Semantic AI Explanations Using Knowledge Graph Traceability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI models in complex domains like manufacturing lack transparency, making it difficult for human operators and domain experts to understand the reasoning behind predictions, leading to a lack of confidence and increased resource consumption in model validation.
Innovation Solution
A method and system that generate semantic explanations by leveraging knowledge graphs to trace predictions through both problem and solution domains, providing comprehensive and context-aware explanations that bridge the gap between mathematical operations and real-world concepts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If complex AI models are used to improve prediction accuracy, then prediction accuracy is improved, but model transparency deteriorates
Solution Approach 1:
The patent introduces an explanation generation module as an intermediary between the complex AI model and the user. This module translates the internal workings of the complex model into understandable explanations using domain knowledge graphs, allowing the system to maintain high prediction accuracy while improving transparency through the mediating explanation layer.
2Measurement precision
If complex AI models are used to improve prediction accuracy, then prediction accuracy is improved, but validation effort increases
Solution Approach 1:
The patent performs validation-related tasks in advance by generating explanations during the model operation phase. The explanation generation module proactively creates transparency artifacts that can be used for validation, eliminating the need for separate validation efforts and reducing time loss.
3Loss of information
If domain knowledge graphs are integrated to improve explanation quality, then explanation quality is improved, but system complexity increases
Solution Approach 1:
The domain knowledge graph serves multiple functions within the system: it provides context for interpreting predictions, generates explanations, validates model behavior, and bridges domain terminology with model terminology. This multi-functionality justifies the added complexity by delivering comprehensive benefits across multiple system aspects.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
Disclosed is a method (100, 600) for providing semantic explanations (614) for predictions made by an artificial intelligence (AI) model (610). The method executes code (608) to create/train an AI model using a solution description that maps problem domain knowledge graph (602) attributes to solution domain knowledge graph (604) features. The method receives a prediction output, traces it through the solution domain knowledge graph (604) to identify contributing inputs and intermediate representations, generates a feature mapping matrix, and identifies relevant entities and relationships from the problem domain knowledge graph. The method outputs an explanation (614) comprising at least one of: identified contributing input features and weights, traced path through hidden layers, and relevant problem domain entities and relationships. This approach provides context-rich, domain-specific explanations that enhance understanding of predictions of the AI model.