You are very right to bring that up. That said, I think it doesn’t get to apply in force here; if you have a quota, you’re still looking to get the best out of each required category anyway. That is, you’re not *replacing* the “I’m looking for the most competent people” metric, you’re just throwing an additional requirement on there.

Well, you could argue that you’re not changing it much (is this what you’re arguing? I’m having a hard time telling), but there is going to be some cost for not “just looking for the most competent people” (although that assumes one was doing that correctly, which as you have pointed out, one probably wasn’t).

Yes, you can’t really read the conditional probabilities that way, if we’re being strict about things and pretending that we can only assume a uni-directional link without further evidence. But I suspect that in real life we can do so. Here’s a supporting anecdote (the post is long and mostly irrelevant; search for the paragraph starting with “I’ve told this story before” and read from there).

It’s a very interesting story — I’ve seen it before, actually, but thank you for reminding me of it. It’s an interesting strategy and it seems to have paid off here, but I must admit I’m not entirely at ease with it. One would hope that there would be some way to modify it so as to get the benefits of it without having to pay the corresponding costs. But it’s possible that such a thing might not exist — it’s hard to get people to truly put in as serious an effort when nothing is actually at stake — and that something like this is the best we can do.

I’m wondering if maybe this is getting to the heart of our disagreement here. I have no problem using such positive-discrimination methods — so long as there is a solid argument for doing so, that it’s really the best way, that you can’t avoid it and get the same good results. To my mind, the fact that a method uses any sort of discrimination is a significant cost that needs to be overcome; these aren’t a style of solution you just toss around casually.

As such, to my mind, arguing for positive discrimination bears a burden of proof. So I expect people arguing for it to make solid, explicit arguments that shows they’ve seriously thought about it, have anticipated objections, and really don’t see a better way — or at least lampshades the gaps in the argument and asks us to spot them the point, to let us know they’re not an idiot.

But this is not what I see. What I see is the people making flimsy arguments and then papering over the gaps by calling evil anyone who tries to point out the holes. Sometimes these arguments are backed up by data, but, well, data is just the start of an argument, it’s not an argument in and of itself. I mean, honestly, I don’t know statistics. I don’t know how to judge whether you’ve done your statistics right. So I will spot you the data. What I want to see is solid argument structure, because that seems consistently to be lacking! (Then maybe once that’s down I’ll worry about the solidity of the data.)

And, you know, in a lot of these cases, I can pretty easily see how to take these people’s arguments and improve them and fill in many of the holes myself. But when you’ve annoyed me by making a flimsy argument and acting as if there are no real issues in it, then, well, I don’t want to make your argument for you. I want to disagree with you out of spite, in the hopes that maybe it’ll inspire you to stop making such flimsy arguments. Unfortunately, this never seems to work.

I’m hoping this maybe makes it a bit clearer where I’m coming from?

Report comment

This is a decent argument that diversity can serve as a measure, but just in case you’ve forgotten…

(I am not saying you are necessarily wrong; I am saying that if you have not accounted for Goodhart, you are making a mistake.)

You are very right to bring that up. That said, I think it doesn’t get to apply in force here; if you have a quota, you’re still looking to get the best out of each required category anyway. That is, you’re not *replacing* the “I’m looking for the most competent people” metric, you’re just throwing an additional requirement on there.

(Also it’s only really an argument that the current state of affairs is suboptimal; trying to read a target out of it seems like something to be wary of.)

Yes, you can’t really read the conditional probabilities that way, if we’re being strict about things and pretending that we can only assume a uni-directional link without further evidence. But I suspect that in real life we can do so. Here’s a supporting anecdote (the post is long and mostly irrelevant; search for the paragraph starting with “I’ve told this story before” and read from there).

Report comment

For example, radfems and lesbian separatists claim to be “feminist”, but they’re of the “men should basically be killed” crazy sect.

Now I think you’re being unfair to the radfems! At the very least they point out concerns the libfems don’t. I’m sure the libfems have answers to these concerns, but as someone who isn’t intimately familiar with these arguments, I don’t feel comfortable dismissing the radfems out of hand.

So, for one, that kind of problem definitely crops up. I’ve seen in this very blog comments talking about radfem as if it was a mainstream view, rather than a bugfuck-nuts offshoot.

Part of this might be variance in environments, but let’s not forget the other factor I’ve already mentioned — once you’ve entered the trap, it becomes difficult to distinguish between different strains of feminism, because you’re required to obey all of them. (As well as whatever superficially pro-woman nonsense can pass itself off as feminism.)

Report comment

This whole time I’ve been talking about removing the effects and consequences of implicit and societal biases on individuals; my argument for better diversity in itself is just an observational one, that when I’ve seen selection processes that purposely blind themselves, they come up with better gender/race/etc ratios than an unblinded one does, suggesting that we leave a lot of value on the table by default due to biases. This suggests that we can target that second-level effect, the increased diversity itself, as a more visible and easier-to-purposely-optimize thing, and get the underlying effect (increased competence) to come along with it.

This is a decent argument that diversity can serve as a measure, but just in case you’ve forgotten…

(I am not saying you are necessarily wrong; I am saying that if you have not accounted for Goodhart, you are making a mistake.)

(Also it’s only really an argument that the current state of affairs is suboptimal; trying to read a target out of it seems like something to be wary of.)

I’m going to assume that you’re just pretending to be a sociopath in that comment, ironically or whatever. I don’t think it deserves any further response.

This is a good idea, Ialdabaoth does this fairly often.

Report comment

Report comment

These researchers did not use “non-standard methodology” in terms of what you would see in actual mathematics, fields like machine learning and so on. If anything they applied techniques that other psychological researchers are inexplicably decades behind on.

This is a point in favor of all psychological research along these lines just being bad, including that which yields the result that men and women don’t differ in personality, because the underlying data is invalid. So more studies would have to be constructed from the ground up, but it’s not a unique criticism for what these researchers did.

Report comment

Report comment

Report comment

Report comment

Report comment