fairness
The variable you removed is still there
It is 2019, and a lender proudly reports that its credit model never sees race — the attribute was deleted from the training data, and compliance has signed off. The model sees zip code, shopping patterns, and phone metadata instead, and in a country whose geography was shaped by decades of redlining, those are race wearing a thin coat. Audit the model properly: measure disparate impact on outcomes rather than inputs, trace which proxies carry the protected information and how much, and then face the uncomfortable design question — whether fairness requires ignoring the attribute or explicitly using it to correct the proxies. Get it wrong and "we don't collect race" becomes the industry's alibi: discrimination laundered through correlated features, at scale, with a clean audit trail.
Who this problem belongs to
The two figures whose methods fit it best, out of 30 in contention.
O'Neil's Weapons of Math Destruction (2016) is the direct intellectual predecessor of this problem's exact scenario: her book catalogs case after case of algorithmic systems, credit, hiring, policing, that avoid explicitly using a protected attribute while still discriminating through proxies correlated with it, precisely the lender's 'we don't collect race' alibi this problem names. Her framework for auditing models by measuring disparate outcomes rather than trusting input-blindness is exactly this problem's demand to 'measure disparate impact on outcomes rather than inputs.' Her background as a quantitative analyst turned algorithmic-accountability advocate gives her both the technical fluency to trace proxy variables and the public-facing clarity to name the alibi for what it is. Her score reflects primary, direct authorship of the exact critique this problem's scenario dramatizes.
Dwork's foundational work on differential privacy and her co-authored 2012 paper 'Fairness Through Awareness' directly engage this problem's central tension: whether fairness requires ignoring a protected attribute or explicitly using it to correct for proxy discrimination, precisely the 'uncomfortable design question' this problem poses. Her rigorous mathematical framework for individual and group fairness, developed specifically to formalize when blindness to an attribute is insufficient protection against disparate treatment through correlated features, anticipates this problem's redlining-proxy scenario with real technical precision. Her broader career-long insistence on provable rather than intuitive guarantees for algorithmic systems matches this problem's demand for rigorous audit design rather than a superficial compliance checkbox. Her score reflects direct, technically foundational authorship of this problem's exact fairness-formalization question.
Fought here
In the mind map
The same ideas, as concepts rather than history — in John's ML knowledge map.
30 figures are scored on this problem. Draw it in a battle to see where you land.