Using AI to create a selfie was quite an eye-opening experience. I utilized prompts that not only described my physical attributes, but also my personality and characteristics. Some prompts included features such as long brown hair and brown eyes. Although the image created is not extremely accurate, as our facial features have many differences (face shape, skin tone, etc.), I do find it quite impressive that the image created is relatively similar to my actual appearance, and looks like someone I am related to! The girl in the image looks as though she is a university student and a relatively fit person, wearing athletic shoes. This represents the prompts that I included (appreciation of fitness and a love of running). The image created does not reflect all of the prompts included, such as my strong sense of faith (there is no representation of Christ in the image), or my love of music (an example of this could be headphones). Some of the other prompts that I found were not represented in the image were: my occupation, my appreciation for my family, and my desire to work in advertising using my degree in media studies. For these reasons, I don’t believe the image created is fully-aligned with my self-perception, as it lacks depth.
Art-breeder AI was the tool I used to create this image. I also tried the same prompts with other AI platforms, such as Adobe Firefly and Google Gemini. However, I found that the images created on both of these platforms were far too general and inaccurate. One aspect of this assignment that I found notable is that in the image created, the girl appears to have three feet! This is an excellent example of how it is very important to fact-check when using AI for work and academics. Overall, I found this assignment extremely relevant to the concepts we are learning in MST 499, particularly the evolution and utilization of AI in media, academics, and the workforce. As AI continues to evolve, I believe it is very important to be aware of how to operate various AI tools, as they can be extremely useful. Additionally, I find it important to be aware of the pitfalls of AI, such as the fact that it is not always accurate, so it is important to use AI as a tool, and not an instrument to fully rely on. This concept relates to one of the readings assigned in Module 3; Recursively Summarizing Books with Human Feedback, in which authors Jeff Wu et. al. evaluate the strengths and weaknesses of using AI to summarize fiction novels. Wu et. al. conclude that although AI is an extremely useful tool for data summarization, it is important that the information created using AI is fact-checked by humans to ensure that the information is both precise and accurate. AI is a tool that lacks the emotions humans experience in order to create deep, meaningful art. This notion became relevant to me as I completed this assignment. This is not to say AI cannot create art; however, it is important to note when using AI as a platform for creativity.
Wu, J., Ouyang, L., Ziegler, D. M., Stiennon, N., Lowe, R., Leike, J., & Christiano, P. (2021). Recursively summarising books with human feedback. arXiv preprint arXiv:2109.10862.
Lily, thank you for this thoughtful reflection. Your decision to include prompts representing not just physical appearance but personality, values, and aspirations shows sophisticated understanding that identity is multidimensional. Your observation that the image "lacks depth" and looks like "someone I am related to" suggests something important about the limits of algorithmic representation.
ReplyDeleteI'm particularly interested in the pattern you've identified: the AI represented fitness and student identity but ignored faith, music, family, and career aspirations. I think this partly reveals what AI systems can "see" versus what remains invisible to them. Why are some dimensions of identity more algorithmically legible than others? This is rich territory for exploration in Assignment 3.
Your observation about the three feet is excellent and deserves deeper analysis. This isn't just about fact-checking; I think it reveals how AI generates images through pattern recombination without actually understanding what it's creating. How does this relate to questions about whether AI can truly represent human identity?
Excited to see you develop these insights further!