Imagine an AI acting like a biased judge who makes unfair decisions based on a person's background or appearance. This work creates a digital toolkit that acts like a set of corrective glasses, helping the AI see everyone fairly. It ensures that automated systems don't accidentally discriminate against people based on multiple traits at once.
AI systems often discriminate against minority groups based on gender, race, or age, leading to legal risks and unfair service denial. Current tools are often too simple and cannot handle complex data like images or networks.
The MAI-Bias toolkit, an open-source collection of tools and methods to discover and fix multi-discrimination in tabular, network, and visual AI data.
If you are a digital lending platform dealing with unfair loan rejections — this project developed a fairness-aware system that reduces bias in credit scoring and debt repayment. This ensures a fairer process for applicants across different demographics.
If you are an identity verification provider dealing with high failure rates for minority groups in KYC — this project developed tools for face verification systems. This improves access to online services for underrepresented groups.
If you are an academic search engine dealing with visibility gaps for certain scholars — this project developed methods to measure and mitigate intersectional biases in citations. This creates a more equitable academic network.
Based on available project data, the tools are provided as open-source methods and an integrated open-source toolkit called MAI-Bias, meaning there is no direct purchase price mentioned.
The project tested its solutions through pilots in finance, identity verification, and academic evaluation, suggesting it is designed for real-world industrial application.
The project results, including the MAI-Bias toolkit, are released as open-source, though specific license types are not detailed in the provided text.
The tools promote the accountability of AI systems regarding protected attributes, which helps businesses meet legal requirements to avoid discrimination.
The MAI-Bias toolkit combines new methods with third-party fairness libraries, suggesting it is built to work alongside existing AI components.
The project features a strong commercial orientation with a 38% industry ratio, including 5 industry partners and 6 SMEs. The collaboration is geographically diverse, spanning 11 countries, and blends technical AI expertise with social science and ethics specialists to ensure the tools are grounded in real-world human needs.
Contact the Centre for Research and Technology Hellas (CERTH) in Greece.
Contact us to explore how to integrate the MAI-Bias open-source toolkit into your AI pipeline.
