By working with the two main inputs to machine

learning $($input data & better defined output

goals$)$ we have just seen that in this case, it was

possible to:

reduce the bias signficiantly, with a

very. minor decrease in the

algorithms accuracy

eliminate the bias, but the algorithms

accuracy will decrease decisively

create the false impression that the

bias is gone, but, in reality, one

cannot do anything about biased

machines