Consolidated Explainability for AI Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current AI systems lack comprehensive explainability, as explanations from different components often conflict or are inconsistent, making it difficult to consolidate explanations and resolve conflicts, especially in end-to-end AI systems handling complex tasks like energy management in communication networks.
Innovation Solution
A computer-implemented method and apparatus that acquire and combine explanations from multiple sources, forming a set of constraints to generate a planning problem, which determines a consolidated explanation by resolving conflicts and finding a consistent solution using a combined planner and explainer module.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If multiple explanations are provided from different AI components, then the coverage of explanation is improved, but the consistency and reliability of the explanation deteriorates due to conflicts between explanations
Solution Approach 1:
The patent introduces a higher-level reasoner that acts as an intermediary to receive, evaluate, and reconcile explanations from multiple subsystems. This mediator consolidates the explanations and resolves conflicts by determining whether the overall explanation is acceptable or identifying conflicting parts, thereby maintaining both comprehensive coverage and consistency.
Solution Approach 2:
The patent merges multiple explanations from different subsystems into a unified explanation structure. By combining the explanations at the higher-level reasoner and using a planning problem formulation, the system integrates diverse component explanations into a single coherent narrative that addresses the overall intent.
2Measurement precision
If a component-oriented explanation approach is used, then the detail level of explanation is improved, but the overall system-level explanation capability deteriorates
Solution Approach 1:
The patent segments the explanation task into component-level explanations generated by individual subsystems and a system-level explanation synthesized by the higher-level reasoner. This segmentation allows detailed component explanations to be produced independently while the higher-level system integrates them into a coherent system-level narrative.
Solution Approach 2:
The patent adds a hierarchical dimension to the explanation structure by introducing a higher-level reasoner that operates above the component level. This dimensional addition enables the system to simultaneously maintain detailed component explanations and generate overarching system-level explanations without increasing component complexity.
3Loss of information
If explanations from multiple subsystems are collected, then the comprehensiveness of explanation is improved, but the conflict resolution difficulty increases
Solution Approach 1:
The higher-level reasoner serves as a mediator that systematically evaluates explanations from multiple subsystems against the overall intent and goal state. This intermediary structure provides a organized approach to conflict resolution by comparing explanations in the context of the planning problem, reducing the complexity of reconciling multiple perspectives.
Solution Approach 2:
The system implements feedback mechanisms where the higher-level reasoner evaluates component explanations and determines acceptability. This feedback loop allows the system to identify conflicts, request clarification or alternative explanations from subsystems, and iteratively refine the overall explanation until consistency is achieved.
Data Source
AI summary
There is provided a method for consolidating explanations associated with actions proposed based on a current state of a system and an intent. The method includes acquiring first and second explanations, the first and second explanations being associated with a proposed action or with different actions, wherein each of the first and second explanations includes one or more constraints, combining constraints from the first and second explanations to form a set of constraints D, generating a planning problem P=<K, A, I, G, Cost>, wherein K consists of a set of predicates F and the set of constraints D, wherein A represents a set of possible actions, I represents an initial state of the system, G represents a goal state of the system, and Cost represents cost values associated with each constraint, and determining a solution for the planning problem.