Bias & Representation

Highlighting issues of bias and representation in AI-generated creative works.

Devon arrived at college already having established a photography business. He took photographs of weddings, graduation ceremonies, and even his friend's band. So, when he arrived in college, he knew exactly how he wanted to further his photography career. However, in his second-year photography studio course, his professor introduced them to an AI program that could generate photographs based simply on a short text description. Devon has always been interested in how photography and tech intersect and was eager to try out this new way of expanding his creativity.

Devon signed up for a pro account on the AI tool and spent long hours generating photographs based on the themes he was working on in his course. He was initially wowed by what the program was producing. Each one was a slight variation on his theme, and he could produce variations on those variations. It sparked his creativity and helped him generate ideas for future projects.

As his project advanced, he spotted reoccurring patterns in the images that were being produced. They didn't allow him to create the kind of diverse imagery he had in mind for his project, no matter how many times he changed his prompts. Also, the photos, especially those that included people in them, look like stereotypical caricatures that didn't provide a variety of races, genders, and backgrounds that were culturally distinct from each other. Devon aimed to demonstrate how rich human experiences could be, so this limitation was frustrating, and he couldn't figure out a way around it. For that reason, Devon began researching why the AI tool produced this kind of bias.

Devon also heard others debating this issue. Some argued that AI-generated images expose the inherent prejudices in our society and the art world. They saw this as an opportunity to spark meaningful discussion about these topics. Others pointed out that these biased tools inadvertently allow artists to perpetuate harmful stereotypes. This group advocated for boycotting AI tools until they could be more representative.

Devon contacted some photographers from often-underrepresented groups to learn their perspectives on how datasets are created and their feelings about how they were being represented, not represented, and misrepresented. Some were offended, while others wanted to use AI tools to create fantastical and non-realistic images that would, in effect, challenge the conventional notions of identity and representation. These conversations made Devon realize that the issue was far more complex than he initially thought, involving questions of artistic freedom, social responsibility, and the very nature of representation in art.

He found several helpful articles that reviewed how AI-generated content produces bias and diverse representation. At the core of this problem lies the fact that AI algorithms need to learn from image datasets that contain massive amounts of photographs that often represent only certain and often privileged groups in society. Devon learned even more about this on several podcasts focusing on AI's ethical implications in art. The fact that these data sets contained many more images from specific demographics led to the images being produced being an overrepresentation of those groups.

He was determined to focus on this issue. He experimented with different ways of generating images that reflected more balanced and diverse data sets, especially those representing a white spectrum of humanity.

Devon presented his final project at the end of this semester. It was unique among his classmates. It showed AI-augmented photographs and a written essay about the importance of adding more diversity to the datasets that are part of these AI tools. He hoped his classmates and other photographers would follow ethical standards as they produced photographs with AI tools.

What do you think?


Questions for Discussion

  • In what ways can an AI tool improve a photographer's creativity?
  • Are there any risks in using an AI tool to create a photo?
  • Why do we need to call out and eliminate the biases produced by AI-generated content?
  • How does dataset content that is used to train AI algorithms impact who becomes represented in the images that are generated?
  • How can the creative community ensure that the future of AI in art is diverse and inclusive?
  • What role can photographers and other creatives play in shaping how AI tools are ethically used in their fields?
  • What new skills should a photographer acquire to avoid producing new biases when they use an AI tool?
  • What did you learn from this scenario that you can apply to your process?

Incomplete list of resources that deal with bias