The Blind Algorithm: Invisible by Default, Verifiable by Right

Abstract

In the age of generative AI, invisibility is a feature, not a bug. The best algorithmic systems disappear into use—shaping decisions, curating information, structuring choices without friction. But invisibility without verifiability is domination. When subjects cannot verify the processes that affect them, they are governed without knowing it and without recourse.

This paper introduces the Blind Algorithm as an architectural pattern for what I term epistemic agency: the capacity to verify—not merely trust—the algorithmic processes that shape one's life. The framework serves dual functions: enabling institutions to prove the fairness of their decisions, and enabling individuals to verify the influences that shaped what they believed were autonomous choices.

Drawing on Foucault's governmentality, Bourdieu's symbolic violence, and building on recent work on algorithmic governance (Rouvroy & Berns, Zuboff, Yeung), the paper argues that verification is not opposed to invisibility but its necessary complement. The paper contributes: (1) a four-regime typology of algorithmic governance, (2) an architectural pattern—the Blind Algorithm—for verifiable decisions, and (3) the Audit Dependency Profile (ADP) as a conceptual bridge to distinguish internalized compliance from internalized integrity.

This is a theoretical and architectural proposal, not a finished system or empirical validation. Organizational justice research and one legal case (Brazil's Portaria 612) serve as illustrative anchors, not comprehensive evidence.

Keywords: algorithmic governance, governmentality, Foucault, Bourdieu, ISO 24153, auditable randomization, procedural justice, epistemic agency, generative AI