AI Bias Reduction via Open-Closed Source Comparison
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Open source and closed source AI systems may exhibit biases in their search results, which can lead to undesirable outcomes, such as intentional biases favoring specific entities.
Innovation Solution
A bias reduction AI system is introduced, which includes interfaces to both open source and closed source AI systems. A bot performs automated searches using both systems with the same terminology, and a bias reduction engine compares the results to detect any predetermined biases. The system then shares data between the AI systems to reduce biases and assigns scores and certificates based on bias levels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If open source AI systems or closed source AI systems are used independently, then each system can operate with its own data and algorithms, but biases may be injected into search results that cannot be detected or reduced
Solution Approach 1:
The patent combines multiple AI systems (open source and closed source) into a unified architecture where they interact and compare results. The bias detection system merges outputs from different sources to identify discrepancies that indicate bias injection, thereby improving reliability while managing complexity through structured integration.
Solution Approach 2:
The patent introduces intermediary components including a bias detection engine and a result aggregation module that mediate between individual AI systems and the final output. These intermediaries analyze results from multiple sources, detect biases through comparison, and produce corrected aggregated results, acting as buffers that improve reliability without requiring complete system redesign.
2Measurement precision
If multiple AI systems are integrated to detect biases, then bias detection capability is improved, but the system complexity and computational overhead increase
Solution Approach 1:
The patent segments the bias detection function into distinct modular components: individual AI systems generate results independently, a comparison module identifies discrepancies, and a bias determination module classifies the type of bias detected. This segmentation allows precise bias detection through specialized subsystems while managing overall system complexity through clear functional separation and standardized interfaces.
Data Source
AI summary
Artificial intelligence (AI) systems may manipulate search results about an entity to inject a predetermined bias into the search results. A bias reduction artificial intelligence (AI) system and method may perform the same automated search about the entity on open source and closed source AI systems. The AI-generated search results for the open and closed source AI systems may be compared to determine differences in results. The differences may be analyzed to determine attempts by the AI systems to manipulate search results about the entity to inject predetermined bias. If an attempt at predetermined bias is identified, the bias reduction AI system may reduce the predetermined bias by sharing data about the entity between the open source AI system and the closed source AI system to cause the system to use machine learning to update the algorithms and the first and second data sets. Incentives may also be provided to the open and closed source AI systems to reduce biases.