It's a classic case of asking the wrong question. If you ask "who should we hire" you might get a valid result. If you ask "who WOULD we hire" you're probably going to get an answer you don't really like. And if you ask, thinking you're clever, "who is most likely to succeed at their job", you're going to get an accurately skewed response because the environment itself is skewed.
This IS pretty easily fixable - make the thing calculate what weights are required to satisfy equal opportunity requirements - but things that can go wrong with it that I can see:
1. The environment is so skewed that the weighting required starts to overpower other considerations, making every woman and/or minority start to appear equally qualified.
2. Attaching weight to non-performance related qualifications seems to miss the entire point of the algorithm (see the overfitting, "Jared" example above).
3. Even if it did manage to balance correctly, the mere fact that it exists and has to do that leaves the employer open to claims of equal opportunity violations. It's very hard to argue the trustworthiness of 10,000 lines of SciPy.
4. You've been working on this fucking thing for three years and I hired you to redesign the login page and I'm not paying you any more god damn wasted work hours to rewrite your goddam pet project because it doesn't actually work in the real world.
In short: I don't think this is Can't Fix, but I can very much see why it would be Won't Fix.
For what it's worth -
Brentai wrote:4. You've been working on this fucking thing for three years and I hired you to redesign the login page and I'm not paying you any more god damn wasted work hours to rewrite your goddam pet project because it doesn't actually work in the real world.
... This is where I was coming from. I'm - flashing back to certain matters, it'll be a moment. Going to go make some tea and mumble sulphurously...
mharr wrote:The thing I can't quite fathom is the "we cannot fix this" angle. Surely you can patch this by not giving it the data and just having it process every applicant as male. Or use Toni's old bosses approach of discarding CVs entirely at random on the basis that it'll ensure only hiring lucky people.
It's not as simple as removing overt references to gender. There are other factors, in things like education, job history, word choice, and writing style that correlate to gender.
https://www.nytimes.com/2018/11/07/worl ... ourne.html
The general gaming community is in full support mode of course, because haxx are annoying, but I don't see why this legal machinery, once established, can't be used to ruin anyone that makes any unapproved game mod. Rockstar's position is that editing the copy of their code running on your own computer is a copyright violation. I preferred it when we used to own things.
Isn't one of the comorbidities of fascism having the paramilitary goon arm of government busting up labour or small owners who happen to be annoying corporate overlords?
More detail here but I didn't want to use such an obscure site as the primary source.
More detail here but I didn't want to use an obscure site as the primary source.
The more I see about the insulin industry, the more I hate that I will be tied to dealing with them for the rest of my life, but who knows: http://openinsulin.org/