There are two fundamentally distinct research questions at play:
1. Can a computer learn X at all (and how)?
2. Can a computer learn X in an unbiased way (and how)?
Where (1) is clearly a prerequisite for (2). Most of the ML community is currently focusing on (1), and Yann is right in saying that unbiasedness should not be a concern for those focusing on it. And Timnit is right in insisting that more focus should be spent on (2), which is a perfectly valid research question in its own right and not something to be left only to ML engineers.
With that said, not being familiar with Timnit's work and seeing her explode like that did not leave a good impression of her and her cause.
1. Can a computer learn X at all (and how)?
2. Can a computer learn X in an unbiased way (and how)?
Where (1) is clearly a prerequisite for (2). Most of the ML community is currently focusing on (1), and Yann is right in saying that unbiasedness should not be a concern for those focusing on it. And Timnit is right in insisting that more focus should be spent on (2), which is a perfectly valid research question in its own right and not something to be left only to ML engineers.
With that said, not being familiar with Timnit's work and seeing her explode like that did not leave a good impression of her and her cause.
Just my 2c.