The math of who's missing from AI research

Have you ever used an AI tool and felt, just slightly, like it wasn't built with you in mind? Like it was talking past you instead of to you? That feeling isn't in your head. A lot of the time, it traces back to who was in the room.

Let's do a little math together, because the numbers are quieter than the headlines and they matter more.

The numbers nobody puts in the ads

Stanford's Institute for Human-Centered AI tracks who actually builds this stuff. Over the past decade, women have made up fewer than 19% of all AI and computer science PhD graduates in North America. And among the people who go on to teach the next generation? Women are just 16% of tenure-track CS faculty.

Sit with that for a second. Fewer than one in five of the people earning the highest research credentials in AI are women. And the pipeline that trains everyone coming after them is even thinner.

This isn't a story about one bad chatbot. It's about a field that has stayed overwhelmingly male at exactly the levels where the big decisions get made. What questions are worth asking. Whose experience counts as the default. What "normal" looks like in the data.

Why this lands in your lap

Here's the part that's personal. The tools you use every day were mostly designed by people whose lives don't look like yours. When a research team is missing a perspective, that gap doesn't announce itself. It just quietly becomes the baseline. The AI learns the world as the room saw it.

That's how you end up assumed to be the nurse and not the doctor. How your health questions get answered with a man's body as the template. How a tool feels almost-right but not quite hers. None of it requires anyone to be cruel. It just requires the right women to not be at the table.

Linguist and AI researcher Hadas Kotek has spent years documenting how language models follow gender stereotypes, replying that the nurse was late when you say "she" and the doctor was late when you say "he." Her point isn't that the machine is mean. It's that it's trained on a world that already tilts, and it hands that tilt back to us in confident, polished sentences.

One thing to try today

Next time an AI tool gives you an answer that assumes who you are, push back out loud. Ask it straight: "Why did you assume that?" Watch how it responds. Noticing the tilt is the first quiet act of refusing to accept it as normal. And if you've ever pictured a future in tech or research, let this be your nudge. The field doesn't need you to be the one who's missing. 💛

Quote to sit with

"The potential for harm is simply too large to ignore." — Hadas Kotek

💌 Sarah