I am not a lawyer, but I worked on a project at Facebook years ago that touches on the underlying case law here. I spent a full year of my life bashing my head

I am not a lawyer, but I worked on a project at Facebook years ago that touches on the underlying case law here. I spent a full year of my life bashing my head against opaque legal precedents to understand how to build a product that met legal requirements and user needs. Maybe this comment will prove useful in distilling some of the nuance here so you don’t have to spend 12 months of your life doing what I did. This decision confirms a nearly-unanimous legal consensus that existed before this ruling about the legality of disparate treatment vs disparate impact. Disparate treatment is enshrined as illegal in other legislation like the fair credit and lending act. It basically states that the use of protected class data for most use cases is illegal. I know the most about how this is interpreted in the context of private companies, but my understanding is that the embargo on the data use is much broader. Disparate impact, by contrast, is a legal theory without much case law to support it, but it asserts that what is illegal is not the use of the protected class data, but unequal outcomes. This is super thorny territory. On the surface, disparate treatment (which again, is what the court upheld here) appears to only reinforce the racist, sexist, unequal status quo. And personally, I agree, but it at least prevents overt discrimination on the basis of race, sex, and other important human characteristics. The law right now makes it illegal to outright say that you are treating someone else differently on the basis of those traits. It doesn’t prevent all the other ways to still be racist and sexist and awful, though. So you might say, let’s go with disparate outcome, then! And this is indeed what I first thought. But there are huge problems here, too. First, the only way to ensure that outcomes are equal is … to measure and report on the very traits we think are sensitive. This wouldn’t just be invasive, it would entrench the collection and measurement of this information. I can think of a lot of ways that bad actors could misuse petabytes of accurate racial demographic information in ways that would make current ML-based inferences and regression correlations laughably indirect. Then there’s the issue of whether or important things in our society should be equal at all. Take the example of sex and employment. There are professions that are heavily male or female dominated. How would you feel if a colleague of the opposite sex with no training in your discipline was offered 2x as much compensation as you to perform the same task because society insisted that your profession have exactly equal representation? I think of myself a bit of a bleeding heart liberal, but I don’t think that’s a desirable outcome. And then there are the second and third order effects to consider. Is the fact that most commercial truck drivers are men a reflection of different preferences, social norms, or something altogether different? I have no idea and I’m not about to go all Charles Murray and act like I do. My goal in writing such a long post isn’t to make an apologist defense for what seem to me a clearly socially regressive finding from the Supreme Court, just that finding a better alternative is really, really hard. So let’s not get discouraged by this setback. Let’s find a better alternative and get that written into law.