Summary: The Bias in the Machine

Denise Warren is fifty-nine and has applied for 247 software engineering positions in six months. Thirty years of experience, updated skills, strong references. Three callbacks. The rest disappeared into AI-powered applicant tracking systems that scored her out before any human saw her resume. Her graduation year, her years of experience, even her email domain triggered flags in algorithms trained on historical hiring data that skews young.

Machine learning systems learn patterns from training data. If that data reflects decades of age discrimination, the algorithm learns to discriminate at scale, speed, and invisibility no human recruiter could match. Proxies for age work even when age is not asked: graduation year correlates almost perfectly with age, years of experience exceeds what younger candidates could have, technology platforms listed signal career stage. A 2025 Stanford study found systematic bias against older women across multiple AI hiring platforms.

The applications extend beyond hiring. In job advertising, AI determines who sees which opportunities; Facebook settled a lawsuit over allegations that its platform allowed employers to exclude older users from seeing job postings. In healthcare, risk models may weight age in ways neither medically justified nor transparent. In insurance and credit, proxy variables correlating with age shape lending decisions and premiums. In fraud detection, algorithms flag transactions by older adults as suspicious based on expectations of how they should behave.

The regulatory gap is vast. The EEOC issued guidance in 2023 that employers are liable for discriminatory outcomes from AI tools, but enforcement has been minimal. New York City requires bias audits of automated hiring tools. Colorado’s AI Act takes effect in 2026. Most jurisdictions require nothing. The class action Mobley v. Workday, proceeding through federal court, may establish precedents for holding AI vendors accountable.

The algorithm does not hate Denise for being fifty-nine. It learned from a world that does, and now it scales that learning to millions of decisions per day. The same technology that amplifies bias could be designed to detect and correct it. The question is whether the will exists to make that choice.