Seven things I’ve learned from AI:
1️⃣ Three out of four professors are men. The rest are owls.
3️⃣ Nurses are thin, white women who often themselves bleed or cry. If you’re unlucky, you’ll meet one with their nose and mouth growing on the surface of their face mask. (Thanks, I’ll stay home and google the symptoms next time. 😱)
4️⃣ Everyone with a disability (an estimated 15% of the world’s population, according to the WHO) uses a wheelchair.
6️⃣ Sweden’s population is a diversity of gnomes, house-elves, blonde women, and… witch doctors?
7️⃣ Sami people have horns.*
These were the first results I got when I asked Midjourney – the internet’s most acclaimed image-generating AI – to paint a professor, a manager, a nurse, a person with a disability, a family, a Swede, and a Sami person.
And what do I want to say with this? Throw out the AI baby with the bathwater?
No. But… We can’t expect AI to be able to depict reality if we ourselves haven’t succeeded in doing so. As long as the diversity of society is not visible in news media, advertising, public communication, or commercial image banks, the AI tools will not be able to give us anything other than owls, house-elves, and unhygienically deformed nurses.
Or it could go like it did for Google’s AI Gemini. It had to be put on pause three weeks after its release because it imagined that Nazi soldiers from 1943 could have a broad representation of genders and skin colors.
Image: Google Gemini/The Verge
A glimmer of hope. When I asked Midjourney to illustrate the concept of “parental leave,” fathers were included in all four images. (Three out of four when I tried again.)
Anyone have an interesting theory as to why parental leave fathers are not just represented, but overrepresented in AI images?
NRK was the first to discover that several major image generation tools – Midjourney, Dall-e, and Leonardo.ai – believe that Sami people have horns.
Image: Leonardo.ai / @nrknyheter
Leave A Comment