“Can a robot write a symphony? Can it turn a canvas into a beautiful masterpiece?”
-i, ROBOT (Proyas, 2004)
Creating AI selfies meant traveling out of my comfort zone to both work with controversial technology and reflect on what pieces of myself to share. This was hard because most people, myself included, have very intentional versions of themselves that they share with different audiences (Hogan, 2010) - so I decided to explore multiple by looking at the way I represented myself on various social media versus illustrating the “real me” in order to dive into both the possibilities and limitations of AI and of the selfie.
Methodology
Step 1: Create a prompt
Using the same physical description each time for consistency
I took the different information that LinkedIn (Picture 1) and Instagram (Picture 2) broadcasted as well as the personal traits and themes that I felt best embodied who I was (Picture 3) and turned them into tangible aspects that were integrated into a written prompt.
Step 2:
Plug into Adobe Firefly's AI photo generator.
Step 3:
Get a perfect picture first try. Just kidding; refine prompt and try again. And again. And again.
Step 4:
Get a decent picture on attempt 20-something and work with it as re-prompting could be terrible for the environment (Ren and Wierman, 2024).
Picture 1: Inspired by my LinkedIn
Picture 2: Inspired by my META profiles
Picture 3: The version of me you won’t find on social media.
Here, I explored more personal themes which clearly contrasted with my social media renditions, illustrating just how little I was previously sharing with various audiences online.
Round 2
My goal was to further explore how AI could communicate complicated ideas. Adobe Firefly was not cutting it because it struggled to simultaneously include many visual components.
Step 1: Think deeper on how to better represent complex themes.
Journal. Meditate. Cry into the beyond for help because actually representing myself requires digging deep .
Step 2: Take all my traits and behaviors that genuinely feel important to my lived experience and turn them into a prompt.
Step 3: Pick a different AI
This time I used Google Gemini.
Step 4: Refine and create on repeat until satisfied. (Sorry, environment)
Picture 4: The Real Me (Attempt #2)
Does yoga
While this felt like an accurate picture, the process itself brought on many considerations about AI and selfies.
Successes
- AI was able to quickly generate images. Media can broadcast information and interact with norms (Liu et al, 2018) which AI can amplify by reducing barriers like time and cost.
- Free recources also make media creation accessible, which beneficially increases the accessibility of public communications and contributes to a shift in communications culture (Manovich, 2001).
Limitations
- Quality photo generation involves trial and error. As Wu et al. (2021) argue, AI works best when accompanied by human guidance in prompting and re-prompting. While this is a time-consuming and frustrating limitation of AI, it also emphasizes the value of human intelligence in AI generation (that’s a win for us).
- AI models vary in quality. I had to switch halfway, because representing more complex themes required intelligent integration of multiple items. This suggests that when we use AI, we have to remember its limits and try multiple providers.
- Algorithm Bias. Every photo I generated had an attractive person in it - even when I asked added “I am of average attractiveness” on the prompt. The clear bias towards traditionally attractive rendering could skew users' and consumers' idea of beauty, creating negative implications for their self-worth (Kim and Kim, 2023). I also had to specifically state my skin colour, otherwise it assumed I was white. This signifies a need for more inclusive algorithm development and increased overall transparency with AI systems.
- The ownership and privacy of your content is not guaranteed when using AI (Gunter, 2023).This speaks to a larger ethical issue of increasing surveillance and privacy loss in digital spaces that underpins calls for increased digital literacy for users and better regulation around content ownership.
This was a chance to explore Sue Thomas’ concept of transliteracy in practice, which examines how messages are impacted as they are translated across different mediums (Thomas, 2011). While this transition required reduced information as I simplified a detailed academic analysis into something more concise, it also involved the integration of additional media (like gifs), humour, and informality that could connect me to my intended audience. Ultimately, AI selfies and blog posts were both exciting mediums that left room for lots of personalization and self-exploration in order to make interesting content that would suit the format and intention.
I answer yes, it can - but humans will be at the centre of it.
References
fzynodin. (2023). You’re weclome [Gif]. In Tenor.com. https://tenor.com/view/youre-welcome-gif-25058970
Gunter. (2023, August 21). AI, the new frontier – opportunities and challenges - Research Information. Research Information. https://www.researchinformation.info/analysis-opinion/ai-new-frontier-opportunities-and-challenges/
Hogan, B. (2010). The Presentation of Self in the Age of Social Media: Distinguishing Performances and Exhibitions Online. Bulletin of Science, Technology & Society, 30(6), 377–386. https://doi.org/10.1177/0270467610385893
Kim, D., & Kim, S. (2023). Social Media Affordances of Ephemerality and Permanence: Social Comparison, Self-Esteem, and Body Image Concerns. Social Sciences, 12(2), 87. https://doi.org/10.3390/socsci12020087
Liu, F., Ford, D., Parnin, C., & Dabbish, L. (2017). Selfies as Social Movements. Proceedings of the ACM on Human-Computer Interaction, 1(CSCW), 1–21. https://doi.org/10.1145/3134707
Manovich, L. (2001). What is new media? In The Language of New Media. MIT Press. https://dss-edit.com/plu/Manovich-Lev_The_Language_of_the_New_Media.pdf
Proyas, A. (Director). (2004). i, ROBOT [Film]. 20th Century Fox.
Ren, S., & Wierman, A. (2024). The Uneven Distribution of AI’s Environmental Impacts. Harvard Business Review. https://hbr.org/2024/07/the-uneven-distribution-of-ais-environmental-impacts
Thomas, Sue. (2011, July 7). Transliteracy lecture by Sue Thomas. YouTube. https://www.youtube.com/watch?v=r02W5aTmJgI
Wu, J., Ouyang, L., Ziegler, D. M., Nisan Stiennon, Lowe, R., Leike, J., & Christiano, P. F. (2021). Recursively Summarizing Books with Human Feedback. ArXiv (Cornell University). https://doi.org/10.48550/arxiv.2109.10862
Comments
Post a Comment