This is mostly a short rant.
The US has been busy playing musical chairs, and it looks like some things concerning diversity, equity, and inclusion (DEI) were left standing. You must be familiar with the general drama, so I’ll spare you the recap; but also, I don’t think this post has much to do with politics anyway.
What got my attention were some bizarre — hopefully niche — takes in my broader research and academia circles. Without pointing fingers, people have been saying that we should finally stop wasting resources on fair (lEfTiSt) machine learning research. To me, it’s a Simone Biles level of mental gymnastics, so let’s unpack.
What is fairness
Not gonna lie, the definition of fairness on Wikipedia doesn’t make my life easier. While it talks about algorithmic bias, and sensitive attributes, it largely focuses on people. I think it’s fair (no pun intended) to say that a substantial amount of research on fairness has been focused on social injustice. For example on the basis of gender, or ethnicity.
What it boils down to, is that we want to address some bias (and/or imbalance) in the dataset that leads to disproportionately bad outcomes for some part of the population. Population as in your data, not necessarily people. So, to me, equating fairness with DEI is misguided. A more interesting way of looking at it as applied fairness, i.e. a specific domain.
Fairness is better cancer detection in young people who are under-represented as cancer patients. It’s object detectors in cars that continue to work although you moved across Europe and everyday things don’t look the same. It’s energy demand prediction that accounts for rare, extreme weather. It’s crop monitoring systems that work with non-industrial varieties of fruits and vegetables.
”It isn’t fairness, it’s just good models?” Tomayto, tomahto. This type of goodness is what we call fairness ¯\_(ツ)_/¯
Focus on the big picture
I know that I’m preaching to the choir. Still, if you’ve been of the opinion that the main point of fairness in machine learning is DEI, then perhaps this post gives you another perspective. Even if it doesn’t, well, get a grip.