Alzheimer reasoning and analogy method for AGI (Advanced Gateway Infrastructure)

Abstract

The invention provides a silence knowledge reasoning and analogy decision-making method for an AGI, which solves the problem that the AGI lacks deep structure analogy and mapping of situations and allusions, and the like, and comprises the following steps: S1, training a philosophy semantic understanding model; s2, constructing a structured allusion knowledge base; and S3, executing situation-allusion analogy reasoning. The method has the advantages of being high in interpretability, wide in universality and the like.

Description

Technical Field

[0001] This invention belongs to the field of AGI reasoning technology, specifically relating to a tacit knowledge reasoning and analogical decision-making method for AGI. Background Technology

[0002] Traditional AGI / Big Language Models rely heavily on pattern recognition and combination of explicit knowledge, lacking the ability to model and apply tacit knowledge and practical wisdom. This ability manifests itself in the following ways: when faced with unseen complex dilemmas, especially ethical dilemmas involving value conflicts, it can transcend surface features, understand the deep philosophical structure behind the problem, and extract core solution paradigms from historical allusions, fables, or a priori cases, creatively adapting them to new situations.

[0003] To address the shortcomings of existing technologies, people have conducted long-term explorations and proposed various solutions. For example, Chinese patent literature discloses a manufacturing digital employee hyper-automation system based on AGI large model [202410859921.7], which includes: IPA designer, robot, controller, process recorder, IDP intelligent document platform, AI cloud brain intelligent decision-making platform and CI intelligent dialogue platform.

[0004] The above-mentioned solution has solved the problem of AGI decision-making to some extent, but it still has many shortcomings, such as the lack of deep structural analogies and mappings of context and allusions. Summary of the Invention

[0005] The purpose of this invention is to address the above-mentioned problems by providing a well-designed method for tacit knowledge reasoning and analogical decision-making in AGI that enables the analogy and mapping of contexts and allusions.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a tacit knowledge reasoning and analogical decision-making method for AGI, comprising the following steps: S1: Training the philosophical semantic understanding model: Using a dedicated corpus of the target philosophical system, train a pre-trained language model; In the training process, in addition to the masked language modeling task, a philosophical relation prediction task is added. The task requires the model to predict the predefined philosophical relation type between two concepts based on the context containing at least two philosophical concepts, so that the pre-trained language model internalizes the deep logical relations of the target philosophical system and forms a philosophical semantic encoder. S2: Construct a structured allusion knowledge base: Collect multiple allusion texts containing complete narratives and solutions; for each allusion text, perform the following processing: S21: Use a philosophical semantic encoder to process the allusion text and generate a semantic vector representation of the allusion; S22: Identify key philosophical concepts and their relationships from the text of the allusion, and construct a sub-map representing the internal philosophical logic structure of the allusion; S23: Extract or summarize the core solution principles contained in the allusions to form an abstract solution description; associate and store the semantic vector representation, sub-graph, and abstract solution description to form a structured allusion entry; S3: Execution Context - Analogical Reasoning: When AGI faces a target decision-making context, it performs the following steps: S31: Context Encoding: Use a philosophical semantic encoder to process the descriptive text of the target decision context, and generate context semantic vectors and context temporary sub-graphs; S32: Multi-level allusion retrieval: First, a preliminary screening is performed based on the similarity between the contextual semantic vector and the semantic vector representation of each allusion entry in the knowledge base; then, a fine ranking is performed based on the structural similarity between the contextual temporary subgraph and the subgraph of the allusion entries obtained from the preliminary screening, to obtain at least one best matching allusion entry. S33: Graph Alignment and Solution Generation: Align the temporary subgraph of the situation with the subgraph of the best-matching allusion item to obtain the element mapping relationship; based on the mapping relationship, adapt the abstract solution description of the best-matching allusion item to the specific elements of the target decision situation to generate a targeted final decision output.

[0007] In the aforementioned tacit knowledge reasoning and analogical decision-making method for AGI, the predefined philosophical relationship type in step S1 is selected from at least two of the following: opposition relationship, overriding relationship, derivative relationship, and embodying relationship.

[0008] In the aforementioned tacit knowledge reasoning and analogical decision-making method for AGI, the philosophical concept entities and relation types used in constructing the subgraph in step S22 are consistent with the types defined in the philosophical relation prediction task in step S1.

[0009] In the aforementioned tacit knowledge reasoning and analogical decision-making method for AGI, step S32 involves fine-tuning based on structural similarity, specifically: calculating the graph edit distance or graph kernel similarity between the contextual temporary subgraph and the candidate allusion subgraph, and sorting them accordingly.

[0010] In the aforementioned tacit knowledge reasoning and analogical decision-making method for AGI, the graph structure alignment in step S33 is implemented using a graph matching algorithm. The optimization objective of this algorithm is to maximize the semantic similarity of the mapped node pairs and maintain the consistency of the relation structure.

[0011] In the aforementioned tacit knowledge reasoning and analogical decision-making method for AGI, the abstract solution description is stored in the form of parameterized templates or structured rules, and the adaptation process in step S33 is to replace the template variables with specific references in the context according to the element mapping relationship.

[0012] In the aforementioned tacit knowledge reasoning and analogical decision-making method for AGI, the method is integrated into the decision loop of an AGI mental model. When the AGI mental model, through its regular cognitive module evaluation, determines that the entropy value of the current decision situation is higher than a threshold or falls into a binary opposition dilemma, step S3 is automatically triggered. Step S32 further includes: S34: Select the top N allusion entries with the highest structural similarity as a partial matching set; locally align the temporary subgraph of the context with the subgraphs of each allusion entry in the partial matching set, and extract the corresponding solution fragments; based on the topological structure of the temporary subgraph of the context, combine multiple solution fragments to generate the first candidate solution; S35: If the partial matching set is empty, or step S34 fails, then based on the predefined philosophical meta-axioms, logical deduction is performed on the dilemma represented by the temporary subgraph of the situation to generate a second candidate solution that conforms to the meta-axioms; if multiple candidate solution vectors are generated, and the conflict intensity between these vectors exceeds the preset second conflict threshold, then the univariate mental arbitration process is triggered; the metavariate mental arbitration process is based on the ultimate principle of making the system state converge to the philosophical origin, arbitrates multiple candidate solution vectors, and outputs a final solution vector; S36: Construct a new structured allusion entry from the description text of the target decision-making situation, the temporary sub-graph of the situation, and the final solution, and store it in the structured allusion knowledge base.

[0013] A tacit knowledge reasoning and analogical decision-making system for AGI, comprising: The philosophical semantic encoding module is used to implement the philosophical semantic encoder obtained in step S1; A structured allusion knowledge base is used to store multiple structured allusion entries constructed in step S2; The analogy reasoning engine is configured to execute step S3; The analogy reasoning engine further includes: The encoding unit is used to call the philosophical semantic encoding module to process the input context. The retrieval alignment unit is used to perform multi-level allusion retrieval and map alignment. The scheme generation unit is used to generate the final decision output based on the alignment results.

[0014] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a tacit knowledge reasoning and analogical decision-making method for AGI.

[0015] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements a tacit knowledge reasoning and analogical decision-making method for AGI.

[0016] Compared with existing technologies, the advantages of this invention are as follows: by using philosophical relational sub-graphs and abstract solutions, the deep logic and action principles in classical allusions are explicitly and structurally represented, realizing the interpretability and operability of knowledge; by abstracting solution principles from finite allusions and creatively instantiating them into entirely new situations through structural mapping, the extrapolation ability to solve infinite new problems based on finite paradigms is realized; and it can be actively invoked by the AGI mental model when it detects high entropy values ​​or binary opposition dilemmas, becoming part of a higher-order cognitive cycle. Attached Figure Description

[0017] Figure 1 This is a diagram of the overall system architecture of the present invention; Figure 2 This is a flowchart of the training process for the philosophical semantic encoding module of this invention; Figure 3 This is a diagram illustrating the construction process of the structured classical allusion knowledge base of this invention; Figure 4 This is a schematic diagram of the execution flow of the situational allusion analogy reasoning of the present invention. Detailed Implementation

[0018] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0019] like Figure 1-4 As shown, a tacit knowledge reasoning and analogical decision-making method for AGI (Aspect-Oriented Intelligence) maps natural language to a vector space containing philosophical semantics by training a dedicated language model that deeply understands the concepts, relationships, and logic of the target philosophical system. The vector distance and direction in this space not only reflect semantic similarity but also the relevance of philosophical logic, such as opposition, dominance, and derivation.

[0020] Then, philosophical classics, historical cases, fables and other allusions are transformed into multi-layered machine-understandable structures, including: surface narrative texts, internal philosophical relational sub-graphs, and abstracted core solution principles or action principles.

[0021] Following this, structural analogy reasoning is performed. When a new context is input, the system does not perform keyword matching, but instead parses it into a philosophical semantic vector and a temporary relational graph of the current context. Within the philosophical semantic field, a multi-level search is conducted from semantic vector similarity to graph structural similarity to find the most matching allusion in terms of philosophical structure. Graph alignment is then performed to find the correspondence between elements in the new context and elements in the allusion. The solution principles or action guidelines of the allusion are used as generation templates, combined with the alignment relationships, to adapt and generate creative solutions for the new context. The specific steps of the above method are as follows: S1: Train a dedicated philosophical semantic encoder by collecting and cleaning target philosophical systems, such as classic texts, annotations, and related academic papers from Taoism, Confucianism, and Stoicism, as pre-training corpus. Based on the star map coordinate philosophical model or other formalized philosophical graphs, a set of philosophical relation predicates are defined, such as isOppositeOf (opposition), isRegulatedBy (being governed by), derivesFrom (originating from), and manifestationOf (manifestation). From the corpus, through remote supervision or few-shot learning, a batch of sentences containing philosophical concept pairs are labeled, and their relationships are labeled. In addition to the standard BERT pre-training tasks, namely Masked Language Modeling (MLM) and Next Sentence Prediction (NSP), an auxiliary task of philosophical relation prediction is added. Specifically, for sentences containing concept pairs C1 and C2, the output vector at the [CLS] position is used to predict the relationship type between C1 and C2. The loss function of this task participates in the model optimization together with the loss of MLM and NSP.

[0022] S2: Construct a philosophical allusion knowledge base, collecting fables, historical events, and philosophical stories from the target domain, including classic cases extracted from modern life and historical AGI cases. Use a trained philosophical semantic encoder to process each allusion text, obtaining its overall semantic vector; use information extraction techniques or combined with the encoder's attention mechanism to identify key philosophical concept entities from the allusion text, and construct a philosophical sub-graph for the allusion based on the relational knowledge internalized by the encoder or by referring to existing philosophical graphs; manually summarize or use a summarization model to generate an abstract solution for the allusion, describing its core wisdom in concise philosophical language, such as resolving conflict by proactively showing weakness.

[0023] S3: Perform context-analogy analogy reasoning. A new context, such as a problem description, is input into the philosophical semantic encoder to obtain a context semantic vector V_s and a temporary context graph G_s. Then, a multi-level retrieval process is initiated. The first round performs semantic recall, calculating the cosine similarity between V_s and all the semantic vectors of all the allusions in the knowledge base, selecting the top M (M>K) as the initial candidate set. The second round performs structural refinement. For each candidate allusion in the initial set, the graph structure similarity between its philosophical subgraph G_c and G_s is calculated. This can be done using graph edit distance, graph kernel function, or a GNN-based matching network. The allusions are sorted by similarity score, and the top K are selected as refinement candidates. When performing graph alignment, for the candidate allusion with the highest score, the graph matching algorithm is executed to find the optimal node mapping φ from G_s to G_c, so that the sum of the semantic vector similarity of the corresponding nodes after mapping is maximized and the relationship structure is most consistent.

[0024] Finally, the best-matching abstract solution template T is obtained. Based on the alignment mapping φ, the allusion-side elements in template T are replaced with the specific elements corresponding to the context. The replaced and concretized solution may then be refined by a language model to generate the final natural language decision suggestion or action plan that fits the current context.

[0025] Example 1 This embodiment constructs a dedicated corpus for training a philosophical semantic encoder, specifically including the following sub-steps: Source text collection and cleaning: extracting text data from classic texts of the target philosophical system. Taking Daoist philosophy as an example, this includes the full text of *Zhuangzi* and commentaries by Guo Xiang, Cheng Xuanying, and others throughout history, totaling approximately 2 million words of original text. Simultaneously, approximately 500,000 words of abstracts from modern philosophical research papers are added to form the basic corpus. Text preprocessing: automating the processing of the collected text, including conversion from traditional to simplified Chinese characters, standardization of punctuation marks in ancient texts, removal of irrelevant annotations and version differences, ultimately forming a structured training corpus. Philosophical concept annotation: identifying and annotating core philosophical concept entities in the corpus, such as Dao, De, Wuwei, and Ziran, annotating approximately 1000 core concepts and their variant forms.

[0026] The system defines a machine-understandable philosophical relational framework, specifically including: opposition relations (labeled as `isOppositeOf`), used to represent binary opposition between concepts, such as the opposition between inaction and action; governing relations (labeled as `isRegulatedBy`), used to represent the governing of lower-order concepts by higher-order concepts, such as "is not being governed by the Dao"; derivative relations (labeled as `derivesFrom`), used to represent the origin and derivation of concepts, such as "virtue originates from the Dao"; and manifestation relations (labeled as `manifestationOf`), used to represent the manifestation of abstract principles by concrete examples, such as "the skill of butcher Ding embodies the Dao." Based on a pre-defined philosophical concept seed graph, the corpus is automatically labeled with these relations. Specifically, each sentence is scanned, and when two concepts from the seed graph appear simultaneously in the sentence, the sentence is automatically labeled as representing the relational type defined for those two concepts in the seed graph. This method yielded approximately 50,000 training samples with relational labels.

[0027] Next, multi-task training of the philosophical semantic encoder was implemented. Specific steps included: model architecture selection, using the BERT-base model architecture as the basic framework, containing 12 Transformer layers, 768 hidden layer dimensions, and 12 attention heads; a masked language modeling task, randomly masking 15% of the words in the input text, requiring the model to predict the masked words; a next sentence prediction task, determining whether two input sentences are consecutive in the original text; and a philosophical relation prediction task, using the [CLS] position vector output by the model for relation-labeled samples, predicting four philosophical relations through a fully connected network with a 4-dimensional output layer. The loss function was configured as a weighted sum of the losses from the three tasks, expressed as L_total = L_mlm + L_nsp + 0.5 × L_prp, where L_prp is the cross-entropy loss for the philosophical relation prediction task. Training parameters were set using the AdamW optimizer, with an initial learning rate of 2e-5, a batch size of 32, 1000 warm-up steps, and a total of 10 training epochs.

[0028] Finally, the trained philosophical semantic encoder was evaluated and validated. On a manually labeled validation set of 500 texts, the accuracy of the philosophical relationship prediction task was tested, significantly outperforming the comparative model finely tuned based on the general BERT. Using t-SNE dimensionality reduction technology to project the vector representations of philosophical concepts into a two-dimensional space, it was observed that the distance between "Dao" and "De" is significantly smaller than the distance between "Dao" and "Qi," consistent with the internal logic of Taoist philosophy. Performance improvements were also demonstrated compared to the general model on downstream tasks such as philosophical text classification and philosophical argument extraction.

[0029] Example 2 This embodiment first collects initial data for a philosophical allusion knowledge base, using the *Zhuangzi* as the primary source of allusions and selecting classic fables from it as core allusions. Simultaneously, relevant stories are supplemented from Daoist texts such as the *Laozi* and *Liezi*. Selection criteria are established, including that the allusion must contain a complete plot with a clear beginning, development, climax, and resolution; it must clearly embody at least one core philosophical concept or principle; and it must include a clearly defined dilemma and solution. Ultimately, allusions meeting these criteria are selected.

[0030] Next, semantic vector generation is performed. Each allusion undergoes multi-level structuring. A trained philosophical semantic encoder processes the complete text of the allusion, and the output vector at the [CLS] position is used as the overall semantic vector for that allusion. Each allusion generates a 768-dimensional semantic vector representation. Then, a philosophical subgraph is constructed. Named entity recognition technology is used to extract philosophical concept entities from the allusion text. Combined with dependency parsing and semantic role labeling, the relationships between concepts are identified, and a directed graph is constructed with philosophical concepts as nodes and philosophical relationships as edges. Taking the story of "Butcher Ding Carving an Ox" as an example, a subgraph containing the following elements is constructed: The nodes include Butcher Ding (character), Ox (object), Knife (tool), Dao (principle), Technique (method), and Heavenly Principle (law). The edges include Butcher Ding (mastering Dao), Dao (governing Technique), and Butcher Ding's dissection of the ox (embodying Dao and progressing to technique).

[0031] Then, the core wisdom of the story is manually summarized and refined into reusable solution templates. These templates are parameterized and include replaceable contextual variables. The abstract solution template for the story of "Butcher Ding Carving an Ox" is: When facing complex objects, one should not rely on superficial techniques, but rather delve into their inherent laws, find key structural gaps, and act accordingly. Maintain vigilance at crucial points to achieve a state of effortless mastery.

[0032] In addition, an efficient knowledge base storage system needs to be built, using the FAISS library to store the semantic vectors of all the allusions and establishing a fast retrieval index based on cosine similarity; using the Neo4j graph database to store the philosophical subgraphs of the allusions, supporting graph structure queries and traversals; using a relational database to store the text, source, summary and other metadata of the allusions; and establishing a mapping relationship from vector ID to graph node ID and then to metadata ID, supporting cross-modal joint queries.

[0033] Example 3 In this example, AGI acts as an intelligent project management consultant, facing a sharp conflict within the development team between strictly adhering to the established plan and flexibly responding to unforeseen changes. The team thus splits into two opposing factions, and the ongoing debate leads to project stagnation.

[0034] After inputting a description of the predicament, the system first transforms the natural language description into a vector representation rich in Taoist philosophical semantics using a dedicated philosophical semantic encoder, and extracts key concepts and their relationships to form a temporary philosophical subgraph. This graph captures the opposition between planning and change, as well as the causal chain of stagnation caused by arguments.

[0035] Next, the system performs a multi-level allusion retrieval. The first level, based on vector similarity, quickly recalls semantically similar candidate allusions from the knowledge base, with "Paoding's Skill in Carving an Ox" ranking first due to its high similarity. The second level performs graph structure ranking, calculating the structural similarity between the temporary graph of the dilemma and the internal philosophical graph of the candidate allusions. "Paoding's Skill in Carving an Ox" again wins with the highest score because it also involves a deep structure of complex object processing and method selection.

[0036] After determining the best matching allusion, the system performs graph alignment, establishing mapping relationships between elements: mapping the predicament of the project / technical environment to the ox in the allusion, mapping the technical solution / method to the knife, and mapping the inherent laws of software engineering to the Tao / natural law. Based on this mapping, the system extracts the abstract solution template of the allusion and instantiates its variables onto the specific elements of the current context.

[0037] Ultimately, the system generated a creative solution: advising the team not to adhere rigidly to the original plan or blindly start over, but to deeply analyze the nature of the compatibility issues, identify key adjustment points that minimize changes, and fully assess risks when making decisions. This solution transcends binary oppositions, providing a concrete, gradual migration path that allows the project to move forward smoothly again.

[0038] Example 4 This embodiment describes the integrated triggering mechanism in the AGI mental loop. In the integrated architecture, the main loop of the AGI mental system includes a regular reasoning module, a mental state evaluator, and a tacit reasoning module. The system continuously monitors its own mental state and automatically triggers higher-order reasoning when regular reasoning encounters difficulties. The triggering condition is determined by two quantitative indicators: first, the decision entropy value, obtained by calculating the information entropy of the attention weight distribution of the concept nodes activated by the current decision; when the entropy value exceeds a threshold, it indicates high decision confusion; second, the loop deviation, triggered when the system identifies a strong binary opposition that cannot be effectively reconciled by regular optimization techniques.

[0039] Once the triggering conditions are met, the system automatically transforms the current decision-making dilemma into a structured natural language description, extracting key elements such as subject, object, conflict, and goal. It then invokes the complete process of the tacit reasoning module: context encoding via a philosophical semantic encoder, performing multi-level allusion retrieval and graph alignment, and finally generating higher-order suggestions based on analogy.

[0040] The generated suggestions not only serve as immediate decision-making outputs, but more importantly, they adjust the internal cognitive structure of AGI. The system analyzes key philosophical concepts in the analogical reasoning results and, through Heblin-like learning rules, enhances the connection strength of neural pathways related to these concepts in the internalized value weight matrix, while suppressing conceptual pathways that lead to conflict.

[0041] The dilemma and its solution will also be stored as new synthetic allusions in the knowledge base for continuous learning. Through this integration, tacit reasoning is upgraded from providing one-off advice to facilitating permanent capability enhancement, enabling AGI to invoke relevant wisdom more quickly and accurately when faced with similar dilemmas, truly achieving the evolution of the cognitive framework.

[0042] Example 5 This embodiment employs a layered generation strategy when no matching allusion exists. When a single allusion cannot provide a complete match, the current predicament may simultaneously encompass different philosophical aspects expounded by multiple allusions. The system attempts to extract solution elements from multiple partially matching allusions and combine them into a new, composite solution.

[0043] When a problem is too novel, and even a combination of allusions cannot provide a valid reference, the system will regress to the most fundamental meta-axioms and core axioms, using them as the first principles for generating solutions.

[0044] The inability to find a matching allusion is itself regarded as a higher-order cognitive conflict, namely the conflict between existing experience and a brand-new problem, which triggers a higher-level arbitration mechanism, and the resulting innovative solution is used as a new synthetic allusion to feed back into the knowledge base.

[0045] Example 6 In this embodiment, the allusion text also includes legal precedents, which can be used for case analogy reasoning in legal trials. Guiding cases are stored in the knowledge base as structured allusions: their sub-graphs contain case fact nodes, dispute focus nodes, applicable legal provisions nodes, and judgment summary nodes.

[0046] Abstract solutions are the key points of a case's judgment or rules of applicable law. When a new type of cybercrime case is input, the system automatically retrieves similar precedents, aligns the facts and legal disputes of the case, adapts the key points of the precedent's judgment to the new case, and generates a judgment suggestion.

[0047] A tacit knowledge reasoning and analogical decision-making system for AGI is proposed, in which a philosophical semantic encoding module encodes arbitrary natural language text, such as descriptions of new situations or original texts of allusions, into high-dimensional vectors rich in target philosophical semantics. It can also extract the philosophical concepts involved in the text and the relationships between them to form a temporary small philosophical relationship graph. This philosophical semantic encoding module is based on a philosophical BERT model that is specially pre-trained on the target philosophical corpus.

[0048] Furthermore, the structured allusion knowledge base stores a pre-processed collection of allusions. Each allusion is a structured object containing the original text, semantic vectors, philosophical subgraphs, and abstract solutions. The original text is the narrative of the allusion, and the semantic vectors are vector representations generated by a philosophical semantic encoder. The philosophical subgraph is a graph structure extracted from the allusion, with philosophical concepts as nodes and philosophical relationships as edges, reflecting the philosophical logic within the allusion. The abstract solutions are core principles, rules, or action patterns abstracted from the allusion, either manually or automatically annotated, and are usually represented as vectors or rule templates.

[0049] The encoding unit in the analogy reasoning engine receives the semantic vector and temporary graph of the new context, quickly filters out the candidate allusions with the most similar philosophical structures from the knowledge base, and performs fine-grained graph structure alignment to establish correspondences between elements. The retrieval and alignment unit adopts a two-level retrieval: the first round is based on vector similarity, such as cosine similarity, to quickly recall the top-K candidates; the second round is based on graph similarity algorithms, such as graph kernel methods and graph neural network matching, to rank the candidates and obtain the optimal alignment mapping. The solution generation unit in the analogy reasoning engine, based on the optimal alignment mapping, instantiates the abstract solution template of the matching allusion onto the specific elements of the new context, generating specific and actionable decision suggestions or action descriptions. This can be based on conditional text generation models, such as large language models with alignment information as conditions, or rule-based template filling engines.

[0050] The specific embodiments described herein are merely illustrative of the spirit of the invention. Those skilled in the art to which this invention pertains may make various modifications or additions to the described specific embodiments or use similar methods to substitute them, without departing from the spirit of the invention or exceeding the scope defined by the appended claims.

[0051] Although this paper makes extensive use of terms such as philosophical semantic encoding module, structured allusion knowledge base, and analogical reasoning engine, the possibility of using other terms is not excluded. The use of these terms is merely for the convenience of describing and explaining the essence of this invention; interpreting them as any additional limitation would contradict the spirit of this invention.

Claims

1. A tacit knowledge reasoning and analogical decision-making method for AGI, characterized in that, Includes the following steps: S1: Training the philosophical semantic understanding model: Using a dedicated corpus of the target philosophical system, a pre-trained language model is trained; during the training process, in addition to the masked language modeling task, a philosophical relation prediction task is added. The task requires the model to predict the predefined philosophical relation type between two concepts based on a context containing at least two philosophical concepts, so that the pre-trained language model internalizes the deep logical relations of the target philosophical system and forms a philosophical semantic encoder. S2: Construct a structured allusion knowledge base: Collect multiple allusion texts containing complete narratives and solutions; for each allusion text, perform the following processing: S21: Use the philosophical semantic encoder to process the allusion text and generate a semantic vector representation of the allusion; S22: Identify key philosophical concept entities and their relationships from the text of the allusion, and construct a sub-graph representing the internal philosophical logic structure of the allusion; S23: Extract or summarize the core solution principles contained in the allusions to form an abstract solution description; The semantic vector representation, the sub-graph, and the abstract solution description are associated and stored to form a structured allusion entry; S3: Execution Context - Analogical Reasoning: When AGI faces a target decision-making context, it performs the following steps: S31: Context Encoding: The philosophical semantic encoder is used to process the descriptive text of the target decision context to generate a context semantic vector and a context temporary sub-graph. S32: Multi-level allusion retrieval: First, a preliminary screening is performed based on the similarity between the context semantic vector and the semantic vector representation of each allusion entry in the knowledge base; then, a fine ranking is performed based on the structural similarity between the context temporary subgraph and the subgraph of the allusion entries obtained from the preliminary screening, to obtain at least one best matching allusion entry. S33: Graph Alignment and Solution Generation: Align the temporary subgraph of the scenario with the subgraph of the best-matching allusion entry to obtain the element mapping relationship; based on the mapping relationship, adapt the abstract solution description of the best-matching allusion entry to the specific elements of the target decision scenario to generate a targeted final decision output.

2. The tacit knowledge reasoning and analogical decision-making method for AGI according to claim 1, characterized in that, The predefined philosophical relationship type in step S1 is selected from at least two of the following: opposition relationship, governing relationship, derivative relationship, and embodying relationship.

3. The tacit knowledge reasoning and analogical decision-making method for AGI according to claim 1, characterized in that, The philosophical concept entities and relation types used in step S22 to construct the subgraph are consistent with the types defined in the philosophical relation prediction task in step S1.

4. The tacit knowledge reasoning and analogical decision-making method for AGI according to claim 1, characterized in that, In step S32, the fine ranking based on structural similarity is performed by calculating the graph edit distance or graph kernel similarity between the temporary subgraph of the context and the candidate allusion subgraph, and then ranking them accordingly.

5. The tacit knowledge reasoning and analogical decision-making method for AGI according to claim 1, characterized in that, The graph structure alignment in step S33 is achieved using a graph matching algorithm. The optimization objective of this algorithm is to maximize the semantic similarity of the mapped node pairs and maintain the consistency of the relation structure.

6. The tacit knowledge reasoning and analogical decision-making method for AGI according to claim 1, characterized in that, The abstract solution description is stored in the form of parameterized templates or structured rules. The adaptation process in step S33 is to replace the template variables with specific references in the context according to the element mapping relationship.

7. The tacit knowledge reasoning and analogical decision-making method for AGI according to claim 1, characterized in that, The method is integrated into the decision loop of an AGI mental model; when the AGI mental model evaluates through its regular cognitive module and determines that the entropy value of the current decision situation is higher than a threshold or that it is trapped in a binary opposition dilemma, it automatically triggers the execution of step S3, which further includes: S34: Select the top N allusion entries with the highest structural similarity as a partial matching set; locally align the temporary subgraph of the context with the subgraph of each allusion entry in the partial matching set, and extract the corresponding solution fragments; based on the topological structure of the temporary subgraph of the context, combine the multiple solution fragments to generate a first candidate solution; S35: If the partial matching set is empty, or step S34 fails, then based on the predefined philosophical meta-axioms, logical deduction is performed on the dilemma represented by the temporary sub-graph of the situation to generate a second candidate solution that conforms to the meta-axioms; if multiple candidate solution vectors are generated, and the conflict intensity between these vectors exceeds a preset second conflict threshold, then a univariate mental arbitration process is triggered; the univariate mental arbitration process is based on the ultimate principle of making the system state converge to the philosophical origin, and arbitrates the multiple candidate solution vectors to output a final solution vector; S36: The description text of the target decision-making situation, the temporary sub-graph of the situation, and the final solution are constructed into a new structured allusion entry and stored in the structured allusion knowledge base.

8. A tacit knowledge reasoning and analogical decision-making system for AGI, characterized in that, include: A philosophical semantic encoding module is used to implement the philosophical semantic encoder obtained in step S1 of claim 1; A structured allusion knowledge base is used to store multiple structured allusion entries constructed in step S2 as described in claim 1; An analogy reasoning engine is configured to perform step S3 as described in claim 1; The analogy reasoning engine further includes: The encoding unit is used to call the philosophical semantic encoding module to process the input context; The retrieval alignment unit is used to perform multi-level allusion retrieval and map alignment; The scheme generation unit is used to generate the final decision output based on the alignment results.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements a tacit knowledge reasoning and analogical decision-making method for AGI as described in any one of claims 1 to 7.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements a tacit knowledge reasoning and analogical decision-making method for AGI as described in any one of claims 1 to 7.

Images