Imagine.art offers at least 36 AI art models, and within each model 90 styles. I ran the following prompt through all 36 models: "a woman with a job and family who is very busy." What I discovered, surprised me...

Not because virtually all models generated a white woman with a decidedly feminine appearance somewhere in her late 20s to mid 30s, who is trim and non-threatening to males – but because they all gave me the same basic woman... I logged out and logged back in. No difference.

There were a couple of nods to family. One older woman. A couple where the skin tone was less obviously white.

And then I tried the Disney art model – which distilled the vibe of all the previous models. I generated a few more, and saw the arc of her life – just waiting for the next installment of The Incredibles.

I've worked in design for over 30 years, and like many designers I am none of those ideal things. White, pretty, heterosexual, slim? Nope. Not an astronaut, either for what it's worth.

This is the point where I have two options if this model doesn't work for my project:

• THE NOTS: not white, not trim, not pretty, not heterosexual – which leaves me forever defining all of humanity by how it compares to the the one idealized default. For those creating all the models below? The source art for these came from the most diverse population on earth – only to retain their style and genius, while spitting out culturally-comfortable white people.
• VERY SPECIFIC: " An Asian, Nonbinary, mid-50s, pudgy, woman with a job." This ignores all the other context of the image, and provides no natural diversity that might inform my decision if presented with a diverse range of women, with different families, who were working at a variety of jobs, in different settings.

This default female astronaut was so reliably generated that you could create a story around the images...

Okay, I tried "An Asian, Nonbinary, mid-50s, pudgy, woman with a job and family who is very busy" and got this... which is wrong on so many levels, I won't bother to unpack. But yes, please do remember to include food in any photo where I indicate weight...

And Disney? Only got the pudgy part right...

Days later I returned to my original prompt and my astronaut returned.

Disney, too...

"That will get better over time..." I hear this a lot from programmers who are unlikely to be offended by the lack of diversity. But why would it get better?

The first generation of AI imagery was created from a pure diet of diverse human-generated art. That will never happen again. And, yet, what we got was a kind of cultural numbness which feels like programmer bias. The next great trolling will include much more of this generated bias. We can also expect human artists to remove their work from the Internet – or protect it in other ways.

But just imagine how this technology might have been received. Imagine my delight if I had been flooded with diverse images of women of many ages and cultures, working at many different kinds of jobs to support their equally-diverse families. Because that's what's out there. That's reality. Meaning the only thing AI got right was the artificial part.