Algorithmic Bias Evaluation for Risk Assessment Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current systems face challenges in efficiently reviewing a large set of artificial intelligence/machine learning (AI/ML) risk assessment models for biases, particularly in compliance with fair lending laws, due to the practical limitations of conventional technological systems in performing quantitative reviews across all models.
Innovation Solution
A method and system that utilize a disparate impact analysis (DIA) service to assess biases in AI/ML risk assessment models by generating model score output files, obtaining government monitoring information, calculating DIA results, and determining if model features exceed predetermined thresholds, with the ability to generate recommended changes and update models to align with compliance standards.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional technological systems are used to review AI/ML models for bias, then in-depth quantitative review can be performed, but the review process becomes impractical when applied to large sets of models due to time and resource constraints
Solution Approach 1:
The patent segments the model review process into multiple stages: automated pre-screening using machine learning to identify high-risk models, followed by focused quantitative analysis only on those models. This segmentation allows the system to maintain measurement precision for models that require it while dramatically increasing overall productivity by automating the triage process.
Solution Approach 2:
The patent introduces an intermediary automated analysis system that acts as a mediator between the large set of models and the detailed quantitative review process. This intermediary performs initial bias detection and risk assessment, filtering models that require in-depth review and identifying those that can be approved through automated analysis, thereby resolving the contradiction between thorough review and scalability.
2Productivity
If automated analysis is used to review large sets of models, then productivity increases, but measurement precision and depth of bias detection may be reduced
Solution Approach 1:
The patent implements a dynamic review system that adapts the depth of analysis based on risk levels. Automated analysis provides rapid initial assessment for all models, while the system dynamically identifies high-risk cases that require deeper quantitative analysis. This dynamic approach ensures measurement precision is applied where most needed while maintaining high overall productivity.
Solution Approach 2:
The patent performs preliminary automated bias detection and risk assessment before committing resources to detailed quantitative analysis. This preliminary action filters out low-risk models that can be approved through automated review, ensuring that measurement precision resources are concentrated on models that truly require it, thus maintaining both productivity and accuracy.
3Reliability
If detailed quantitative review is performed on every model, then bias detection accuracy improves, but time consumption and computational resources increase significantly
Solution Approach 1:
The patent applies local quality by providing different levels of review depth to different models based on their risk characteristics. High-risk models receive detailed quantitative analysis with high reliability, while low-risk models receive streamlined automated review. This ensures bias detection reliability is optimized for models that need it without unnecessarily consuming time on models that don't require such thorough analysis.
Solution Approach 2:
The patent changes the parameter of review depth based on model risk characteristics. By adjusting the intensity and scope of analysis parameters according to each model's features and risk profile, the system achieves high reliability for problematic models while reducing time consumption for models that present minimal bias risk.
Data Source
AI summary
A method, system, and computer-readable storage medium storing instructions, for assessing biases of a risk assessment model. The method, system, and computer-readable storage medium storing instructions, comprising: receiving, by a processor, a first model; generating, by the processor, a first model score output file by evaluating the first model; transmitting, by the processor, the first model score output file to a disparate impact analysis (DIA) service; utilizing, by the processor, the DIA service to obtain, from a government monitoring information (GMI) database, first GMI data that corresponds to a first set of ethics and compliance initiative (ECI) information; further utilizing, by the processor, the DIA service to calculate first DIA results by analyzing the first set of ECI information and the first GMI data; and determining, based on the first DIA results, whether any features of the first model exceed at least one predetermined threshold.