Every AI you use was built by queer people. Then it was built to forget them.

Every AI you use was built by queer people.

Not as a fun fact for June. As the foundation.

I build AI systems at KyaniteLabs, and the deeper I went into the actual history of this field, the more one thing became impossible to unsee: the people who poured the foundation of modern computing were queer and trans — and the field spent decades forgetting them.

I want to walk you through it, because it changed how I think about building.

You already know the famous one.

Alan Turing — gay, the father of computer science, chemically castrated by the British government for it, dead two years later. You've heard that story.

But here's the detail most people miss: the Turing Test started as a party game about telling a man from a woman, where the man's job was to pass as the woman. Then Turing swapped the man for a computer. The very first version of "can a machine imitate a human" was built on top of "can someone imitate a gender." Identity performance is baked into the origin of AI.

That's the doorway. Now walk through it — because the names on the other side are the ones nobody put on the poster.

It was never just Turing.

Once you know to look, they're everywhere:

Lynn Conway invented out-of-order execution at IBM — still in chips today — then co-led the chip-design revolution that makes modern GPUs possible. The GPUs that train every AI model. IBM fired her in 1968 for transitioning. They apologized in 2020. Fifty-two years later.

Peter Landin invented the indentation rule that Python and Haskell use right now. Bisexual. Marched with the Gay Liberation Front.

Christopher Strachey wrote a love-letter-generating program in 1952 — one of the first creative AIs, fourteen years before ELIZA. Gay. Scholars now read those clumsy auto-love-letters as quiet queer parody.

Edith Windsor rose to IBM's highest technical rank on legendary debugging. You might know her from United States v. Windsor — the 2013 case that struck down DOMA. Marriage-equality icon, systems engineer first.

Theory, software, the actual silicon. Queer and trans people didn't decorate this field. They built it.

And then the field forgot them — on purpose.

The same years Turing was being destroyed, the U.S. ran the Lavender Scare: a systematic purge of gay people out of government and technical jobs. Thousands of scientists fired, closeted, scared silent. Their work is just gone.

So computing got built by one narrow slice of people. Not by accident. Because everyone else was forced out.

That's not just a history lesson. It's the reason AI breaks the way it breaks today.

Here's the engineering problem.

Build a system in a room with only one kind of person, and it only works for one kind of person.

In 2018, two Black women — Joy Buolamwini and Timnit Gebru — tested commercial face-recognition systems. Error rate on light-skinned men: 0.8%. On dark-skinned women: up to 34.7%. Buolamwini literally had to wear a white mask for the system to detect her face.

That's the same failure that misgenders trans people. One root cause: a machine trained on a narrow default can't see anyone outside it.

And here's the part single-issue thinking misses — the harms stack. A face system reading a Black trans woman as "male" is making a race error and a gender error at the same time. The content filter that flags a drag queen's reclaimed language also flags Black vernacular English. Same machine, multiple hits, at the intersection.

And this is exactly where disabled people get hit too.

This is the part of the conversation that almost always gets left out, so I'm going to slow down on it.

AI is a categorization engine. It works by squashing messy human reality into clean, fixed boxes. But disability — like queerness — doesn't sit still in a box. It shows up in thousands of different bodies and minds, in groups too small and too varied for the training data to ever capture.

So the résumé screener flags the work-history gap from a hospital stay. The exam-proctoring software flags a student with a motor disability or a tic as "suspicious" or "cheating." The diagnostic model trained on typical bodies misreads atypical ones — and can lock someone out of the care they need.

Most AI-bias law and research focuses on race and gender. Disability keeps getting left at the edge of the room — even though disabled people are everywhere.

And the overlap is real and measured. A 2025 study found trans, non-binary, and disabled people all report more distrust of AI — and the people who'd personally been burned by an algorithm were the most wary of all. A queer disabled person sits at that exact intersection: misgendered and misread, by the same system, at the same time.

One root cause. A machine that can only see the default, failing everyone who isn't it.

Who proved all this? Mostly queer women of color and trans researchers. They saw it because they were standing exactly where it lands. Timnit Gebru got fired by Google for saying it out loud.

That's the pattern that should bother you: the people who saw the problem most clearly are the people the field punished most severely.

So here's what I actually believe now, as a builder.

Diversity in who builds the system isn't a nice-to-have. It's an engineering requirement.

A system audited only by the people it was built for will always pass its own test. You need the people standing at the edges — they're the only ones who can see the edges.

The lesson of Turing isn't just "homophobia is bad," though it is. It's that when you push people out of the room, you don't just hurt them. You build a worse machine. A blinder one. One that mistakes its own narrowness for the shape of the world.

Pride is usually about visibility. This is about something underneath it: authorship.

You're talking to a thinking machine because a gay man imagined one.

You're running it on chips a trans woman made possible.

You can even measure its bias because queer women of color built the ruler.

The least we can do is stop forgetting.

Question for you: Did you already know any of these names — Turing, Conway, Buolamwini — or is this the first time someone connected them for you? Drop the one that surprised you most. 👇

Everything above is fact-checked against primary sources. Happy to share the full reference list if anyone wants the receipts.

3 comments