Algorithmic Fairness Analysis System with PII Anonymization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques require customized analysis of data and handling of sensitive information, such as personally identifiable information (PII) and demographic data, to assess the fairness or bias of algorithms used in evaluating individuals or entities.
Innovation Solution
A computer-implemented method and system that processes data to analyze bias by receiving a first data set with PII, inferring traits from the data, creating a second data set without PII, and analyzing this data set to determine the fairness of an algorithm or process, while ensuring secure handling and anonymization of sensitive information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If customized analysis of data including PII is performed to assess algorithm fairness, then measurement precision of fairness metrics is improved, but loss of information occurs due to removal of sensitive data
Solution Approach 1:
The patent segments the analysis process into two distinct phases: (1) initial analysis with PII to establish baseline fairness metrics and identify potential biases, and (2) subsequent analysis without PII to validate findings and maintain privacy. This segmentation allows the system to achieve measurement precision in the first phase while preventing information loss in the second phase through controlled data handling.
Solution Approach 2:
The system performs preliminary analysis actions by first processing data with PII to establish reference fairness metrics before removing sensitive information. This preliminary action captures essential fairness patterns that can then be analyzed in subsequent phases without PII, ensuring measurement precision is achieved before information loss occurs.
2Reliability
If PII is retained in data sets for fairness analysis, then reliability of fairness assessment is improved, but object-generated harmful factors increase due to privacy risks
Solution Approach 1:
The patent introduces an intermediary controlled environment that acts as a mediator between PII-containing data and fairness analysis processes. This intermediary environment enables reliable fairness assessment by controlling access to PII through security protocols, while preventing direct exposure of sensitive information that would create privacy breach risks.
Solution Approach 2:
The system creates an inert or controlled environment for processing PII during fairness analysis. This environment isolates sensitive data from external access and potential misuse, allowing reliability of fairness assessment to be maintained while eliminating the harmful factors of privacy breaches through environmental control.
3Reliability
If secure handling procedures are implemented for sensitive data, then reliability of the system is improved, but device complexity increases
Solution Approach 1:
The patent implements universal security protocols that serve multiple functions: they protect PII during storage, control access during analysis, and manage data transformation processes. This multi-functionality achieves reliable data security assurance while reducing device complexity by consolidating security operations into a unified framework rather than requiring separate complex systems for each security concern.
Data Source
AI summary
A computer-implemented method for processing data may comprise receiving, from a user, a first data set including at least one item of personally identifiable information (“PII”) for a plurality of individuals, wherein the first data set includes information related to an outcome of an algorithm or process; analyzing the received first data set to infer at least one trait of the plurality of individuals; storing a second data set in a data store, wherein the second data set includes the at least one trait, and wherein the second data set does not include the at least one item of PII; analyzing the second data set to determine a fairness of the algorithm or process; and providing a report to the user related to the determined fairness of the algorithm or process.