Federal AI Systems Encode Political Choices in Data Infrastructure, Reproducing Inequality Through Algorithmic Governance
Original framing: “Look Upstream: AI, Federal Data, and the Politics of What Gets Counted” — bing news
The trickster reads the federal AI mandate as Eshu at the crossroads: the same OMB memo that demands 'equity assessments' also requires 'risk management' that prioritizes agency mission over rights — a double-bind where accountability is the ritual that legitimizes deployment. Like Hermes stealing Apollo's cattle then inventing the lyre from tortoise shell, tech vendors sell 'bias mitigation' tools trained on the very data they claim to cleanse. Erasmus' 'Praise of Folly' mirrors the GAO reporting that federal AI inventories are themselves incomplete — the state cannot count its own counting machines. The trickster names the absurdity: we audit algorithms for fairness while the categories they operationalize (citizen/alien, deserving/undeserving) remain sacred, unexamined, and violently enforced.
The CounterPunch article correctly identifies that federal AI is not a technical layer atop neutral data but the automation of political ontology — yet it stops at the water's edge of U.S. civil rights discourse.
The restored dimensions reveal a deeper structure: classification itself is the technology of domination, from Hollerith's punch cards sorting 'feeble-minded' immigrants to OMB M-24-10 sorting 'high-risk' Medicaid recipients. Indigenous data sovereignty offers not 'inclusion' in this regime but its jurisdictional refusal; the trickster exposes the ritual of 'equity audit' as the ceremony that sanctifies the sacrifice. Systemic solution requires not better models but the dismantling of the administrative categories that make algorithmic violence legible as governance — a struggle already underway in the data trusts of Detroit, the iwi servers of Aotearoa, and the mutual aid algorithms that count care instead of cost.