More Than a Glitch: Confronting Race, Gender, and Ability Bias in Tech

There was a distinct moment—the start of the first full paragraph on page 16, to be precise—when I looked up from Meredith Broussard's More Than a Glitch: Confronting Race, Gender, and Ability Bias in Tech, and smiled. She'd been talking about home loans and the various factors that might contribute to a bank's decision to approve or deny a mortgage application. She'd introduced a few considerations, such as credit score and length of employment, and had plotted the two variables against each other on a series of charts. The first chart was a straight line—easy to interpret—but this one looked more like a mountaintop: “a bell curve,” Broussard explained, “aka a nonmonotonic shape” (16). And there it was—a discussion of nonmonotonicity in a trade book. And so I smiled, both because it was classic Broussard—nerdy yet approachable—and because I could tell exactly where she was going: on to a discussion of the multidimensional nonmonotonic relationships that characterize the parameters of machine learning models like the ones that underlie ChatGPT.If that last sentence threw you, but you're still reading along, then this book is for you. More Than a Glitch is filled with clarifying explanations of all sorts of math-y and computational concepts, introduced through stories about real people in the world that make clear why you should care. Crucially, with Broussard's writing, it's never a hard sell. One of my favorite aspects of More Than a Glitch is how you often don't realize that you've entered into one of these technical deep dives until you're already halfway through, since Broussard's informal tone and narrative-driven storytelling does most of the work of bringing the reader along.In this way, More Than a Glitch resembles Broussard's previous work, Artificial Unintelligence: How Computers Misunderstand the World (2018), but with the increased urgency and heightened stakes of the present moment. In Artificial Unintelligence, for example, Broussard took her readers to an empty parking lot in South Philadelphia to observe a field test for an early self-driving car (verdict: unimpressive even then), and then back in time to the decks of the Titanic for a hypothetical “survivor exercise” that demonstrated the basic steps involved in supervised machine learning (verdict: masterful pedagogical feat!). Fast forward to the present, and actual self-driving cars have come and (partially) gone from the streets of San Francisco after one company's robo-taxi ran into a woman and dragged her for nearly twenty feet, and historical shipwreck scenarios are being replaced by a series of summits at the White House on AI's “existential risk.” While it would seem deeply contradictory (and in fact is deeply contradictory) that the people most concerned about our future AI overlords are the very same who are continuing to bankroll companies like Cruise and Waymo, even as their cars fail to detect pedestrians when they are in the process of running them over, these fears and these actions are both of a piece with the technological hubris—that defining characteristic of Big Tech under late capitalism—that sets More Than a Glitch in motion.Broussard begins the book by describing how we have arrived at this dystopian place, a virtual landscape littered by “thousands of abandoned apps and defunct websites and failed platforms,” with millions of dollars lost to technologies that “marketers promised would ‘revolutionize’ and digitize everything from rug-buying to interstellar travel” (3). Her explanation for this digital debris field: technochauvinism, “a kind of bias that considers computational solutions to be superior to all other solutions” (2). Technochauvinism was a concept that Broussard introduced in Artificial Unintelligence, but in More Than a Glitch, she raises its stakes: “Embedded in this bias is an a priori assumption that computers are better than humans—which is actually a claim that the people who make and program computers are better than other humans,” she asserts (3). The danger of technochauvinism, as we have seen time and time again—and which gets Broussard to the central argument of her book—is that the people who make and program computers are just as susceptible to the “very human factors like self-delusion, racism, bias, privilege, and greed” (3) as anyone else—and arguably more so, given the money and power that continue to accrue around them, enabling their perpetual rise and insulating them from the real-world consequences of technologies that they create.The main argument of More Than a Glitch is exactly that: that the racism, sexism, ableism, and other biases embedded in our technology systems “are more than mere glitches; they're baked in from the beginning” (4). Furthermore, because these are “structural biases,” Broussard rightly asserts, “they can't be addressed with a quick code update” (4). Instead, what is required is a wholesale examination of “how the biases of the real world take shape inside our computational systems” (4). This examination is what Broussard sets out to do.The introductory chapters give way to a series of story-driven case studies, each structured around a specific algorithmic system and the specific people who experience its harm. Broussard's journalistic training shines through in how she employs dialogue and narrative detail in order to draw new insight from otherwise familiar examples from the AI-ethics canon. The story of Robert Williams, a Black automotive supply salesman from the Detroit suburbs who was falsely arrested because of an algorithmic mismatch between his face and that of a robbery suspect, frames Broussard's discussion of the racial biases embedded in facial recognition software—a topic first brought to national attention by Joy Buolamwini and Timnit Gebru in their article “Gender Shades: Intersectional Accuracy Disparities in Commercial Gender Classification” (Proceedings of Machine Learning Research, 2018) and related advocacy work. Robert McDaniel, a Black Chicago man identified by a crime-risk assessment algorithm as being likely to be involved in a shooting (46), becomes Broussard's way into an examination of how predictive policing software amplifies the racism permanently entrenched in the US policing system, an issue that rose to public prominence through Cathy O'Neil's Weapons of Math Destruction (2016). And the experience of Jonathan Ferguson, a transgender technical writer from the UK, becomes a pathway into an exploration of the trans-exclusionary nature of binary classification systems, as initially identified through the archival work of Mar Hicks in “Hacking the Cis-Tem” (2019). In Broussard's accounts of these issues, however, it's the people and not the systems who take center stage. It's a compelling rhetorical tactic that refuses to allow anyone off the hook for believing that the harms these people experienced are mere exceptions rather than the rule.As the book unfolds, Broussard layers in interviews with subjects of algorithmic harm, contributing additional examples and fresh detail to this necessary body of work. She speaks with Isabel Castañeda, a bilingual high school student from Colorado who, when her International Baccalaureate Spanish exam was canceled because of the COVID-19 pandemic, unfairly received an algorithmically predicted low score instead. Broussard allows Castañeda herself to make the case for the injustice of a system that disregarded the evidence of a student who “earned an A+ every semester” in her advanced AP and IB courses yet received a failing grade on the canceled exam based on her school's historical performance (66). Another chapter documents Broussard's conversations with Richard Dahan, an Apple Store employee who, after years of stellar workplace performance, began to experience a series of job-related injustices because he is deaf. Here, Dahan's story opens up a broader discussion about the range of unaddressed issues of accessibility in tech.Already compelling, More Than a Glitch reaches new levels of persuasiveness and poignancy when Broussard places herself center stage. Two later chapters of the book deal with the author's personal experience as a multiracial woman attempting to navigate the US healthcare system—first in her repeated attempts to accurately document her racial identity in her medical providers’ electronic records system, and second, and more acutely, in her experience of being diagnosed with breast cancer via a machine learning algorithm. Broussard investigates her own diagnosis as she would any other lead: by verifying claims made by corporations, tracking down evidence along circuitous paths, and interviewing sources for supplementary information. Her account of her attempts to obtain copies of her mammography scans in order to audit the AI system that diagnosed her with cancer is carried by a narrative of journalistic sleuthing, involving covert screenshots, additional files that could only be transmitted via CD-ROM, and the fortuitous involvement of her downstairs neighbor, a data scientist working on breast cancer detection AI. The result is a story that keeps the reader continually engaged while its broader stakes—the incursion of AI, with its racial and other biases, into the healthcare system—continue to compound.Broussard is explicit about the fact that the problems with AI that she discusses throughout the book are highly complex and are deeply embedded in structural inequalities expressed at a societal level. “Unfortunately, there isn't a magic switch that we can flip to make technology work for everyone, everywhere,” she states at the outset (7). But nor does she simply throw up her hands. The final chapters of the book pivot from problems to potential solutions that may strike readers as incommensurate with the scale and scope of the challenges so deftly explored. Broussard is explicit about this as well: she begins the section by calling out “solutions journalism” for setting up society to “expect that there is an easy answer to computational problems” (157). In its place, she proposes “technology that is in the public interest” (158). Broussard outlines the history of public interest technology and locates examples in US government initiatives and EU regulatory policies, as well as in a range of algorithmic auditing approaches. She then surveys some examples of early successes: an antidiscrimination lawsuit leveled at the algorithm used by the NFL to determine brain injury payouts, the EU AI Act (which will soon go into effect), and individuals who have turned to social media to ensure public accountability. It's hard not to come away from the book with the sense that these efforts will need to increase by multiple orders of magnitude to fully rebalance the current asymmetries of money, power, and influence that characterize Big Tech, but that is also the point. “Tech is racist and sexist and ableist because the world is so,” Broussard clearly states (188). And for this reason, this world—our world—has always required the efforts of activists and individuals, students and scholars, citizens and their governments, to continually work to “get us closer to a world that is truly more just” (188).