Machine Learning (ML) systems are increasingly utilized in high-stakes decisions. While progress exists in the fairness domain, existing evaluation techniques generally assess decisions using only a single protected attribute, overlooking the intersection of multiple sensitive attributes and resulting in compounded inequities. Intersectional discrimination is inherently difficult to perceive due to the complex nature of overlapping attributes (e.g., race, gender, socioeconomic status), rendering traditional fairness tests ineffective. This paper proposes a conceptual framework that treats fair evaluation as an optimization problem solved using a Multi-Population Genetic Algorithm (MPGA) with dynamic bounds. The framework introduces two co-evolving populations—one that maximizes bias and one that minimizes bias—to collaboratively search for and discover nuanced subgroups. Dynamic bounds are introduced to effectively constrain the evolutionarily guided search space, balancing subgroup diversity, statistical power, and discrimination sensitivity. This application of evolutionary computation provides a scalable and adaptable pathway for exploring fairness, organized around four key blocks: sensitive attribute selection, generating subgroup rules, fitness evaluation, and reporting findings. The framework serves as a foundation for future empirical work, complementing existing fairness audits, and facilitating the development of more inclusive, robust, and transparent AI systems.