Earlier quoted context omitted.
>Some things are going to be hard to explain. Some skin tones, hair styles, accents, pitches, timbres, writing styles, etc. may have more/less separable entropy, and that’s going to inevitably make models seem racist/sexist for some applications. We all know that if you have garbage as an input the output can be anything. So a professional data scientist should do his job and analyze the data first , identify the pro…
If all you are wanting to know is "will this work" it is way to onerous to focus on getting a perfect dataset. If it gets into a product though, that is really pretty bad, but it may be a consequence of their development lifecycle or any other org issue rather than a failure of this or that data scientist.
If the definition of work is relaxed and equivalent to "is better then random and our marketing can sell it"
If you are a professional and you data implies that race predicts school grades and you say "it works" instead of "something is not right, let me check the financial status or house size/family size" then you are an impostor, you plug libraries with code from Stack Overflow until it works, doing the right job would give you better results and fix the bias. If you work for Google,Facebook or other giants I am sure there is not an excuse that we don't have the money to hire competent people and cleanup the data or gather correct data.
Think about it, because some person/company is incompetent you make life harder for millions of people, why we try to find excuses for not doing the correct thing, we need to demand that this AIs with big consequences are created using mathematically correct methods and when someone reports potential problems we need to stop excusing the guys that want to silence them because bad PR.