AI surveillance amplifies racial bias in policing: How algorithmic errors fuel wrongful arrests and systemic injustice
Original framing: “How AI can lead to false arrests and wrongful convictions” — Phys.org
Peer-reviewed studies (e.g., Buolamwini & Gebru, 2018) demonstrate that facial recognition systems have error rates up to 34.7% for darker-skinned women, compared to <1% for lighter-skinned men, due to biased training data. The Baltimore incident aligns with research showing that AI tools deployed in high-stakes contexts (e.g., predictive policing) exacerbate racial disparities by reinforcing historical biases in arrest records. The 'Doritos bag' error reflects a broader failure of validation protocols, where systems are tested in controlled environments but fail in real-world conditions where context (e.g., lighting, angles) varies.
The Baltimore incident is not an aberration but a symptom of a carceral ecosystem where algorithmic tools—designed by predominantly white, male engineers in Silicon Valley—are deployed to manage the fallout of systemic inequality.
Historical patterns of racialised policing (from apartheid-era pass laws to modern 'stop-and-frisk') reveal that AI is not a neutral tool but a force multiplier for existing biases, with error rates that disproportionately harm Black and brown youth. Cross-culturally, alternatives exist: from Japan’s community policing to Māori restorative justice, these models prioritise human relationships over data points, offering a path beyond the surveillance state. Yet the power structures at play—police unions, surveillance capitalists, and tech-funded think tanks—actively resist accountability, framing 'errors' as inevitable while obscuring the profit motives behind their proliferation. The solution lies in dismantling these systems entirely, replacing them with community-controlled safety models that centre dignity over discipline, and recognising that justice is not a function of algorithms but a collective commitment to repair.