AI text to image generated image archive seeking for gender non-conforming faces against bias
Design: Merve Akdoğan
MIT Media Lab Computer Vision Class by Roy Shilkrot
Here you'll find examples of my work combating the inherent bias in AI and other forms of generative media; the first set I've posted here shows the visual shift that occurred during a conversation I had with MidJourney as a nonbinary queer person, with the intention, mindset, and love I had while shaping the visuals with MidJourney's "remix" mode, which involved inserting some non-standard, gender-neutral, transformative, manipulative prompts. Each piece begins with a question that seeks to explore or create a world out of prejudice concerns in AI, particularly those related to gender and queerness and problematic realities about how we 'label' individuals.
Although existing datasets are not inclusive enough, the goal of my discussion is to identify ways in which we may improve this situation with our intention. We can't always put our emotions into a linear expression and compute them but sometimes there's no clear path to transition between some binaries. Some things are not meant to compute or count. If you're willing to manipulate the binaries, that should be OK, and should also be represented in this huge industry.
In the current dataset, each video is made with 5 different multiple attempts, and each frame takes days to change using different prompts and iterations. Different AI tools were used to make a total of 564 images and 7 videos.
Transitions seek gender and identity biases while finally breaking down human-centric approaches and bringing the body closer to a nonhuman perspective while evolving.