Everyone has heard of this recidivism prediction software used by American judges that penalizes African American populations and, more recently, the Apple Pay Card algorithm giving men a higher credit limit than women, despite equivalent incomes. These are examples of unintentional racist and sexist discrimination that have attracted mistrust and discredit on technological solutions designed to accelerate processes and, paradoxically, optimize decision-making by reducing the part of the subjectivity of any human arbitration. These algorithmic systems that we think of as "objective" actually have three points of weakness: On the one hand, the algorithms themselves: Most application developers do not use learning algorithms that they have personally created to measure. In open access, these generic algorithms have for the most part been developed by scientists whose priority is to validate the precision of their mathematical model and to avoid over-learning, and not to ensure the generalization in all fairness. Thus, not only were none of these algorithms designed with an explicit objective of non-discrimination, but they were developed by a singularly homogeneous population.