Book Review - Unmasking AI: My Mission to Protect What Is Human in a World of Machines by Joy Buolamwini
- 1 day ago
- 5 min read
Joy Buolamwini (2024). Unmasking AI: My Mission to Protect What Is Human in a World of Machines
Book review by Angelica Kalika
Dr. Joy Buolamwini is an award-winning researcher, poet of code, and activist. She founded the Algorithmic Justice League and now is an Inaugural Senior Fellow of Ethics in AI at the University of Oxford. Her work at the MIT Media lab, where she received her PhD in 2022, was the focus of a Netflix documentary Coded Bias. Her research focuses on algorithmic bias after facial recognition software blocked her own face from detection. Only after dawning a white mask was she recognized. She realized this was a systemic issue in the types of data AI was sourcing from. Essentially, minoritized communities were not recognized or were misidentified because of unrepresentative data in the algorithm. Once she found and reported these results there were mixed reactions from big tech.
The overall issue - could algorithmic bias impact more than just facial recognition technology? Buolamwini originated the term Coded Gaze to describe what she was witnessing. She says it is much like the male gaze, but here the focus is not on commodification of the female body, but how this gaze reflects only its creator's view of the world and is already embedded into our data systems. AI is a reflection of these biased and prejudice data sets.
Much of the book takes us through the internal dialogue of Buolamwini, and her struggle of balancing her passion for software design, artistry, innovation, with using tech for social good. But she questions what does doing good mean, and whom does it do good for? She explores concepts outside of computer science to help her ethically understand what direction her work should take her: Intersectionality, colorism, ableism, patriarchy, power asymmetries, and a term she introduces - Excoded. Buolamwini was excoded in systems - where it doesn’t recognize or include you in the design. This exposes the excoded to direct harm, damage, and/or danger from AI frameworks.
She takes us step-by-step through her projects that expose the coded gaze in her masters and then PhD program at MIT and opens up about her inner ethical turmoil in choosing images for her Gender Shades project. A history of photographic/filming of black communities is first analyzed to showcase that misrecognition here is not due to the technical aspects of a camera - that the camera is not neutral. Buolamwini discussed what led her to choose gender and/or skin type measures in the study - and all of the steps she took to ethically choose and label images in her quantitative research methodology. Using the Fitzpatrick Skin-Type Chart, her study found that it was at the intersectionality of race and gender that these algorithms had significant bias against those with darker skin and those that presented as women - with significant rates of misidentification error. Those that had lighter skin and presented as male had a significantly lower misidentification error. Her seminal peer-reviewed paper, “Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification” was co-authored by Dr. Timnit Gebru and published in 2018.
Starting with a marshmallow and toothpicks metaphor Buolamwini takes the reader by the hand and discusses how AI networks and LLM deep machine learning is constructed. It is through layered neural pathways, built to mimic the human brain, that we uncover the dangers of AI if vastly applied to society, “Neural does not equate to neutral” (54). Buolamwini also takes the time to lay out the differences in facial recognition technology with 1. Facial Verification - finding a match to a specific face, verses 2. Facial Identification - one face matched against a large data set of faces. It’s important to understand the difference when advocating for policy - it must be broad but specific enough to know exactly what the legislation is regulating.
Inspired by tenants from a housing community in Brooklyn, NY fighting against the installation of facial recognition technology in their building, Buolamwini’s work leads her to create the Algorithmic Justice League (AJL). AJL works to fight for algorithmic justice by providing resources to activists dealing with the impacts of biased algorithms. With a lack of laws at the state and federal level, each case is a battle. The book goes into several case studies where the implementation of AI causes false arrests, job loss, and medical apartheid - all particularly within marginalized non-white communities.
As a first generation American, her lived experience is woven into the fabric of the story. From her parent’s support of her coding journey to the bullying she dealt with in school, to adulthood where her poetry echoes her past, present, and future. Even though the story is centered on unmasking AI and the potential foreboding future that could await us unless we act with policy, it's also a very personal account of what the coded gaze means to Buolamwini herself. The book culminates with Buolamwini advising the President of the United States, Joe Biden, on the harms of AI and its bias on Black, Asian, and Indigenous communities, and advocating for biometric rights for all Americans.
The key takeaways that Buolamwini leaves us with, along with beautiful analytical poetry, is The Cost of Inclusion and The Cost of Exclusion. Inclusion - improve the data and advocate for a more inclusive AI. But that could lead to more risk associated with being seen. Exclusion - Don’t improve the software, but there are risks with misidentification and symbolic annihilation. Buolamwini stresses that both have downfalls, and anyone assessing these systems should look at these issues from a sociotechnical point of view. Understanding the human-computer interaction function is just one component, but the technological interdependence in our lives, culture, and communities is a critical component when creating and implementing AI.
This work goes beyond classroom friendly. One feels they are having an engaging conversation with the author personally, creating space for all of us in-between the lines. You feel immersed in her world, and are brought to tears through the ups-and-downs of her successes and stresses. This roller-coaster journey through the annals of AI is a must-read for those that are seeking to understand the dangers of unregulated and untested AI, graduate school life, the importance of citing those whose work inspired you, and a deep-dive into the coded bias that surrounds us all. More importantly, this book is a directive and call-to-action for representation in the tech industry, AI regulations, and consumer/citizen privacy protections.
Byline:
Angelica Kalika, University of Colorado Boulder
Angelica Kalika, PhD, is an Assistant Teaching Professor at the University of Colorado Boulder, Journalism Department, in Boulder, Colorado. Her research interests include nonprofit media, media business models, and news influencing.
Comments