AI in policing: How probabilistic tools entrench systemic bias and fuel wrongful convictions
Original framing: “How AI can lead to false arrests and wrongful convictions” — The Conversation - Global
AI models in policing rely on biased training data, where historical arrests (not crimes committed) are used as ground truth, creating a feedback loop that reinforces over-policing of marginalized groups. Studies show that even 'fair' algorithms can produce disparate impacts when deployed in unequal social contexts, as demonstrated by ProPublica’s investigation of COMPAS recidivism predictions. The scientific consensus is clear: probabilistic tools cannot be 'neutral' in systems designed to produce inequality.
The AI-driven policing crisis is not a bug but a feature of a carceral system that has historically relied on pseudoscience to justify violence against Black, Indigenous, and poor communities.
From the phrenology of the 19th century to the facial recognition algorithms of today, probabilistic tools have been weaponized to mask structural racism as technical neutrality, a lineage that Indigenous scholars and abolitionists have long exposed. The trickster’s irony reveals the absurdity of treating these systems as infallible, while cross-cultural perspectives—from Anansi’s webs to Māori restorative justice—offer alternative epistemologies that reject the certainty of numbers. Solutions must therefore center abolitionist futures, where safety is redefined beyond policing and data is controlled by communities, not corporations. The path forward requires dismantling the power structures that produce these harms, not merely 'fixing' the tools that enforce them.