Explainable AI Ontology Reasoning Provenance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing explainable AI systems are inefficient in providing detailed analysis of their decision-making processes, often lacking sufficient granularity and confidence levels, which can lead to inadequate user trust and inability to identify errors in AI-driven recommendations, especially in critical contexts like healthcare and finance.
Innovation Solution
The development of an AI tool that uses data-driven ontology and inferencing rule construction to provide granular explanations, incorporating unsupervised learning and both forward and backward chaining techniques to derive logical reasoning and confidence levels, enabling users to understand and verify AI decisions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional machine learning techniques are used to produce output labels from feature vectors, then the system can make decisions efficiently, but the system lacks detailed documentation of how and why the label was produced
Solution Approach 1:
The patent segments the decision-making process into distinct components: feature extraction, label generation, and provenance tracking. By separating these functions, the system can maintain efficient automated decision-making while simultaneously documenting the reasoning process through forward and backward chaining mechanisms that trace the logical path from features to labels.
Solution Approach 2:
The patent introduces provenance information as an intermediary element that connects the input features to the output labels. This intermediary layer captures the reasoning process without interfering with the efficient operation of the machine learning model, allowing users to trace and verify the logical steps taken to reach a decision.
2Reliability
If existing explainable AI techniques are used, then some explanation is provided, but the granularity of verification is insufficient for users to audit the recommendation
Solution Approach 1:
The patent adds a new dimension to explanation by introducing provenance tracking that operates alongside the traditional feature-label mapping. This additional dimension provides multi-level verification capability, allowing users to examine the reasoning process at different granularities from high-level decision logic to specific feature contributions, thereby enhancing auditability without excessive complexity.
3Loss of information
If machine reasoning techniques based on forward or backward chaining are used, then some reasoning provenance is provided, but complete provenance of the reasoning process is not achieved
Solution Approach 1:
The patent merges forward chaining and backward chaining techniques into a unified provenance tracking system. Forward chaining captures the deductive reasoning from features to labels, while backward chaining traces the necessary conditions from labels back to features. By combining these complementary approaches, the system achieves complete reasoning provenance that documents both the forward derivation and backward verification of decisions.
4Ease of operation
If accuracy or confidence level alone is used to determine decision quality, then simple evaluation is possible, but it is insufficient to determine whether human intervention is needed as the value of making the right or wrong decision varies greatly
Solution Approach 1:
The patent extends the evaluation framework by introducing additional parameters beyond accuracy and confidence level. The provenance information enables assessment of reasoning validity, feature relevance, and logical consistency. These additional parameters allow the system to differentiate between high-confidence incorrect decisions and low-confidence correct decisions, providing a more nuanced basis for determining when human intervention is necessary.
Data Source
AI summary
Examples of artificial intelligence-based reasoning explanation are described. In an example implementation, a knowledge model having a plurality of ontologies and a plurality of inferencing rules is generated. Once the knowledge model is generated, based on a real-world problem, a knowledge model from amongst various knowledge models is selected to be used for resolving a real-world problem. The data procured from the real-world problem is clustered and classified into an ontology of the determined knowledge model. Inferencing rules to be used for deconstructing the real-world problem are identified, and a machine reasoning is generated to provide a hypothesis for the problem and an explanation to accompany the hypothesis.