A hiring algorithm that rewards the safest match, real wage data that shows the same job doesn't mean the same pay, a country where the rules changed overnight, and a global scoreboard that shows just how far that spans — six investigations, the same underlying pattern: systems built on the past keep producing the past.
Hiring algorithms don't actually find the best talent — they find the safest match based on historical data. Pick a candidate and watch the exact same 25 jobs reorder around them.
Zip code quietly stands in for race and income. Wealthy tech-hub zip codes get +20; lower-income urban zip codes get −25 — before a human ever opens the resume.
Names the model reads as non-Anglo take an automatic −20, echoing the real NBER finding that identical resumes with Black-sounding names got ~50% fewer callbacks.
Target-school keywords, blank "current role" fields, and rigid degree filters reward familiarity over ability — none of it illegal, all of it discriminatory in effect.
U.S. Bureau of Labor Statistics, median weekly earnings of full-time workers by detailed occupation and sex, 2025 annual averages (Table 39). Bars are scaled to the highest-paid role below.
In 2014, Amazon built an experimental AI tool to grade resumes on a one-to-five star scale, trained on ten years of resumes from a heavily male-dominated tech industry.
The model concluded men were simply preferable — downgrading resumes with the word "women's" or all-women's colleges, rewarding masculine-coded action verbs instead.
Amazon scrapped the project entirely. Engineers couldn't reliably strip the bias back out of a model trained on biased history.
A field with a "narrow" gender gap on the surface looks different once race is layered in.
Pay gaps tend to be largest in fields where women are the minority — and the Amazon case shows how that becomes a feedback loop: a model trained on male-dominated history concludes men are "inherently preferable," then perpetuates the inequality it learned from.
High-status, specialized professions are not immune. Architects earn only 64.3% of what men earn; paralegals have the widest percentage gap of any occupation measured.
Software developers look narrow on the surface (94.5%) — but Black women in computer science specifically earn only ~90–92% of what white men earn. Broad categories mask the disparities faced by women of color.
In educational/career counseling, women earn 120.3% of men; in transportation/distribution management, 116.7%. These stand out precisely because they're rare — the national average is 82.1%.
Algorithms encode the past quietly. Laws can erase decades of progress instantly. Here's what happened to women's rights in Afghanistan before and after August 2021.
World Economic Forum, Global Gender Gap Report 2025. Each score runs 0 to 1, where 1.0 is full equality between men and women — economic participation, education, health, and political empowerment, combined into one overall rank.
A country's overall score can hide the texture of daily life. The punchiest facts behind the number, by category.
A fictional podcast conversation between two thirteen-year-old girls — one in Colorado, one in Kabul — imagining what the data above actually sounds like as a conversation.