Read on Twitter
I have some thoughts on Y*nn Lec*nn’s disappointing response on dataset bias (1/12)
I’m writing a review piece right now where I argue (with Marion Fourcade) that the “coding elite” are a quasi-occupational class that is increasingly powerful and also ill equipped to intervene in domains where they are nonetheless intervening...so more coming soon (2/12)
Like many elite computer scientists he seems invested in the idea that bias especially resides in humans and wherever humans do things to data (or in model building) there is especially a problem with bias (see his comment on “hand-crafted features” ...
... and pointing to deep learning systems which are a form of “unsupervised” learning). This is a baseline assumption or a hunch resting on the idea that humans == bias (3/12)
He states that there is something called a “generic” architecture, even a “generic” objective that we can presume to be bias-free. What could that possibly be? This is merely the way computer science sees its work as domain non-specific and thus apolitical
on this point see Ribes et al fantastic piece on the logic of domains - https://journals.sagepub.com/doi/abs/10.1177/0306312719849709">https://journals.sagepub.com/doi/abs/1... (4/12)
However, a supposedly generic “optimization” model used in criminal justice reframes justice in particular ways, e.g. as the management of scarce resources. Inherently such interventions impose a bias towards prediction, minimizing uncertainty, and optimizing resources (5/12)
If that’s all you’ve ever known perhaps it’s hard to see how that skews the world in weird ways when it becomes a totalizing framework that defines preferred ways of acting. A great critique of that is found in Harcourt, Against Prediction - http://www3.law.columbia.edu/bharcourt/against-prediction.html">https://www3.law.columbia.edu/bharcourt... (6/12)
Also he poses his response as a “guess” – why bother in the face of a challenge from someone like @TimnitGebru who’s actively done research on this among many other people? There’s clearly some motivated reasoning there – it would be better for his work if what he says were true.
After each of his “guesses" he says "prove otherwise." As others have argued, a positivist epistemology would rather argue that the burden is on him to “disprove” it. The baseline assumption would not be lack of bias/racism, but the reverse. (8/12)
Apologies to the computer scientists I know and respect but computer science has never been very strong on epistemology – either recognizing that it implicitly has one, that there might be any other, or interrogating its weaknesses as a way of understanding the world. (9/12)
*** Especially as it now tries to claim the social world as something it can model and intervene in this is a looming and very serious problem *** (10/12)
You can spend your time during heightened awareness of racial discrimination and social inequality doing more of what you are already good at or you could read a bit more broadly (11/12)
Clearly he’s been doing the former or he would already know how obnoxious and derailing it is to tell a person of color that they didn’t critique him politely enough or to throw around the word ‘emotional’ in a conversation with a woman. Or maybe that was on purpose. 
https://abs.twimg.com/emoji/v2/... draggable="false" alt="🤷♀️" title="Achselzuckende Frau" aria-label="Emoji: Achselzuckende Frau">(12/12)
