I took to DALL-E and experimented with the various labels I might use to describe myself. My first mission was to create a selfie that looked like me, followed by other factors that play into my identity. The features that I think are most prominent in my external appearance are my curly hair, my wire-framed glasses, and a dimple when I smile.
I prompted DALL-E first with “Middle Eastern young woman,” including the features I listed. This presented me with a very feminine figure, doe eyes, full chest, and a shy smile. I found it particularly interesting that the model chose to represent the figure in this way, as my beginning prompt was mostly vague and did not include descriptors of a build or what emotions to show.
I learned that my initial prompt was not descriptive enough, and it also did not include other aspects of my identity that could make this generated selfie present in a familiar way. I used a few different prompts including “young, Libyan, Muslim women who are Media Studies and enjoy coffee and music” and sometimes using “North African” instead of Libyan and “University of Alberta” student instead of Media Studies student.
(DALL-E 3, OpenAI, 6 Mar. 2026, labs.openai.com.)
I learned it was important to create a new chat if I wanted a fresh generation. When I wanted to adjust a detail in the selfie, DALL-E was not often cooperative. For example, when I wanted to steer away from the hyperfeminine visual, I asked the model to add more androgynous features. The solution was adding a sharper jaw, but the rest of the figure was the same. This showed me that generative AI models have rigid understandings of what a certain category means, so it’s difficult to shed once that understanding becomes the foundation of what it’s creating.
This idea also taught me that including certain props or specific locations in the prompt is equally important as they add more accurate identifiers. Research done by Michael Dezuanni and his colleagues highlights that a selfie is a way for us to communicate with the world and create our own impact. In my experience, the details I wished to include are ways I wanted to communicate with my audience: that I value learning and academia, I enjoy music and that it's a big part of my daily life, and that I like to connect with my community through a third space like coffee shops.
The understanding of ethnicity and gender by generative AI models is larger than me to challenge, but it was important to acknowledge. Research by Jane Prophet and Dan M. Kotliar gives us insight into how large language models have established categories that can be used to describe people and these definitions are often simplified down and come from a gender-biased and Western perspective. When we generate a digital representation, it means we are building our identity based on definitions that are decided for us instead of by us. This can mean certain aspects of identities will not be recognized in new media — only certain representation will be acceptable while the diversity seen in real-life will be further marginalized.
In order for me to translate my critical analysis, I imagined I was creating a script for a podcast (with the added mindfulness of grammar and punctuation). It’s substituting academic jargon for conversational phrases so it reads like a coffee conversation rather than an academic lecture. The goal is for you, the audience, to be just as informed about the key points through a brief and engaging article. By adapting my analysis for the blog, it gave space for my voice to shine through as blogging’s success comes from the intimacy and community we’re granted. McLuhan’s medium is the message that tells us that the format of the blog is what makes the content impactful, by the tone and language we use and including visuals and personal experiences to make a more engaging read. It also becomes a digital representation that adds to my selfie in hopes to connect with you readers.
Comments
Post a Comment