[ericjang]

This is a sensitive topic, so I hope my question comes off as genuinely curious and not invalidating to the OPs experience, but rather a general question about how applicant selection ought to work. I wonder if the OP has some qualities that are weakly or strongly correlated with age (e.g. number of "older silicon valley institutions" one has worked at), and YC was selecting on those factors rather than explicitly filtering based on age.

My question is: if there exists some correlation between a protected class and some other variable X, is it moral to select based on X?

What happens if X is highly predictive of some conventional success metric (e.g. ability to raise a series A)?

Things that come to mind: redlining, use of zip code in credit score prediction, etc. If we deem these immoral, is the implication that we should perform selection within protected classes, and within non-protected classes (e.g. p(X|is protected class)), and guarantee equality of outcome (or non-discrimination of protected classes based on X)?

Note that I'm not saying YC was doing this - might just be standard unconscious bias.

[caddemon]

IMO it isn't inherently immoral to select based on some correlate if it's also correlated with success, but I think it depends a lot on the details. How important is the opportunity, what is the typical acceptance rate, are there analogous opportunities that are selected for using different metrics, etc. I think YC is okay on this front - kind of like access to Harvard instead of access to higher education in general.

That's just my opinion, and it is a tough, subjective question. However I think where things often go wrong in practice is that the metrics aren't actually correlated with success, or are very weakly so! I am not familiar with the YC process, but my experience with selective applications in general is that institutions are slow to update their methods, and often don't really back up their decisions with data.

Of course it is an extremely hard problem to predict start up success in the first place, that's pretty much the whole YC business model. So they should have a lot more incentive to determine the metrics that hold some real signal than a university or even a FAANG does. Which makes me think there are legitimate reasons they ask these questions, but I do hope they've considered the tradeoff in how much signal is obtained versus how much bias may be introduced.

As a bit of an aside, this is why I like the idea of pushing for more diversity between institutions instead of just within. If different places were more meaningfully different we should see more diversity in metrics used and hopefully this would result in a fairer landscape across the population without needing to enforce overbearing rules on individual institutions. Of course that makes the assumption that a particular set of metrics won't end up naturally dominating, but I'm fairly confident in that. Different people thrive in different environments, and honestly the measurements we currently have are full of uncertainty anyway.

[aabhay]

If you are a gatekeeper for an industry where protected classes are inversely correlated with success, then you should strongly consider the negative effect your gatekeeping has on the world.

[caddemon]

Having a medical degree (for example) is inversely correlated with protected classes, so I'm not sure it is entirely up to industry gatekeepers to fix this. Having a college degree also is, so without a major overhaul of the way medical school works, there is only so much that can be done at that point either.

I agree it is something people need to think about more, and in the case of software engineering it is more feasible to directly address, as the degree is much less critical. I'm all for some companies introducing training programs to hire promising potential SWEs without much prior background. But just dropping all criteria that correlate with a protected class is not a feasible solution, and it is not always the case that people making these decisions are having a net negative impact.