Comment Re:The problem is that the AI gets things wrong (Score 1) 465
But in this case there seem to be enough cultural differences between races to skew the results. I'm guessing it's something along the lines that white people in the learning set were more likely to lie about stealing while hungry, while the black people were more honest. Since the AI didn't have data about race, it couldn't use it while learning - it had to average things out which means (since there were probably more white people in the learning set), that mostly 'hardened criminals' responded that stealing was ok.
What these results basically show us that there ARE differences between the races, and that including race machine learning could lead to LESS racist outcomes.