An Ethical AI Framework for Identifying, Auditing and Mitigating Bias in AI Models

Abstract

Artificial Intelligence systems are extensively used across many domains such as healthcare, finance, and recruitment. Many of these systems are trained on sensitive or biased datasets, that results in unethical and unfair decisions taken by the AI models. Additionally, this makes user data accessible and puts privacy at risk. In order to overcome these challenges in AI systems, this study proposes an interactive and open-source platform that identifies, audits and mitigates the biases detected and gives a clear and fair understanding about the problems that have been resolved. Furthermore, the platform delivers an explainable scorecard and a community crowdsourcing option along with regeneration of synthesized data, privacy risk evaluation, proxy bias detection that is implemented using Cramér's V correlation measure, fairness audits, and retraining capabilities. The system provides a clear visual report with suggestions and is user-friendly. The system allows users to clean their datasets ethically, which in turn fosters the development of ethical and fair AI solutions.

Ähnliche Arbeiten