From search bars to chatbots: a short history of AI bias

Have you ever asked a chatbot something and felt, somewhere in the answer, like it wasn't really built with you in mind? That little flicker isn't you being too sensitive. It's a pattern. And it has a history.

Long before chatbots, a researcher named Safiya Noble started typing into a search bar. In her book Algorithms of Oppression, she wrote about searching the words "black girls" and getting back something degrading instead of, you know, actual girls. She showed that search engines aren't neutral mirrors. They learn from us, ads and all, and they hand our worst habits back to us dressed up as objective results. Around the same time, Latanya Sweeney at Harvard ran a quieter experiment: she found that searching names more common among Black Americans was more likely to surface ads implying an arrest record. The machine wasn't malicious. It had just absorbed the world and repeated it at scale.

Fast forward, and the bar where we type our questions changed shape. It became a chatbot that talks back. But the pattern came along for the ride. In 2024, UNESCO published a study testing popular large language models and found regressive gender stereotypes baked into their answers. Women turned up tied to words like "home" and "family"; men to "business," "career," "executive." One open model cast women in domestic roles around four times as often as men. Same lineage. New interface.

Why this is yours to know

Here's why it matters for you specifically. These tools now sit in the middle of real decisions: the resume you tweak, the email you draft, the late-night question you'd be embarrassed to ask a person. If the thing answering you quietly assumes a woman's place is smaller, that assumption can shrink your options before you even notice. You deserve to use these tools with your eyes open, not to be quietly managed by them.

Knowing the history is its own kind of power. When an answer feels off, you don't have to wonder if you're imagining it. You can name it: this is a known pattern, documented by women like Noble and Sweeney who saw it coming. That naming is the opposite of gaslighting yourself.

One thing to try today: next time a chatbot gives you advice about your work, your money, or your worth, ask it a follow-up. "Would you say the same to a man in my position?" Watch what shifts. You're not being difficult. You're auditing.

Quote to sit with:

"At the end of the day, online advertising is about discrimination... The question is when does that discrimination cross the line from targeting customers to negatively impacting an entire group of people?" — Latanya Sweeney

💌 Sarah