Counterfactual Explainability for Bias in ML Decisioning Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning-based decisioning models are often opaque, leading to trust issues and fairness challenges due to their complexity and lack of transparency.
Innovation Solution
A system and method that includes obtaining a decisioning dataset, detecting favorable records closest to unfavorable records, computing bias intensity metrics, generating an explainability artifact, and displaying it to highlight factors contributing to bias, enabling model reconfiguration to mitigate bias.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If machine learning-based decisioning models are used to improve decision-making efficiency, then productivity is improved, but the black box nature of these models causes loss of information about decision-making processes and undermines user trust
Solution Approach 1:
The patent introduces counterfactual explanations as an intermediary mechanism that bridges the gap between the black box decisioning model and users. These explanations provide insights into how decisions were reached by showing what changes in input features would have led to different outcomes, thereby maintaining productivity while reducing information loss and enhancing user trust
Solution Approach 2:
The system implements feedback loops where decision outcomes and their explanations are fed back to users. This allows users to understand the reasoning behind decisions and potentially correct biased or erroneous outcomes, thereby maintaining efficiency while improving transparency and trust in the decision-making process
2Measurement precision
If complex machine learning models are deployed to improve decision accuracy, then measurement precision is improved, but device complexity increases and fairness becomes difficult to ensure
Solution Approach 1:
Counterfactual explanations serve as an intermediary that simplifies the interpretation of complex model decisions. Instead of requiring users to understand the complex internal workings of the model, the system provides intuitive explanations showing what specific changes would alter outcomes, thereby maintaining accuracy while reducing perceived complexity
Solution Approach 2:
The patent segments the complex decision-making process into interpretable components through counterfactual explanations. By breaking down decisions into specific feature changes and their impacts, the system maintains measurement precision while making the decision process more understandable and easier to audit for fairness
3Ease of operation
If machine learning models operate as black boxes to maintain simplicity of operation, then ease of operation is improved, but user trust deteriorates due to lack of transparency
Solution Approach 1:
The counterfactual explanation system acts as an intermediary layer that maintains the simplicity of model deployment while building user trust. Users can continue to interact with the model through simple interfaces while receiving explanatory feedback that enhances their understanding and trust in the decision-making process
Solution Approach 2:
The system provides self-service explanations automatically generated by the model itself. This maintains ease of operation as users don't need to separately query or interpret complex model internals, while simultaneously building trust through transparent, automatically provided explanations of decision rationale
Data Source
AI summary
A system, method, and computer-program product includes obtaining a decisioning dataset comprising a plurality of favorable decisioning records and at least one unfavorable decisioning record; detecting, via a machine learning algorithm, a favorable decisioning record of the plurality of favorable decisioning records that has a vector value closest to a vector value of the unfavorable decisioning record; executing a counterfactual assessment between the favorable decisioning record and the unfavorable decisioning record; generating an explainability artifact based on one or more bias intensity metrics to explain a bias in a machine learning-based decisioning model; and in response to generating the explainability artifact, displaying the explainability artifact in a user interface.