needs communicated with colleagues), and equipment data (such as data from connected industrial sensors and medical testing equipment) in real time during the human work process, realizing comprehensive work information input.

[0063] Function 2: Personalized Interaction Module. Through the operation habit learning unit, it records the user's daily interaction methods with the system (such as preference for voice commands or text commands, and the location of frequently used operation buttons), and adjusts the interaction response logic (such as prioritizing the voice recognition function for users who prefer voice interaction; fixing the "Quick Filter" button in a prominent position on the interface for users who frequently use it).

[0064] In addition to the aforementioned multimodal information acquisition module, a brain-computer interface adapter unit can be added to support connection to brain-computer devices and directly read the user's simple work intentions (such as "start the quality inspection program" or "pause data acquisition"). This is suitable for scenarios where both hands are occupied (such as operators on industrial assembly lines).

[0065] The core processing layer consists of a general capability engine (natural language understanding unit, computer vision unit, and logical reasoning unit), a scene transfer engine (multi-scene model library and lightweight training unit), and a collaborative decision-making engine (task decomposition unit, human-machine division of labor unit, and dynamic adjustment unit).

[0066] Function 1: The general capability engine provides basic AI capability support. For example, the natural language understanding unit can parse the work requirement text input by the user (such as "compile this week's project progress report") and extract the core task objectives; the computer vision unit can recognize product images in industrial quality inspection and image data in medical scenarios; the logical reasoning unit can analyze the dependencies between tasks (such as "writing a report requires collecting data from various departments first, and then statistically analyzing key indicators").

[0067] Function 2: The scene transfer engine solves the scene adaptability problem. The multi-scene model library pre-stores basic models of various fields (such as basic models of industrial quality inspection, medical diagnosis, and office document processing). When switching scenes, the lightweight training unit only needs to input a small amount of scene-specific data (such as 10-20 defect sample images of a new product) to fine-tune the basic model without retraining, thus achieving rapid scene transfer.

[0068] Function 3: The collaborative decision engine enables deep human-machine collaboration. The task decomposition unit breaks down complex work tasks (such as "completing the entire process of quality inspection of automotive parts") into multiple sub-steps (such as "part size inspection → surface defect identification → assembly gap measurement"). The human-machine division of labor unit assigns work roles based on the difficulty of the sub-steps and the advantages of human capabilities (such as "AI performs size inspection and defect identification, while humans are responsible for manual verification of complex assembly gaps"). The dynamic adjustment unit monitors the task progress in real time.

[0069] Application adaptation layer: consists of domain application modules (industrial auxiliary sub-module, medical auxiliary sub-module, office auxiliary sub-module, education auxiliary sub-module, etc.) and personalized configuration modules.

[0070] Function 1: Domain Application Modules transform the capabilities of the core processing layer into specific application functions for different work scenarios in different fields. For example, the Industrial Assistance Submodule provides full-process assistance from "defect identification → data statistics → report generation", while the Medical Assistance Submodule provides auxiliary functions from "image analysis → medical history integration → diagnostic suggestion generation".

[0071] Function 2: The personalized configuration module adjusts the output format of application functions according to the differences in user abilities and work habits. For example, it provides an "rigorous text generation" mode (more professional terminology and clear logical structure) for academic paper authors, an "active text generation" mode (increases the proportion of popular online words and short sentences) for marketing copywriters, "detailed knowledge point explanation" (step-by-step analysis of formula derivation process) for students with weak learning foundations, and "in-depth expansion explanation" (connecting interdisciplinary knowledge points) for students with extra learning capacity.

[0072] In one embodiment, the electronic device includes a computing unit that can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) or a computer program loaded from a storage unit into a random access memory (RAM). The RAM may also store various programs and data required for device operation. The computing unit, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0073] Multiple components in an electronic device are connected to an I / O interface, including: input units such as keyboards and mice; output units such as various types of displays and speakers; storage units such as disks and optical discs; and communication units such as network interface cards (NICs), modems, and wireless transceivers. The communication unit allows the device to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0074] The computing unit can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of computing units include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit performs the various methods and processes described above, such as methods S1-S4. For example, in some embodiments, methods S1-S4 may be implemented as computer software programs tangibly contained in a machine-readable medium, such as a storage unit. In some embodiments, part or all of the computer program may be loaded and / or installed on the device via ROM and / or a communication unit. When the computer program is loaded into RAM and executed by the computing unit, one or more steps of methods S1-S4 described above may be performed. Alternatively, in other embodiments, the computing unit may be configured to perform methods S1-S4 by any other suitable means (e.g., by means of firmware).

[0075] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0076] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0077] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0078] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.< / c> < / t>

Claims

1. A method for assisting human work based on artificial intelligence, characterized in that, The method includes the following steps: Acquire multimodal data input by the user, preprocess it, and generate structured task data; The structured task data is initially decomposed to generate an initial set of subtasks. The logical dependency strength, granularity adaptability and decomposition priority of each task in the initial set of subtasks are calculated. The task decomposition is optimized to generate a list of subtasks. Human-machine capability profiles are quantified, generating human capability vectors and artificial intelligence capability vectors respectively. Subtask skill requirement vectors are generated based on the subtask list. Skill fit is calculated based on the matching degree between the capability vectors and the subtask skill requirement vectors. An allocation utility function is constructed by combining the skill fit, load penalty term, and risk reward term. Based on the allocation utility function, human-machine task allocation is performed on each subtask in the subtask list under the constraints of the allocation decision rules. During the execution of auxiliary tasks, the progress and quality of task execution are monitored in real time, and the allocation of human and machine tasks is dynamically adjusted.

2. The artificial intelligence-based human work assistance method according to claim 1, characterized in that, The calculation of the logical dependency strength includes the following steps: For any subtask in the initial subtask set Extract its direct predecessor subtasks from the initial subtask set and determine the number of predecessor subtasks. ; Construct a Bayesian network, and calculate subtasks based on the dependency execution data of similar historical tasks and the preceding subtasks. Dependency trigger probability ; Based on the number of preceding subtasks and the probability of triggering dependencies Calculate the strength of logical dependencies: ， in, For subtasks The strength of logical dependencies This represents the total number of subtasks in the initial subtask set.

3. The artificial intelligence-based human work assistance method according to claim 1, characterized in that, The calculation of the particle size fit includes the following steps: Based on historical task data, K-means clustering is used to obtain the time intervals that a single role can efficiently complete, and the median of these time intervals is taken as the optimal time threshold. ; Calculate the estimated execution time of subtasks based on historical execution time data of similar tasks. ; Granularity fit is calculated based on the estimated time consumption of subtasks and the optimal time consumption threshold: ， in, For subtasks Granularity fit.

4. The artificial intelligence-based human work assistance method according to claim 1, characterized in that, The calculation method for the disassembly priority is as follows: ， in, , As weight, For subtasks The strength of logical dependencies For subtasks Granularity fit For subtasks Disassembly priority.

5. The artificial intelligence-based human work assistance method according to claim 1, characterized in that, The task decomposition and optimization specifically includes the following steps: When the decomposition priority is greater than or equal to the preset priority threshold, the subtask corresponding to that decomposition priority will not be adjusted. When the decomposition priority is less than the preset priority threshold, the relationship between the corresponding thresholds of logical dependency strength and granularity fit is further determined: If the logical dependency strength is less than the first preset threshold but the granularity fit is greater than or equal to the second preset threshold, merge similar subtasks, recalculate the granularity fit and decomposition priority, or adjust the task dependency relationship and recalculate the logical dependency strength and decomposition priority. If the logical dependency strength is greater than or equal to the first preset threshold but the granularity fit is less than the second preset threshold, granularity calibration is performed. When the granularity is too coarse, the subtask is split a second time. When the granularity is too fine, similar subtasks are merged, and the logical dependency strength, granularity fit, and splitting priority are recalculated. If the logical dependency strength is less than the first preset threshold and the granularity fit is less than the second preset threshold, reconstruct the subtask, delete redundant subtasks, and recalculate the logical dependency strength, granularity fit, and decomposition priority. If the logical dependency strength is greater than or equal to the first preset threshold and the granularity fit is greater than or equal to the second preset threshold, the corresponding subtask will not be adjusted.

6. The artificial intelligence-based human work assistance method according to claim 1, characterized in that, The calculation of skill suitability includes the following steps: Human-machine capability profiles are quantified, and human capability vectors and artificial intelligence capability vectors are generated respectively. Each element in the human capability vector and artificial intelligence capability vector represents the skill value of the human and artificial intelligence for a certain skill, respectively. For each subtask in the subtask list, mark the core skill requirements and generate a subtask skill requirement vector. Each element in the subtask skill requirement vector is a skill requirement indicator value. A skill requirement indicator value of 1 indicates that the subtask requires this type of skill, and a skill requirement indicator value of 0 indicates that the subtask does not require this type of skill. After normalizing the human ability vector and the artificial intelligence ability vector based on skill type, the skill probability corresponding to each skill type is calculated. Based on the skill probabilities, the skill importance weights are calculated using information entropy. The normalized skill importance weight is obtained by multiplying the skill importance weight by the skill requirement indicator value of the corresponding skill in the subtask skill requirement vector. The normalized skill importance weights are multiplied and summed with the corresponding skills in the human ability vector and the artificial intelligence ability vector, respectively, to obtain the human skill fit and the artificial intelligence skill fit.

7. The artificial intelligence-based human work assistance method according to claim 1, characterized in that, The allocation utility function is expressed as follows: ， in, Indicates whether the task is performed by a human or artificial intelligence; Subtasks The implementing body is Skill suitability at the time; , Indicates the adjustment coefficient; Indicates the implementing entity The load, For load penalty items; For subtasks The implementing body is The risk factor at any given time is determined based on the corresponding risk level. This is a risk-based reward item.

8. The artificial intelligence-based human work assistance method according to claim 1, characterized in that, The allocation decision rules include: Basic rule: Select the entity with the largest allocation utility function value as the execution role; Constraint rule: When the risk level is greater than the preset level, if the subject with the largest allocation utility function value is artificial intelligence, it will be forcibly converted to a human.

9. The artificial intelligence-based human work assistance method according to claim 1, characterized in that, The aforementioned real-time monitoring of task execution progress and quality, and dynamic adjustment of human-machine task allocation, specifically includes the following steps: During task execution, progress and quality indicators are calculated in real time, and the progress-quality coupling coefficient is calculated based on the progress and quality indicators. When the progress-quality coupling coefficient is greater than or equal to the adjustment threshold, the human ability vector and the artificial intelligence ability vector are updated, and the human-machine task allocation is redistributed. The adjustment threshold is determined based on the risk level.

10. A human work assistance system based on artificial intelligence, characterized in that, The system includes: Data acquisition and preprocessing module: used to acquire multimodal data input by the user, and generate structured task data after preprocessing; Task decomposition module: used to perform preliminary task decomposition on structured task data, generate an initial subtask set, calculate the logical dependency strength, granularity adaptability and decomposition priority of each task in the initial subtask set, optimize task decomposition, and generate a subtask list. Task allocation module: used to quantify human-machine ability profiles, generate human ability vectors and artificial intelligence ability vectors respectively, generate sub-task skill requirement vectors based on the sub-task list, calculate skill fit based on the matching degree between the ability vector and the sub-task skill requirement vector, construct allocation utility function by combining skill fit, load penalty term and risk reward term, and allocate human-machine tasks to each sub-task in the sub-task list based on the allocation utility function and under the constraints of allocation decision rules; Real-time adjustment module: Used to monitor the progress and quality of tasks in real time during the execution of auxiliary tasks, and dynamically adjust the human-machine task allocation.

Images