The Distributional Audit

In the AI Story

The proposal has direct precedent in environmental and social impact assessment regimes. Before a new port can be built, environmental impact assessments are legally required in most jurisdictions. They don't prevent construction; they make costs legible to the political process that approves or constrains the project. No comparable framework exists for cognitive infrastructure, despite AI's comparable or greater reach into human lives.

The audit's components would include: internal organizational measurement (worker hours, schedule extension, task intensification, turnover linked to AI workload), household-level distributional tracking (via partnered research with workers' families — analogous to occupational health surveillance), supply chain tracing (annotation labor conditions, open-source contribution accounting, energy consumption per productive unit), and longitudinal community monitoring (the Berkeley study extended, institutionalized, and scaled).

The resistance to such audits is predictable. Organizations argue that the measurement is subjective, methodologically unstable, or commercially sensitive. Cowen's historical evidence suggests these objections are structurally identical to the 20th-century objections to common rule wage surveys, workplace injury reporting, and environmental disclosure — objections that were defeated not because they were wrong but because organized political pressure made them untenable.

The audit is not sufficient on its own. Counter-logistical movements have consistently demonstrated that measurement without enforcement mechanisms produces documentation of harm rather than reduction of harm. But measurement is a prerequisite: political mobilization cannot organize around invisible costs.

Origin

The concept synthesizes environmental impact assessment methodology, occupational health surveillance, and supply chain transparency frameworks. Cowen developed the AI-specific application in her 2025 Infrastructure Otherwise seminars, drawing on her collaboration with labor researchers examining platform work.

Key Ideas

Measurement precedes politics. Invisible costs cannot be collectively addressed; the audit's purpose is to render them collectively visible.

The ledger must have two sides. Current productivity metrics track only benefits, producing a structurally dishonest accounting.

The audit extends beyond the organization. The cognitive supply chain crosses borders; so must its distributional accounting.

Measurement without enforcement is documentation, not remediation. The audit is necessary but not sufficient.

Debates & Critiques

Critics argue distributional audits risk reducing complex human experiences to metrics, reproducing the quantifying logic they aim to contest. Cowen concedes the tension but argues that the absence of measurement has produced worse outcomes than imperfect measurement — that the perfect epistemology is the enemy of the political response the situation requires.