After I published yesterday’s post, I kind of went down the AI-art rabbit hole. It’s addictive like no one’s business since it does keep challenging you to come up with more and better inputs for the software to generate an image that meets your directives.
Having said that, it’s irritatingly and yet interestingly falling short of what norms constitute a particular type- be it racial or physical. There’s hardly a sliding scale for establishing a benchmark which is odd given that AI exists in a time when everything is under scrutiny, under a lens (extremely literally) and there aren’t many hardline definitions to denote a specific look, physiognomy or even emotions indicating how an individual “ought” to come across.
But enough of the blather, here’s what went down. I kept trying to generate images of what I might look like given the keywords of #chubbywoman #Indian #hasacold #usingatissue #drinkingtea #lookingoutawindow. I alternated my keywords with #wearsglasses Here is what I got and it made me smile and realize that we honestly have quite a bit of road to traverse still…
I looped in the bestie to see if she wanted to describe a situation that could be turned into an image and she asked for 3 chubby Indian women people holidaying on a beach. Lo and behold, here is what the app generated.
While some of these images can be construed as funny, especially if they deviate quite a bit from reality; they aren’t cringeworthy either.
What is however, very surprising is the lack of diversity in the scope of what the app considers a stereotype.
Another texted-generated art about an Indian woman with curls sitting on a beach in a blue sarong created an image of a woman with a sizeable afro and wearing a saree and looking Indian-ish.
I am still having a pretty good time trying to pin down a few things so I can get more specific outputs and hopefully the feedback mechanism of clicking on a heart or a broken heart for images not of my liking, will help the program learn what passes for realism and what’s caricature.