> How do we correct for the bias if we don't know what the unbiased state would look like?
The goal isn't to choose some specific gender balance and then take whatever steps necessary to produce it, it's to eliminate unfair bias. The way to do that is to find it, not indirectly by looking at ratios, but directly by actually finding it.
High school teachers who say things like "women are no good at computers" to young women should be reprimanded etc. Poll women who didn't go into tech and ask them why not, and if any of the reasons are unjust then change them.
If no one can find anything like that then the gender balance at that point is what it's supposed to be. We're obviously not there yet, but the way we know is because we keep finding things like that, not because the gender balance is uneven.
I agree. But I notice that most discussions on sexism in technology companies focus on the outcomes; gender ratios and pay gaps. It would be interesting to try to measure the bias directly. Hide a bunch of microphones in offices and see if the number of sexist comments is larger at Google than it is at a law firm or a hospital.
Perhaps somebody has already done this. What would be the correct search terms?
> But I notice that most discussions on sexism in technology companies focus on the outcomes; gender ratios and pay gaps.
It's a specific instance of the more general manage-by-metrics disease. The thing you actually want is hard to measure, but it correlates with something that is easy to measure. So instead of doing the hard work to understand what is actually happening in detail, they measure the easy thing and optimize for that instead. Even though doing that frequently breaks the original correlation.
The result is the bureaucracy edition of a paperclip maximizer. You get what you measure instead of what you really want. Or bang your head against the wall, if the thing you measured is actually stickier than the real problem.
> t would be interesting to try to measure the bias directly. Hide a bunch of microphones in offices and see if the number of sexist comments is larger at Google than it is at a law firm or a hospital.
To some extent this is just the same disease. Is sexism supposed to be alright if it turns out there are equally large amounts in both places? Should we be satisfied that it's the root of the problem if there is very little at Google but even less somewhere else?
Stop trying to measure things against other things and just consider them in their own right. Sexism is bad regardless of how common it is. You don't fight it because there is more of it over here than over there, you fight it everywhere because it exists when it shouldn't exist.
Frankly I do not see how 50/50, or any fixed number, can be a representation of the unbiased state. The unbiased state is unknown.
If you look at what the orchestra did and remove factors you know are irrelevant for the applicants then you should move towards a more unbiased state independent of what the number actually is. The problem is to find out what bias you have. One way is to compare the people who apply with the people you hire. You can see what traits you select on, then you can decide if you think those traits are good or bad.
You ensure a fair process (without attempting to overcompensate/correct for other processes out of your control), and whatever comes out is the unbiased state.
You can't escape the fact that (assuming any innate differences whatsoever) equality of opportunity will mean unequal outcomes, and equality of outcomes requires unequal opportunities.