The two key components for designing fair AI systems are consistency in how a score or assessment is given across all races and optimized satisfaction for the user versus a long-term improvement.