Reduced costs refer to the amount by which the objective function coefficient of a non-basic variable needs to improve before it can become part of the optimal solution in linear programming. It is a crucial metric for understanding the potential for optimization and resource allocation in decision-making processes.