Recommendations and results from AI applications treat people in comparable groups even-handedly, with no favouritism, no discrimination and no resulting harm, limiting bias on the grounds listed in the Universal Declaration's second article and in Hong Kong's anti-discrimination ordinances. Because fairness is a social construct with many mathematical definitions, applications are designed to meet specific, chosen fairness goals. Training data is kept free of sample size disparity, selection bias, bias in model design and bias in use and feedback, and the integrity of source data is assured so that invalid or inaccurate data does not produce discriminatory decisions.
The graph holds this control, the 0 it maps to, and the evidence behind each claim, over MCP and REST.