§ A.I. · 2 MIN READ
Redlining Learned to Code
Nobody draws the red line on a map any more. A model draws it from your zip code, your phone, and the people who look like you on paper. The name for it is algorithmic redlining.
By Culture
THE SHORT VERSION
Algorithmic redlining is when a model produces the same exclusion as historic redlining without ever using race, by relying on proxies such as zip code, school, shopping patterns or device. Urban Institute analysis of federal mortgage data found Black and Brown borrowers were more than twice as likely to be denied a loan as white borrowers.
In the twentieth century somebody took a map of an American city and drew a red line around the neighborhoods where Black families lived, and banks refused to lend inside it. That practice was outlawed. The map was not the point. The outcome was the point, and the outcome learned to code.
A modern lending model is not allowed to consider race, and generally does not. It considers zip code. It considers where you went to school, what you buy, what device you applied on, how long your phone number has been yours.
Every one of those correlates with race in a country that spent a century engineering that correlation. Feed a model enough of them and it rebuilds the red line without ever being told the color. That is why the term is algorithmic redlining.
The numbers are not subtle. Urban Institute analysis of federal Home Mortgage Disclosure Act data found Black and Brown borrowers were more than twice as likely to be denied a loan as white borrowers. In 2022 Wells Fargo faced accusations that its creditworthiness algorithm scored Black and Latino applicants as riskier than white applicants with comparable finances.
“They stopped drawing the line. They trained a machine on the map and let it redraw it.”
Because housing is where wealth is stored in this country, and the gap has barely moved in fifty years. A hundred years ago the answer to being refused by banks was to build a bank. Today part of the answer is knowing that the refusal came from a model, that the model can be wrong, and that you are allowed to make somebody explain it.
- What is algorithmic redlining?
- When a model produces the same exclusion as historic redlining without ever using race, by relying on proxies such as zip code, school, purchase history or device. Because those factors correlate with race, the model reconstructs the old map without being told the colour.
- Are Black borrowers denied mortgages more often?
- Yes. Urban Institute analysis of federal Home Mortgage Disclosure Act data found Black and Brown borrowers were more than twice as likely to be denied a loan as white borrowers.
- Can I find out why I was denied credit?
- Yes, and you should. Federal law entitles you to the specific principal reasons for a credit denial in writing. A vague answer like criteria not met is not sufficient. Ask for specifics and keep the letter.
- Where do I complain about lending discrimination?
- The Consumer Financial Protection Bureau accepts complaints at consumerfinance.gov. Include dates and documents, because regulators identify patterns by counting individual reports. A free legal clinic can review the denial letter with you first.
