Explores the multidisciplinary debate surrounding algorithmic fairness, examining how machine learning can be evaluated for moral and legal bias. The text contrasts comparative theories, which focus on statistical parity between groups, with non-comparative views that emphasize accuracy, human agency, and the explainability of automated systems. It highlights the mathematical impossibility of satisfying all fairness metrics simultaneously when social groups have different base rates of a specific behavior. Furthermore, the source investigates how biased or nonrepresentative data can perpetuate historical injustices, even when protected traits like race are excluded. Finally, it addresses the proxy problem, discussing how neutral variables can inadvertently function as substitutes for sensitive attributes.