For my lab, I chose to to deoldify the interior of a girl’s room.
“These studies and projects also seem to promote the idea that computer scientists are more adept at historical research than art historians.”
-Sonja Drimmer
While the process was new to me, the steps were not as complicated as I expected them to be. Overall, this tool has impressed me. However, I’m deeply concerned about the ethical problems that were brought up in class. The social biases that are present in an AI’s training data negatively affect the reliability of the results the generative AI produces. These biases feed into negative stereotypes that could ultimately produce misleading outcomes, contrary to the intended purpose of the AI model. For example, color restoring a person of color’s skin tone to a significantly lighter complexion is problematic and fueled by biases in the training data. The prioritization of utilizing computer and scientific methods over human analytical ones can result in the creation of AI tools that lack historical context. The critical thinking required for connecting various key events in history or analyzing trends is something that can not be developed into an algorithm especially when society constantly changes.
“Even if it is possible to restrict large language models from engaging in fabrication, should we use them to generate Web content? This would make sense only if our goal is to repackage information that’s already available on the Web.”
-Ted Chiang
In addition to training data bias, the training data used may have been obtained unethically. Many platforms use their users’ data to train their AI tools without the consent or knowledge of the owner. Copyright issues arise especially within artistic communities. Even though AI has improved the efficiency of various tasks, the jobs that generative AI have been taking over require human creativity that a computer can not replicate without copying previous work.
4 thoughts on “Deoldify Lab 4”
Leave a Reply
This site uses Akismet to reduce spam. Learn how your comment data is processed.
I really appreciate your analysis, Cynthia, and I think you articulate your thoughts really well. The points you bring up not only about the biased output of AI generations like photo colorization but also its relationship to art and copyrighted training data are really insightful. Connecting the two, I wonder if some of the biases in the training data come from inequities in digitizing art. That is, if diverse creators are already at a disadvantage in sharing and promoting their work online, could this contribute to AI models using their art less in training and thus producing images that further marginalize certain communities?
You bring up a good point about if the training sets are ethically-sourced or not. That’s been a huge conflict in the realm of generative text and image *creation* AI, but I haven’t heard about it in the context of AI photo editing like this. (I’m not 100% tapped into the conversation around photo editing AI, though!) I wonder if people are as concerned about their intellectual property rights when it comes to this application. I would guess “yes,” since it’s still their copyrighted material, but I wonder if there are people who think otherwise!
I also did “Interior of a girl’s room”, and I just thought that the resulting colors looked weird. I’m not entirely sure how accurate they are, especially with the odd green artifacts. I also agree that the ethical problems are concerning, but I didn’t even mention the copyright issues. Do you think that the potential uses outweigh the ethical issues?
Thank you, Cynthia, for your interesting point! I agree with your comment about technology replicating distortions in our society, such as racism. It reminded me of the Netflix documentary Coded Bias and the book Unmasking AI by Joy Buolamwini, which explore the relationship between AI and racism in depth. Furthermore, the issue of copyright is also very relevant. While I personally believe that the way art represents an artist’s worldview is what makes it valuable, I also question whether this value will persist in an increasingly fast-paced society.