A method for training scientific thinking of adolescents by using AI

Abstract

The application provides a kind of adolescent AI scientific thinking training method, it is related to education technology and artificial intelligence interactive technology field, including five steps of constructing and issuing fusion training task, synchronous acquisition and fusion multi-modal behavior sequence, converting behavior sequence into thinking structured data model and visual atlas, circulating judgment based on model matching and multi-dimensional index calculation and finally generating visual report, a complete technical processing closed loop is established. Thus the implicit cognitive activity that cannot be observed in traditional teaching is objectively transformed into visual, calculable structured model through the preset analysis and mapping rules, and then through the method of real-time comparison between the model and the deviation mode library, cognitive bias can be automatically identified, and through the method of fusion atlas evolution rate, deviation frequency change and efficiency improvement amplitude to calculate progress index, quantitative evaluation and decision of training state are realized, which provides an objective basis for teaching feedback.

Description

Technical Field

[0001] This invention relates to the fields of educational technology and artificial intelligence interaction technology, and in particular to a method for training adolescents' AI scientific thinking. Background Technology

[0002] In the current field of cultivating artificial intelligence and computational thinking among teenagers, practical training is crucial. However, existing training methods often rely on pre-set linear teaching procedures or outcome-oriented assessments, which have inherent technical limitations. For example, in terms of evaluating the training process, existing methods generally lack objective means of capturing and analyzing the thought process. The evaluation of training effectiveness usually depends only on the correctness of the final task result, or on subjective methods such as post-task questionnaires and interviews. This makes it difficult for instructors to accurately grasp the learner's true thought path, logical coherence, and cognitive obstacles encountered in the problem-solving process, resulting in highly subjective teaching evaluations and difficulty in providing evidence-based personalized feedback.

[0003] Secondly, in terms of intervention during the training process, existing methods struggle to provide real-time and precise guidance. Because they cannot structurally model and identify ongoing thought processes, existing methods cannot detect and correct cognitive biases in learners (such as neglecting variable control or prematurely drawing conclusions) in a timely manner. Furthermore, guidance and assistance are often delayed, typically relying on the instructor's manual observation and experience, or on a unified explanation after training. This lack of a timely and effective correction mechanism negatively impacts training efficiency and effectiveness.

[0004] Finally, at the training model level, existing training processes are typically static and linear, unable to dynamically adjust based on the learner's real-time status during training. Training content, difficulty, and pathways lack adaptability and personalization, failing to form a closed loop of "assessment-feedback-adjustment." This makes the training process rigid, failing to meet the differentiated needs of different learners and hindering learners from effectively advancing their thinking abilities through continuous, targeted practice. Therefore, this invention proposes a method for training AI scientific thinking in teenagers to address the problems existing in current technologies. Summary of the Invention

[0005] To address the aforementioned problems, the present invention aims to propose a method for training adolescents' AI scientific thinking. This invention provides a method that can objectively capture and quantify the thinking process, identify cognitive biases in real time, and dynamically adjust the training process accordingly, thereby solving the problems existing in the prior art.

[0006] To achieve the objectives of this invention, the invention is implemented through the following technical solution: a method for training AI scientific thinking in teenagers, comprising the following steps: Step 1: Build and distribute a fusion training task related to AI scientific concepts, which requires users to complete the task by coordinating the operation of physical entities and digital interfaces; Step 2: During the user's execution of the fusion training task, the entity operation behavior flow and digital interaction behavior flow are collected simultaneously. Then, the two types of behavior flows are time-synchronized and aligned to generate a fusion behavior sequence with time-series labels. Step 3: Based on the preset parsing and mapping rules, the fusion behavior sequence is converted into a structured data model of thinking, and a visual mind map and its evolution data are dynamically generated and recorded accordingly. Then, snapshots of the visual mind map at key nodes are saved according to preset rules to form a map evolution sequence. Step 4: Match the current structured data model of thinking with the pre-stored deviation pattern feature library to determine whether there is a target deviation pattern, and record the current deviation state. Then, based on the structured data model of thinking, the graph evolution sequence and the fusion behavior sequence, calculate a process progress index that reflects the current training state. If a target deviation pattern is identified and the process progress indicator is below a preset threshold, a corrective training subtask is generated and issued. At the same time, the newly generated behavior sequence is collected and added to the fused behavior sequence. Then, the process returns to step three for the next round of evaluation. Otherwise, end the loop and proceed to step five; Step 5: Based on all the evolutionary data, bias state sequences, and performance indicators recorded throughout the training process, generate a comprehensive and visualized training report, marking the completion of this training task.

[0007] A further improvement is that, in step two, the entity operation behavior flow is acquired by at least one of an image sensor, a radio frequency identification sensor, and a pressure sensor, while the digital interaction behavior flow is acquired by recording the user's operation event logs on the graphical user interface.

[0008] A further improvement is that, in step three, the parsing and mapping rules include: parsing rules for segmenting and labeling continuous behavior streams as behavioral units with cognitive intent semantic labels, and mapping rules for converting behavioral units of a specific label sequence into nodes and edges in a thought-structured data model.

[0009] A further improvement is that the semantic tags include at least one of the following: defining features, setting conditions, performing tests, observing results, correcting parameters, forming hypotheses, and verifying logic.

[0010] A further improvement is made in the following way: In step three, the specific method for recording evolution data is as follows: at preset time intervals or key behavioral nodes, a snapshot of the visualized mind map is saved, and the addition, deletion, and modification information of nodes and connecting lines between adjacent snapshots is recorded.

[0011] A further improvement is that, in step four, the bias pattern feature library includes the following preset patterns: variable control missing pattern, causal attribution single pattern, insufficient test iteration pattern, and hypothesis verification loop break pattern.

[0012] A further improvement is that the method for generating corrective training subtasks is as follows: based on the identified target deviation pattern type, the corresponding task template is called from the preset task template library, and the current task context parameters are instantiated into the template to generate an executable subtask instruction set.

[0013] A further improvement lies in the following: In step four, the specific method for calculating the process progress indicators is as follows: Step S1: Calculate the growth rate of the structural complexity of the mind graph based on the graph evolution sequence; Step S2: Based on the deviation pattern recognition results of the structured data model of thinking, calculate the rate of decrease in the frequency of occurrence of the target deviation pattern; Step S3: Based on the fused behavior sequence, calculate the current improvement in overall task performance; Step S4: After normalizing the growth rate of structural complexity, the rate of decrease in frequency of occurrence, and the current improvement, perform weighted fusion to obtain the process progress index.

[0014] A further improvement is that, in step four, the preset threshold is dynamically adjusted based on the user's historical training data or the benchmark data of the user's group.

[0015] A further improvement is that, in step five, the visualized training report is an integrated dashboard view that simultaneously includes the final state of the visualized mind map, a timeline view showing the key evolution path of the map, and trend charts showing the changes in the process progress indicators and performance indicators.

[0016] The beneficial effects of this invention are as follows: This invention establishes a complete technical processing loop by executing a series of coherent steps: constructing and distributing integrated training tasks, synchronously collecting and fusing multimodal behavioral sequences, converting the behavioral sequences into a structured data model and visual graph of thought processes, performing iterative judgments based on model matching and multi-dimensional indicator calculations, and finally generating a visual report. This objectively transforms implicit cognitive activities, which are unobservable in traditional teaching, into a visible and computable structured model through pre-defined analysis and mapping rules. Furthermore, by comparing the model with a deviation pattern library in real time, cognitive deviations can be automatically identified. The progress indicators are calculated using the evolution rate of the fusion graph, changes in deviation frequency, and the magnitude of efficiency improvement. This achieves quantitative evaluation and decision-making of the training status, providing objective evidence for teaching feedback and solving the long-standing technical problem of subjective teaching evaluation, delayed intervention, and rigid training patterns caused by the inability to quantitatively analyze the thought process. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of the steps of the present invention. Detailed Implementation

[0018] To enhance understanding of the present invention, the present invention will be further described in detail below with reference to embodiments. These embodiments are only used to explain the present invention and do not constitute a limitation on the scope of protection of the present invention.

[0019] according to Figure 1 As shown, this embodiment proposes a method for training adolescents' AI scientific thinking. This embodiment is based on the AI ​​scientific concept of designing training tasks using linear classifiers, and is explained accordingly.

[0020] Specifically, the training task is as follows: "Given four types of plastic building blocks—red circles, red squares, blue circles, and blue squares—as data samples, and a sorting device equipped with a color recognition camera and a weighing sensor as the actuator. Please design a decision rule (i.e., a simple linear classifier) ​​that enables the sorting device to automatically sort the 'red circle' building blocks to area A and the other building blocks to area B. You need to conduct experiments by operating physical building blocks and adjust and test the classification rule on the computer interface." The corresponding hardware consists of a physical workbench equipped with a high-definition top camera (for recognizing the color and shape of the building blocks), an embedded high-precision pressure sensor (electronic scale) (for measuring the weight of the building blocks), and a near-field communication (NFC) reader / writer array (each building block has a unique NFC tag embedded in its bottom). The workbench is divided into a "sample area," a "sorting area A," and a "sorting area B."

[0021] Finally, a computer connected to the control panel is set up to run the dedicated software for this training method and provide a graphical user interface (GUI). The interface includes a parameter setting panel (such as setting colors and weight thresholds), virtual sorting device control buttons, a real-time data feedback area, and a mind map display area.

[0022] This method includes the following steps: Step 1: Constructing and distributing fusion training tasks A fusion training task related to AI scientific concepts is constructed and distributed, requiring users to collaboratively complete the task by manipulating physical entities and a digital interface. Specifically, the training task description is presented to the user simultaneously through a computer's graphical interface and voice prompts. The task explicitly requires users to collaboratively operate physical blocks (such as picking them up, placing them on sensors, and moving them to different areas) and operate the digital interface (such as setting color thresholds and clicking the "Execute Sorting" button) to complete the task.

[0023] Step 2: Synchronously collect and align multimodal behavioral data During the user's execution of the fusion training task, entity operation behavior flow and digital interaction behavior flow are collected simultaneously, wherein the entity operation behavior flow includes: The system continuously captures video streams via a camera and outputs data points in real time using machine vision algorithms, including {timestamp, visual tracking ID, object color, object shape, and desktop coordinates (x, y)}. The event is triggered by an NFC reader when the block is picked up or put down, outputting a data point of {timestamp, NFC tag ID, event type (pick up / put down)}. And outputs {timestamp, weight value (grams)} data points via pressure sensors when the blocks are placed in the weighing area.

[0024] Digital interactive behavior flow works by listening to graphical user interface events and recording {timestamp, control ID (e.g., 'color threshold slider'), event type (e.g., 'value changed'), and the changed value}.

[0025] Then, the two types of behavior streams undergo time synchronization and alignment processing to generate a fused behavior sequence with time-series labels. That is, during the initial data acquisition, both are synchronized with a clock via the Network Time Protocol (NTP). Each data point is tagged with a microsecond-level timestamp from this unified clock source upon generation. The background data fusion module arranges all data points from all sources in ascending order of timestamps, interweaving them into a unified, time-series fused behavior sequence. A fused behavior sequence can then be represented as: [t1: NFC Pick Up - Block 3], [t2: Interface - Color Threshold Slider - Value Change - 0.6], [t3: Camera - Move - Block 3 - (x1, y1)], [t4: Pressure Sensor - Measurement - Block 3 - 120].

[0026] Step 3: Generate a dynamic mind visualization map and its evolution data. Based on preset parsing and mapping rules, the fusion behavior sequence is converted into a structured data model of thinking, and a visual mind map and its evolution data are dynamically generated and recorded accordingly. Then, snapshots of the visual mind map at key nodes are saved according to preset rules to form a map evolution sequence.

[0027] Specifically, it is divided into three parts: behavior analysis, mental model construction, and graph generation. Behavioral analysis: A rule base for analysis is predefined, and this rule base predefines how to aggregate and label raw events as high-level semantic units. Accordingly, the predefined set of semantic tags in this embodiment is: "Define features", "Set conditions", "Execute tests", "Observe results", "Modify parameters", "Form hypotheses and verification logic, for example: Rule R1: If the events [NFC Pick Up - Block X] and [Camera - Move - Block X] occur consecutively, they are aggregated into a [Select Sample] behavior unit and assigned a semantic label "Define Feature". The unit attributes record the sample ID and visual features.

[0028] Rule R2: If the [Interface - Value Change - Color Threshold Slider - v] event occurs, mark it as a [Set Condition] behavior unit with the label "Set Condition" and the parameter v.

[0029] Rule R3: If the [Interface - Click - 'Execute Sorting' button] event occurs, mark it as a [Execute Test] behavior unit with the label "Execute Test".

[0030] Mental Model Construction: A pre-defined mapping rule base is established, which defines how to transform a sequence of behavioral units into a graph-structured mental model. For example: Rule M1: When a [form a hypothesis] unit appears (e.g., the content is "If the color is red and the shape is round, then it is assigned to area A"), a "hypothesis node" is created in the mental model, and the node content records the hypothesis.

[0031] Rule M2: If a [Hypothesis Formulation] unit appears after the [Test Execution] unit, then a "Verification Node" is created in the model, and a "Verified" directed edge is established between the hypothesis node and the verification node.

[0032] Rule M3: If an "Observation Result" unit appears after the "Execute Test" unit (e.g., "Block 3 was assigned to area B, which is not as expected"), then create a "Result Node" and establish a "Generation" edge with the verification node.

[0033] Graph Generation: The above-mentioned structured data model of thought is rendered into a visual graph composed of nodes and connections, displayed in a designated area. Simultaneously, each time a user completes a behavioral unit semantically labeled "Perform Test" or "Form Hypothesis" (a preset key behavioral node), a snapshot of the current visual mind graph is automatically triggered. Each snapshot is compared to the previous snapshot, recording newly added nodes, deleted nodes, nodes with modified attributes, and newly added or deleted connections. This series of snapshots and their differences constitutes the "graph evolution sequence."

[0034] Step 4: Identification of Cognitive Biases and Adaptive Training Cycle The current structured data model of thinking is matched with the pre-stored deviation pattern feature library to determine whether there is a target deviation pattern, and the current deviation state is recorded. Then, based on the structured data model of thinking, the graph evolution sequence and the fusion behavior sequence, a process progress index reflecting the current training state is calculated. If a target deviation pattern is identified and the process progress indicator is below a preset threshold, a corrective training subtask is generated and issued. At the same time, the newly generated behavior sequence is collected and added to the fused behavior sequence. Then, the process returns to step three for the next round of evaluation. Otherwise, end the loop and proceed to step five.

[0035] Specifically, the pre-stored deviation pattern feature library contains four typical cognitive bias graph structure definitions, namely: Missing variable control pattern: It is characterized by a "conclusion" node (such as "all red circles can be sorted") being directly associated with multiple "result" nodes (multiple sorting results), but there is a lack of connection between the conclusion node and the "condition" node (such as "consistent weight"), or the condition node is missing. Single causal attribution model: Its characteristic is that the reasoning path in the mental model is too simple, for example, it only contains a single path of "hypothesis A -> verification -> result", and lacks exploration and elimination of competing hypotheses (hypothesis B, C); Insufficient test iteration mode: Its characteristic is that after a "negative result" node (such as sorting error) occurs, no new "correction parameters" or "new hypothesis" nodes are generated for a long time (such as more than 2 minutes). Hypothesis verification cycle break mode: It is characterized by the existence of "hypothesis" nodes in the model, but no "verification" nodes are connected to them, and the hypothesis is not tested.

[0036] This involves performing real-time subgraph matching between the current structured data model of thinking and the deviation pattern feature library. If a match is successful, a "target deviation pattern occurrence event" is recorded, and the event log includes the time, user ID, and the ID of the matched deviation pattern.

[0037] Then proceed with the following steps: Step S1: Based on the graph evolution sequence, calculate the growth rate of the structural complexity of the mind graph, i.e., take the two most recent snapshots in the graph evolution sequence (e.g., the snapshot at the current evaluation time and the snapshot 5 minutes ago). Calculate the difference in the total number of nodes in the mind model corresponding to the two snapshots, and then divide it by the time interval between the two snapshots (minutes) to obtain the growth rate of "nodes / minute". Step S2: Based on the deviation pattern recognition results of the structured data model of thinking, calculate the frequency reduction rate of the target deviation pattern. This is done by counting the number of times the target deviation pattern occurred (F1) within the current evaluation period (the most recent 5 minutes) and the number of times it occurred (F2) within the previous evaluation period (5-10 minutes ago). The calculation method is as follows: (F2- F1) / F 2; Step S3: Based on the fused behavior sequence, calculate the current improvement in overall task performance. Specifically, extract performance-related data from the fused behavior sequence. In this embodiment, the "overall task performance value" is defined as: (Number of effective operations / Total number of operations) * 100. Effective operations refer to those operations that result in the sorting result being consistent with the currently set rules. The current performance value (based on data from the entire sequence to date) is compared with the user's historical average performance value stored in the database (based on the past 10 training iterations). Therefore, the improvement is calculated as follows: (Current value - Historical average) / Historical average; Step S4: After normalizing the growth rate of structural complexity, the rate of decrease in frequency of occurrence, and the current improvement, a weighted fusion is performed to obtain the process progress index. Specifically, the three calculation results of steps S1, S2, and S3 are respectively subjected to minimum-maximum normalization and mapped to the [0, 1] interval. Then, a weighted sum is performed using preset fixed weights to obtain the final process progress index P. In this embodiment, the preset weights are: rigorous thinking step (S2) 0.5, divergent thinking (S1) 0.3, and execution efficiency (S3) 0.2.

[0038] Finally, a qualifying threshold is set (e.g., P=0.7). The judgment logic is: if (the target deviation pattern is matched and P<0.7), then intervention is performed; otherwise, the loop ends. This preset threshold is dynamically adjusted based on the user's historical training data or the benchmark data of their group.

[0039] If intervention is required, the corresponding template is retrieved from the task template library based on the specific deviation pattern identified. For example, if a "variable control missing pattern" is identified, template T01 is retrieved. Template T01 contains the following: "Your current rule may be ignoring the influence of other factors. Please design an experiment to specifically verify whether 'block weight' affects your classifier's decision, and clearly state which conditions need to be kept the same during the experiment?" The current task context (such as the user-defined rule text) is then automatically filled into the template placeholders, generating specific subtask instructions and issuing them.

[0040] Next, the user performs a corrective subtask, simultaneously collecting all behavioral data generated during this process to form a new behavioral sequence fragment, which is then appended to the end of the previously maintained fusion behavioral sequence. The process then returns to step three. This updated, longer fusion behavioral sequence is then used to re-analyze, model, and generate a new mind map and evolutionary sequence. Then, this step (step four) is repeated for a new round of evaluation. This cycle continues until the exit conditions are met.

[0041] Step 5: Generate the final training report Based on all evolutionary data, deviation state sequences, and performance indicators recorded throughout the training process, a comprehensive visual training report is generated, marking the completion of this training task. Correspondingly, the visual training report is an integrated dashboard view, which simultaneously includes the final state of the visual mind map, a timeline view showing the key evolution path of the map (presented in timeline form, with key nodes marked on the axis (such as the proposal of important hypotheses, deviation identification, and the issuance of intervention tasks), and trend charts showing the changes in the process progress indicators and performance indicators (showing the change trend of the process progress indicator P in the entire training process as a curve graph, while auxiliary charts can be used to show the changes in performance indicators (such as accuracy and efficiency).

[0042] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the present invention without departing from its framework and scope of application, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for training adolescents' AI scientific thinking, characterized by: Includes the following steps: Step 1: Build and distribute a fusion training task related to AI scientific concepts, which requires users to complete the task by coordinating the operation of physical entities and digital interfaces; Step 2: During the user's execution of the fusion training task, the entity operation behavior flow and digital interaction behavior flow are collected simultaneously. Then, the two types of behavior flows are time-synchronized and aligned to generate a fusion behavior sequence with time-series labels. Step 3: Based on the preset parsing and mapping rules, the fusion behavior sequence is converted into a structured data model of thinking, and a visual mind map and its evolution data are dynamically generated and recorded accordingly. Then, snapshots of the visual mind map at key nodes are saved according to preset rules to form a map evolution sequence. Step 4: Match the current structured data model of thinking with the pre-stored deviation pattern feature library to determine whether there is a target deviation pattern, and record the current deviation state. Then, based on the structured data model of thinking, the graph evolution sequence and the fusion behavior sequence, calculate a process progress index that reflects the current training state. If a target deviation pattern is identified and the process progress indicator is below a preset threshold, a corrective training subtask is generated and issued. At the same time, the newly generated behavior sequence is collected and added to the fused behavior sequence. Then, the process returns to step three for the next round of evaluation. Otherwise, end the loop and proceed to step five; Step 5: Based on all the evolutionary data, bias state sequences, and performance indicators recorded throughout the training process, generate a comprehensive and visualized training report, marking the completion of this training task.

2. The method for training adolescents' AI scientific thinking according to claim 1, characterized in that: In step two, the physical operation behavior flow is acquired by at least one of an image sensor, a radio frequency identification sensor, and a pressure sensor, while the digital interaction behavior flow is acquired by recording the user's operation event log on the graphical user interface.

3. The method for training adolescents' AI scientific thinking according to claim 1, characterized in that: In step three, the parsing and mapping rules include: parsing rules for segmenting and labeling continuous behavior streams as behavioral units with cognitive intent semantic labels, and mapping rules for converting behavioral units of a specific label sequence into nodes and edges in a thought-structured data model.

4. The method for training adolescents' AI scientific thinking according to claim 3, characterized in that: The semantic tags include at least one of the following: defining features, setting conditions, performing tests, observing results, modifying parameters, forming hypotheses, and verifying logic.

5. The method for training adolescents' AI scientific thinking according to claim 1, characterized in that: In step three, the specific method for recording evolution data is as follows: at preset time intervals or key behavioral nodes, a snapshot of the visualized mind map is saved, and information on the addition, deletion, and modification of nodes and connecting lines between adjacent snapshots is recorded.

6. The method for training adolescents' AI scientific thinking according to claim 1, characterized in that: In step four, the bias pattern feature library includes the following preset patterns: variable control missing pattern, causal attribution single pattern, insufficient test iteration pattern, and hypothesis verification loop breakage pattern.

7. The method for training adolescents' AI scientific thinking according to claim 1, characterized in that: The method for generating corrective training subtasks is as follows: based on the identified target deviation pattern type, the corresponding task template is called from the preset task template library, and the current task context parameters are instantiated into the template to generate an executable subtask instruction set.

8. The method for training adolescents' AI scientific thinking according to claim 1, characterized in that: In step four, the specific method for calculating the process progress indicators is as follows: Step S1: Calculate the growth rate of the structural complexity of the mind graph based on the graph evolution sequence; Step S2: Based on the deviation pattern recognition results of the structured data model of thinking, calculate the rate of decrease in the frequency of occurrence of the target deviation pattern; Step S3: Based on the fused behavior sequence, calculate the current improvement in overall task performance; Step S4: After normalizing the growth rate of structural complexity, the rate of decrease in frequency of occurrence, and the current improvement, perform weighted fusion to obtain the process progress index.

9. A method for training adolescents' AI scientific thinking according to claim 1, characterized in that: In step four, the preset threshold is dynamically adjusted based on the user's historical training data or the benchmark data of the group to which they belong.

10. A method for training adolescents' AI scientific thinking according to claim 1, characterized in that: In step five, the visualization training report is an integrated dashboard view that includes the final state of the visualization mind map, a timeline view showing the key evolution path of the map, and trend charts showing the changes in the process progress indicators and performance indicators.

Images