Data science offers concrete tools to reduce algorithmic discrimination.
We can reweight a training set so underrepresented groups carry proper signal. We can transform features to break their correlation with sensitive attributes before a model learns. We can add fairness constraints during training, and audit predictions afterward against tests like demographic parity and equalized odds. The fix is not to switch AI off, but to test it by checking whether its score holds up across every group. The encouraging part is that the bias is measurable, and what is measurable can be fixed.
— If Your Name Isn’t Western, AI Could Cost You The Promotion. · Forbes