Composite symbolic and non-symbolic artificial intelligence system for advanced reasoning and automation
A composite AI system and method for advanced reasoning and automation that integrates symbolic knowledge graphs and algorithms with non-symbolic, or connectionist, models such as neural embeddings. A hierarchical architecture enables dynamically distributed, cooperative reasoning through layperson and expert-led challenge-based verification, model blending, model fitness and retraining and selection, comprehensive feedback loops at individual model or model blend or process flow with or without supervision, and specialized routing of processing to account for various operational risk, regulatory, legal, privacy, or economic considerations. Models, datasets, knowledge bases, simulations and simulation components, and embeddings are iteratively refined using knowledge graph elements and model, process, simulation or flow/process optimal hyperparameters which are recorded and tracked. Extraction of symbolic representations from connectionist models links them to curated ontologies of facts and principles.
1 . A computing system for composite artificial intelligence employing an advanced reasoning platform, the computing system comprising:
one or more hardware processors configured for:
generating vector embeddings or representations of input data using a plurality of neural network models trained on large datasets;
storing and retrieving symbolic facts, entities, and relations into a hierarchical knowledge graph using an ontology or schema that defines the types of entities, relationships, and attributes relevant to each of the plurality of neural network models' respective domains;
routing data to neural network models or knowledge sources based on predefined rules or learned policies derived from reinforcement learning;
defining a hierarchy of reasoning tasks and subtasks;
managing and routing processing of reasoning tasks and subtasks to a specific neural network model of the plurality of neural network models based on model-specific certainty thresholds that are set for each task or subtask and iterative verification mechanisms designed to test robustness and reliability of decisions by the specific neural network model for the reasoning tasks or subtasks;
iteratively refining vector embeddings or representations through reinforcement learning by retrieving relevant facts, entities, and relationships from the hierarchical knowledge graph, thereby providing retrieved contextual information;
updating the vector embeddings or representations by incorporating the retrieved contextual information through iterative feedback loops;
combining outputs of the plurality of neural network models using dynamically learned or rule-based weighting schemes;
aligning and harmonizing the vector embeddings or representations across different data modalities;
optimizing hyperparameters of the plurality of neural network models during embedding generation and refinement using information-theoretic guidance principles;
extracting symbolic knowledge from the vector embeddings or representations to generate extracted symbolic representations; and
mapping the extracted symbolic representations to existing concepts and relationships in the hierarchical knowledge graph.
2 . The computing system of claim 1 , wherein updating the vector embeddings or representations comprises applying attention mechanisms, graph convolutions, or knowledge-aware language models to combine the embeddings with the contextual data through reinforcement learning optimization.
3 . The computing system of claim 1 , wherein the one or more hardware processors are further configured for:
designing a composite model architecture that leverages the strengths of individual models from the plurality of neural network models;
developing dynamic weighting schemes that adjust the contribution of each model based on various factors through reinforcement learning; and
implementing techniques including Bayesian model averaging, mixture of experts, or ensemble learning to optimize the model combination architecture.
4 . The computing system of claim 3 , wherein the various factors comprise uncertainty, task complexity, or domain relevance.
5 . The computing system of claim 1 , wherein the one or more hardware processors are further configured for ensuring the security, licensing compliance, provenance tracking, and collaborative development of AI models by implementing secure communication protocols and access control mechanisms.
6 . The computing system of claim 1 , wherein aligning and harmonizing the vector embeddings or representations comprises:
developing algorithms that can synchronize and align the representations of multiple input modalities through the hierarchical knowledge graph;
applying techniques including cross-modal attention, multi-modal fusion, or joint embedding spaces with reinforcement learning optimization; and
utilizing domain-specific knowledge from the hierarchical knowledge graph to ensure the generated representations are consistent and realistic across modalities.
7 . The computing system of claim 1 , wherein mapping the extracted symbolic representations comprises hierarchical ontology alignment and linking methods.
8 . The computing system of claim 1 , wherein the hierarchical knowledge graph comprises multiple interconnected knowledge graph layers with domain-specific organization.
9 . A computer-implemented method executed on an advanced reasoning platform for composite artificial intelligence, the computer-implemented method comprising:
generating vector embeddings or representations of input data using a plurality of neural network models trained on large datasets;
storing and retrieving symbolic facts, entities, and relations into a hierarchical knowledge graph using an ontology or schema that defines the types of entities, relationships, and attributes relevant to each of the plurality of neural network models' respective domains;
routing data to neural network models or knowledge sources based on predefined rules or learned policies derived from reinforcement learning;
defining a hierarchy of reasoning tasks and subtasks;
managing and routing processing of reasoning tasks and subtasks to a specific neural network model of the plurality of neural network models based on model-specific certainty thresholds that are set for each task or subtask and iterative verification mechanisms designed to test robustness and reliability of decisions by the specific neural network model for the reasoning tasks or subtasks;
iteratively refining vector embeddings or representations through reinforcement learning by retrieving relevant facts, entities, and relationships from the hierarchical knowledge graph, thereby providing retrieved contextual information;
updating the vector embeddings or representations by incorporating the retrieved contextual information through iterative feedback loops;
combining outputs of the plurality of neural network models using dynamically learned or rule-based weighting schemes;
aligning and harmonizing the vector embeddings or representations across different data modalities;
optimizing hyperparameters of the plurality of neural network models during embedding generation and refinement using information-theoretic guidance principles;
extracting symbolic knowledge from the vector embeddings or representations to generate extracted symbolic representations; and
mapping the extracted symbolic representations to existing concepts and relationships in the hierarchical knowledge graph.
10 . The computer-implemented method of claim 9 , wherein updating the vector embeddings or representations comprises applying attention mechanisms, graph convolutions, or knowledge-aware language models to combine the embeddings with the contextual data through reinforcement learning optimization.
11 . The computer-implemented method of claim 9 further comprising:
designing a composite model combination architecture that leverages the strengths of individual models from the plurality of neural network models;
developing dynamic weighting schemes that adjust the contribution of each model based on various factors through reinforcement learning; and
implementing techniques including Bayesian model averaging, mixture of experts, or ensemble learning to optimize the model combination architecture.
12 . The computer-implemented method of claim 11 , wherein the various factors comprise uncertainty, task complexity, or domain relevance.
13 . The computer-implemented method of claim 9 further comprising ensuring the security, licensing compliance, provenance tracking, and collaborative development of AI models by implementing secure communication protocols and access control mechanisms.
14 . The computer-implemented method of claim 9 , wherein aligning and harmonizing the vector embeddings or representations comprises:
developing algorithms that can synchronize and align the representations of multiple input modalities through the hierarchical knowledge graph;
applying techniques including cross-modal attention, multi-modal fusion, or joint embedding spaces with reinforcement learning optimization; and
utilizing domain-specific knowledge from the hierarchical knowledge graph to ensure the generated representations are consistent and realistic across modalities.
15 . The computer-implemented method of claim 9 , wherein mapping the extracted symbolic representations comprises hierarchical ontology alignment and linking methods.
16 . The computer-implemented method of claim 9 , wherein the knowledge base comprises multiple interconnected knowledge graph layers with domain-specific organization.