Countering Bias in the Machine: The Importance of Art in AI Development
Researchers like Google’s Alexander Mordvintsev are passionate about emergent phenomena and data visualization. Mordvintsev is credited for introducing the world to DeepDream, an entirely new subgenre of neural network-based art that has transformed how we visualize images in AI. Books like Arthur I. Miller’s The Artist in the Machine: The World of AI-Powered Creativity, argue that computers are as creative as humans but this also includes humanity’s flaws (i.e., bias). Like Mordvintsev other researchers, artists and engineers have tossed their hats in the machine learning AI ring and a view of their profiles tells a far too familiar story about the world of AI development.
Fans of my AI-generated images frequently tell me that they see themselves in my work. According to the research, lot of these individuals are from groups that are recognized as underrepresented in modern AI (ex. machine or deep learning). Black and Latinx people, and women are historically underrepresented, which has led to major societal implications. For example, in 2016, ProPublica showed that software used in the U.S. to predict future criminals was biased against Black Americans. More recently, researcher Joy Buolamwini demonstrated that Amazon’s facial recognition software has a far higher error rate for dark-skinned women.
I addressed the bias issue in a recent Medium article and one question I was asked after I published the account was: How are you able to keep using that AI image generator tool? My answer is two-fold. First, I’m hyper-aware of the existence of bias in the machine (from personal experience) and I’ve been investigating the proliferation of AI image generators and the instances of algorithmic or racial bias in the output (of the tools). Other researchers have been doing the same. I also use art knowledge and visual language to intervene in the process.
Last year, Kathleen C. Fraser, Svetlana Kiritchenko, and Isar Nejadgholi wrote a award-winning paper about racial bias in text-to-image systems. I tried their experiment, as well, and got similar disturbing results. I also noted the authors’ push to develop more effective methods to promote equity, diversity, and inclusion in the output of image generation systems. One of the first steps is creating images of more diverse people doing things like programming or building machines.
I used MidJourney to create the “Rosie” portrait series. The series imagines a future for women’s labor by merging World War II’s Rosie the Riveter, an allegorical cultural icon, with Industry 4.0, which is characterized by rapid change to technology, industries, and society due to the emergence of AI. The “Rosies” acknowledge a historical Digital Divide that refers to the lack of digital skills, which impedes the handling of technology, as well as the ways in which race and class intersect with claims of digital democracy.
The digital divide reflects and amplifies existing social, economic, and cultural inequalities such as gender, age, race, income, and ability. The communities that are most affected by this divide include, according to the UN, women/girls, children/youth, and marginalized communities. Unequal access to digital resources can have a range of adverse effects. It reinforces existing societal, economic, and education disparities, creating cycles of disadvantages. It also results in AI system biases. Representation is very important as far as mitigating algorithmic biases but what does one do when the makers/engineers (producers) of AI-driven products look more like Alexander Mordvintsev than Stephanie Dinkins?
My answer to that is “Art” and that is something people from all walks of life have access to. Teaching young people, especially those from marginalized groups how to intervene in the AI generation process is one strategy. Teaching them how to collect and categorize their own data, and train their own AI models is another approach. Learning how to be creative using AI generator tools is an important starting place. This is how I got into computer graphics after majoring in high school visual art and deciding “computers” was not for me. My teacher showed me that computers and software could be used to make art. This is what brought Alexander Mordvintsev and other AI art pioneers to the table (as well as being in the right place at the right time).
In other words, we need to do more than criticize AI and AI art. We need to be in (or create) spaces where we can develop or re-imagine the tools.