I am curious about good solutions. But I feel like "first, do no harm" is vital here - the reason we're happy to speculate about cryonics is that the worst case is you're still dead, whereas SV is astonishingly productive to the point where we don't want to look at it funny in case it stops working.

And there are specific worries. I've worked at a place that appeared to make a point of hiring equal numbers of men and women - and implemented this by hiring a number of women who simply weren't capable of performing their jobs (whether this was a question of talent or training I don't know). It was bad for me, as in programming incompetent colleagues are not just dead weight but actively harmful. It was bad for those women, who had a frustrating time at a job where they knew they weren't contributing. And it was even worse for the highly-skilled women on the team, as others who hadn't worked closely with them would - understandably, and probably unconsciously - make the calculation that they were probably in the incompetent group, and act accordingly. So naturally, avoiding that particular failure mode is a matter that's close to my heart, and I get defensive when I hear people saying things that I think might lead to that.

Well, obviously no one has come up with full solutions, and whatever solutions there are, they will be slow, because changing culture is slow. But just as democratic societies have come to accept the preposterous notion that women should be allowed to vote, so too will it start seeing engineering as a non-manly profession, and one that's perfectly natural for women, too. Right now, I think there are two things we can do: 1) place more women role models in popular culture who are engineers (I think Hollywood has started doing this), and 2) place more actual women role models in engineering positions.

The latter will only be achieved if the level of hostility women feel in the workplace is reduced, and to do that, employers (and engineers in general) will need to undergo some training. Part of that training (though certainly not all of it) would be learning to recognize sexism. Recognizing sexism is very hard if you're not trained for it; there are two reasons for that: one, people confuse sexism with misogyny and assume that since they're not misogynist, they can't be sexist (while actually most forms of sexism are inadvertent/structural/cultural), and two, because sexism is cultural, we just don't think of things that seem so natural to us, and can't perceive the harm they cause. Because sexism and racism are discrimination + power, the best way to recognize them is not to look for discrimination (which is hard to see, and we don't want to find it because it feels we're being judged or doing something wrong) but for power. Once you know what power is, it is relatively easy to see. Once you learn to see power, you see who has more of it and who has less of it. Once you see that, it's much easier to see whether through action or inaction your organization keeps the current unfair power distribution.

One of the other results I've seen posted on SSC (will try to find the link later) is that sensitivity training, at least as it's been actually implemented in the real world, makes people more discriminatory, not less.