Explainable AI Incident Identification System
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional investigation tools are inefficient and prone to errors when processing witness statements, as they require manual querying and review of numerous images based on incomplete and uncertain statements, especially when multiple statements with variances are involved.
Innovation Solution
An interactive investigation tool that uses machine learning algorithms to analyze text data from witness statements, extract features, compute similarity metrics with explainability labels from media data, and identify relevant records such as LPR records, providing explanations for query results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If manual querying and review of images is used based on witness statements, then the investigation process can be performed with conventional tools, but the process becomes time-consuming and error-prone
Solution Approach 1:
The patent replaces manual mechanical querying and review processes with an automated machine learning-based system. The machine learning algorithm automatically processes witness statements, extracts relevant features, queries databases, and retrieves matching images without human intervention in the core processing tasks, thereby eliminating time consumption and reducing errors associated with manual operations
Solution Approach 2:
The system enables self-service by allowing the machine learning algorithm to autonomously perform the investigation tasks. The algorithm automatically extracts features from witness statements, queries relevant databases, and retrieves matching images without requiring continuous human guidance or manual review of each step, significantly improving productivity while reducing time loss
2Reliability
If multiple queries are made based on incomplete witness statements, then more comprehensive search can be performed, but the process becomes more cumbersome and error-prone
Solution Approach 1:
The patent segments the witness statement into multiple features through automated extraction. The machine learning algorithm divides the text data into relevant feature components (such as temporal, spatial, and object characteristics) and processes each feature separately to query the database, which simplifies the overall querying process while maintaining comprehensive search coverage and improving reliability
Solution Approach 2:
The machine learning algorithm acts as an intermediary between the incomplete witness statements and the database queries. It translates the textual, potentially incomplete statements into structured feature representations that can be reliably queried against the database, thereby improving the accuracy of incident identification without increasing process complexity
3Loss of information
If conventional investigation tools are used, then the system structure remains simple, but the tools fail to provide explanation of query results
Solution Approach 1:
The patent implements feedback mechanisms where the machine learning algorithm provides explanations for its query results. The system feeds back information about why certain images were selected, including the feature matching rationale and confidence scores, thereby reducing information loss and providing transparency without excessively increasing system complexity
Solution Approach 2:
The machine learning algorithm serves multiple functions: it processes witness statements, extracts features, queries databases, retrieves images, and provides explanations. This multi-functionality consolidates what would otherwise require separate systems into a single integrated solution, providing comprehensive functionality including result explanations while managing system complexity through unified processing
Data Source
AI summary
There is provided a computer-based identification method comprising, at a computing device having at least one machine learning algorithm operating therein, the at least one machine learning algorithm configured to provide an explanation of results produced thereby, receiving text data, the text data providing a textual description relating to an incident involving at least one object, extracting a first plurality of features from the text data, computing one or more similarity metrics between the first plurality of features and a respective second plurality of features, the second plurality of features derived from a plurality of explainability labels produced by the at least one machine learning algorithm, the plurality of explainability labels associated with media data obtained from one or more media devices deployed at one or more locations encompassing a location of the incident, and identifying the incident based on the one or more similarity metrics.