Multi-Objective Approach to Balance Fairness and Accuracy

As Artificial Intelligence systems are being deployed in multiple domains, ensuring they exhibit fair and just behaviour is a critical challenge. Multi-objective optimization offers a robust framework for addressing this challenge by simultaneously optimizing conflicting objectives, such as fairness and accuracy. In this work, we leverage causal graphs to model dependencies and identify potential sources of bias. We evolve directed acyclic graphs that represent causal structures, optimizing them for fairness and accuracy using evolutionary computational methods. Our approach employs multi-objective optimization to explore trade-offs between these objectives, enabling the discovery of solutions that balance ethical considerations with performance. Experimental results demonstrate that the multi-objective framework effectively improves fairness while maintaining competitive accuracy alongside building causal graphs. This approach provides a scalable and interpretable solution for mitigating bias in machine learning models, paving the way for more responsible and transparent AI applications