In Week 4, we used DeOldify to colorize black-and-white photos from the Carleton College Archives Digital Collections. The goal was to enhance and “restore” historical images and film footage through AI-generated colorization. I selected the image titled “Oriental decorations – Sayles-Hill Gymnasium – Junior-Senior Prom” and found the results intriguing. The AI-generated colors were predominantly neutral and brown, lacking vibrancy or distinctive character. This outcome highlights an inherent tendency of AI to default to the safest option, avoiding bold or unexpected color choices. Below is a side-by-side comparison of my original and colorized images from today’s exercise (Link to Omeka).
This phenomenon aligns with art historian Claire Bishop’s critique:
“When computer science becomes integrated in the humanities, “theoretical problems are steamrollered flat by the weight of data,” which generates deeply simplistic results.”
Sonja Drimmer, “How AI is Hijacking Art History”, The Conversation
AI-driven restoration tools, while seemingly objective, often reinforce biases and oversimplifications. In the case of colorization, these biases manifest in several ways such as cultural homogenization, AI-generated images tend to standardize visual aesthetics based on Western ideals, marginalizing non-Western artistic traditions, clothing, and symbols. This can contribute to a loss of cultural diversity in digital spaces.
Furthermore, AI colorization does not reconstruct historical reality; rather, it reinterprets and reimagines the past through contemporary technological frameworks. As Sonja Drimmer argues:
“But this effort to ‘bring events back to life’ routinely mistakes representations for reality. Adding color does not show things as they were but recreates what is already a recreation—a photograph—in our own image, now with computer science’s seal of approval.”
Sonja Drimmer, “How AI is Hijacking Art History”, The Conversation
Thus, rather than offering an authentic glimpse into the past, AI colorization reshapes history according to modern assumptions and biases. While these tools can make historical images more accessible and engaging, they also risk distorting the cultural and historical contexts they claim to restore.
4 thoughts on “Lab 4”
Leave a Reply to shearerj Cancel reply
This site uses Akismet to reduce spam. Learn how your comment data is processed.
I think you make some really interesting points, Nina, and I appreciate your argument about AI’s oversimplification and homogenization of imported data. It seems that part of the argument for using AI in this way is that it increases engagement with historical artifacts. Yet, as you rightly argue, AI colorization seeks to create the least distinctive image, and so can create an image duller than the original, as well as one which erases cultural diversity.
I somewhat agree with your point that AI colorization tends to default to safe and neutral tones. I think your point about AI reinforcing biases rather than challenging them is important, and if AI tools are trained on datasets that prioritize Western aesthetics, they risk erasing any cultural nuances in historical images.
I hadn’t thought about training set bias in the context of AI image editing, but you bring up a great point! It seems plausible that through AI’s tendency to stick to the “safe” option, it might be systematically flattening out culturally-significant design choices. I wonder if anyone has written an article on this subject before?
I really appreciated how you tied in the idea that AI’s neutral choices might not capture the full cultural and historical context of an image. Your discussion about the balance between technology and historical authenticity really got me thinking about how we can use these tools without losing the rich diversity of our past. It’s fascinating—and a bit humbling—to consider how even well-intentioned digital restoration can shape history in unexpected ways. Great work on connecting both Claire Bishop and Sonja Drimmer’s critiques to your analysis; it added a lot of depth to your post.