This paper focuses on addressing algorithms that produce biased and discriminative outcomes due to the fact that the data required to train a deep neural network is extracted from the real world and, in most cases, inevitably infected with implicit bias. To this end, a novel model based on the generative adversarial network (GAN) method is proposed to mitigate biases inherent in AI-assisted decision-making, fostering trust without excessive reliance on transparency. Our proposal involves integrating two distinct deep neural networks, namely the generator, and discriminator. The generator functions as the decision-maker, while the discriminator undertakes the role of supervisor to ensure impartial decision-making. We argue that the suggested model offers a strategy for increasing fairness and safeguarding sensitive attributes while also furnishing users with explanations. Notably, the distinguishing feature of this model, in contrast to approaches centered on partial explainability in AI, lies in using a machine for supervising the debiasing procedure and offering explanations. Our supervisory machine possesses the capability to identify and rectify unfavorable correlations, thereby guiding the generator toward unbiased decisions and rendering them more trustworthy.
Rezaei et al. (Mon,) studied this question.