Normally, "fixing" ML systems to "Remove Bias" seems like a pointless exercise (e.g. hiring algorithms that somehow still pick more male candidates in a gender-blind test, when 75% of applicants are male), but here it's done correctly.
It obviously depends on the algorithm but if you train your hiring algorithm based on past hires, and your company has predominantly hired males in the past, your going to end up preselecting males. algorithms aren't bias in themselves but if you want to avoid biased results you have to be very careful on the data sets you use to train them with.
That's why you scrub gender data, and it's still biased. the fact that more men work, period, means that it will always be biased based on that fact.
unless you make a conscious effort to select your dataset to avoid that bias sure. You don't have to give the algorithm ALL of your hiring data. Balance the data you train it on to avoid the bias.