All their AI development scenarios rely on exploitation of racial minorities for innovation---algorithmic governance and logistics turn bodies into commodities for profit---reinforcing a self-perpetuating cycle of racialization that encodes itself into AI.

Labor Tax: How Underpaying Minority Workers Fuels AI Development In their investigative piece The Exploited Labor Behind Artificial Intelligence, Williams, Miceli, and Gebru [WMG22] expose the harsh realities faced by gig workers in the AI industry, who are disproportionately drawn from vulnerable populations. Big tech players in the AI industry such as Amazon and Facebook heavily rely on these laborers, including data labelers, content moderators, and delivery drivers. These workers, despite their indispensable contributions, often endure abysmal pay, constant surveillance, unrealistic quotas, and even safety hazards. Further details from the investigation reveal that data labelers, responsible for critical AI training data, can earn as little as $1.77 per task. In the troubling case of content moderators in Kenya, they are relentlessly monitored to make split-second decisions, even when confronted with disturbing content. These workers, whose labor sustains billion-dollar companies, are unjustly compensated. Izaguirre [I23] discussed a very recent example of labor exploitation within the gig economy 3 from the ridesharing giants, Uber and Lyft. In New York, these companies are set to pay a staggering $328 million to settle complaints that they wrongly imposed taxes and fees on their drivers that should have been covered by passengers. State Attorney General Letitia James unveiled this settlement, emphasizing how Uber and Lyft—which incorporate AI technologies for routing and pricing algorithms—had systematically deprived their drivers of millions of dollars in earnings and essential benefits. The drivers, akin to those engaged in data labeling for AI models, toiled for long hours in challenging conditions while being undercompensated. The repercussions of such labor exploitation have wide-ranging implications, demonstrating the persistence of these issues across different sectors of the gig economy. These drivers, often working long hours in challenging conditions, waited an arduous eight years for justice, emphasizing the enduring time burden in their battle for rights–we will explore the time dimension in the next subsection. Moreover, racial capitalism also rears its head in other ways in generative media AI applications. Startups have built systems to synthesize fake profile photos by exploiting images of minority models without consent or payment. The marginalized provide raw material—their unvalued data and likenesses—to advance technologies from which they do not benefit. An example of this is the Argentina-based design firm, Icons8, which built generative AI systems to synthesize fake profile photos by exploiting minority likenesses [H20]. Though the models’ images provided core training data, they received no payment for this labor, nor any share of the systems’ profits derived from their inputs. Once again, capitalism fueled AI development by commodifying bodies. The models’ data and likenesses became raw materials for tech advancement from which they did not equally gain. This case provides further evidence that we must scrutinize and address the human exploitation underlying much AI progress. Time Surcharge: The Burden on Minorities to Fix Biased Systems In 2019, a pivotal federal study by NIST unveiled the entrenched racial biases within facial recognition algorithms, revealing higher false positive rates for Asian, African American, and Native American faces compared to white counterparts [H19]. This disparity is not merely a technical oversight but a reflection of a broader societal issue where fairness is often assessed through a predominantly white lens, perpetuating and amplifying harmful racial stereotypes. These algorithms, failing more frequently with Black and Asian individuals, not only reinforce the misconception of homogeneity within these communities but also echo Cedric Robinson’s argument that capitalism morphs pre-existing social differences into racial disparities, using technology to further these divides. These biases extend beyond race to gender and age, with algorithms showing elevated error rates for women, the elderly, and the young compared to middle-aged white men. Simone Browne, in Dark Matters [Br15, p. 113], elucidates how these systems perpetuate stereotypes, such as misclassifying Asian males as females and Black women as men, illustrating how AI models, like capitalism, disseminates racial and gender stereotypes on a massive scale under the guise of technological neutrality. The case of Brandon Mayfield, as discussed by Browne, exemplifies how biases in AI systems can have real-world consequences. A veteran and a lawyer who had recently converted to Islam, Mayfield was erroneously matched by FBI algorithms to a fingerprint found on a bag containing a detonating device linked to the 2004 Madrid bombing [Br15, p. 115]. Despite being one of twenty matches, his recent conversion to Islam and status as a veteran, both associated with prejudicial stereotypes, led authorities to single him out. Mayfield’s personal life and beliefs were publicly exposed under a false accusation, and he remained entangled in legal battles to clear his name until November 2010. This wrongful identification underscores how minoritized groups not only bear the brunt of these biases but also face a “Privacy Levy,” paying a significant personal and financial price due to prejudiced technological systems. This systemic bias embedded within AI technologies serves capitalist ends in several ways. Firstly, it perpetuates a cycle where the burden of correcting and navigating biased systems falls disproportionately on marginalized communities, consuming their time and resources—a “Time Surcharge” that benefits the status quo by 4 maintaining existing power dynamics. Secondly, the deployment of such technologies, especially in law enforcement and surveillance, opens lucrative markets for tech companies, capitalizing on the state’s desire to monitor and control, often at the expense of those already marginalized. In essence, these biased algorithms are not just products of existing societal inequalities; they are tools that capitalism exploits to further entrench its dominance, ensuring that the costs of innovation are borne by those least able to bear them, while the profits accrue to those already in positions of power. It is important to point out that the concept of a “Time Surcharge” extends beyond the realm of AI and technological bias, reflecting a broader systemic issue that minoritized communities face across various bureaucratic and social systems. This enduring burden of time, as analyzed in the context of racialized administrative burdens by Ray, Herd, and Moynihan in Racialized Burdens: Applying Racialized Organization Theory to the Administrative State [RHM22], is often utilized as a means to punish and control, reinforcing poor living conditions. Similar to the ways in which biased algorithms require marginalized groups to navigate and correct flawed systems, these bureaucratic processes consume excessive amounts of time, serving as barriers that maintain social and economic inequalities. These practices are reminiscent of historical strategies designed to disenfranchise and exploit marginalized populations, perpetuating a cycle where the costs of bureaucratic inefficiency and technological innovation are disproportionately borne by those least equipped to handle them, while the systems in place continue to benefit those in power. The development and implementation of biased AI systems are stark manifestations of racial capitalism, where technological advancements are leveraged to amplify and profit from racial and social disparities. The need for a concerted effort to address these biases is not just a matter of technical accuracy but of challenging the capitalist structures that incentivize and benefit from the perpetuation of inequality.