The second focus is on algorithmic fairness — that is, the ability to detect, measure, and correct biases in AI systems. The group is working on algorithm audit methodologies tailored to Moroccan institutions: public administrations, banks, insurance companies, judicial systems.
Algorithmic bias can slip in at any stage of the AI system lifecycle: in the training data (if it does not fairly represent all populations), in the definition of objectives (if the optimization goal encodes existing inequalities), or in the interpretation of results (if human decision-makers selectively interpret the model outputs).
The group develops auditing tools accessible to non-specialists—checklists, simple statistical tests, evaluation protocols—so that ethical vigilance is not limited to machine learning experts alone.