Introduction
For this week’s lab, I experimented with the IIIF to retrieve a high-resolution historic photograph from the Carleton College Archives, and then used DeOldify to automatically colorize it. While the results can be visually striking, this process brings to the forefront a key ethical issue: the line between enhancing accessibility to the past and inadvertently rewriting it.
Ethical Issue
AI-driven image manipulation, such as colorizing black-and-white photos, can engage broader audiences. But it can also blur the boundary between authentic historical record and modern approximation—especially if viewers assume that newly added color accurately reflects historical truth rather than an AI’s educated guess.
Quote (from Ted Chiang)
“Think of ChatGPT as a blurry JPEG of all the text on the Web.” (Chiang)
Chiang, Ted. “ChatGPT Is a Blurry JPEG of the Web.” The New Yorker, 9 Feb. 2023,
https://www.newyorker.com/tech/annals-of-technology/chatgpt-is-a-blurry-jpeg-of-the-web
Though Chiang is referring to large language models, the “blurry JPEG” analogy resonates with AI colorization. Just as text-based AI “blurs” details when compressing and re-generating text, colorizing algorithms can produce plausible yet potentially inaccurate images, filling in uncertain areas with “best guesses.” If we mistake these guesses for fact, we risk misrepresenting history rather than uncovering it.
Quote (from Sonja Drimmer)
“How absurd to think that black-and-white photographs from 100 years ago would produce colors in the same way that digital photographs do now.” (Drimmer)
Drimmer, Sonja. “How AI Is Hijacking Art History.” The Conversation, 7 Dec. 2021,
https://theconversation.com/how-ai-is-hijacking-art-history-170691.
Drimmer underscores the danger of assuming any direct equivalence between old photographic processes and modern digital imaging. Our DeOldify output, while captivating, is ultimately an invention of the present, colored by the biases of machine learning models trained on modern datasets. This raises fundamental questions about authenticity: Does adding color “revive” the past, or simply reflect today’s vision of what the past should look like?
Side-by-Side Comparison
Below is a comparison of the original black-and-white photograph from the Carleton College Archives and my DeOldify colorization result. Both images are publicly accessible—one via ContentDM and the other via Omeka.
Conclusion
On one hand, this AI-driven process invites fresh engagement with archival materials, making them feel more immediate. On the other, we must label colorized images carefully to avoid muddying the historical record. As both Chiang and Drimmer caution, the technology’s “blurry” or “absurd” side emerges when we forget that these AI outputs are approximations—invented under modern assumptions, not resurrected from the past. By remaining transparent about these manipulations, we can continue to harness AI’s benefits while preserving the integrity of our historical sources.
