Governance as Optimization vs. Authorization

The Two Framings

Governance as optimization: AI governance is a technical problem. Define risk metrics, set thresholds, enforce compliance. The goal is to constrain AI systems within acceptable bounds.

Governance as authorization: AI governance is a political problem. Determine who decides, through what processes, with what legitimacy. The goal is to ensure AI systems serve democratically authorized purposes.

Key Differences

Dimension	Optimization Frame	Authorization Frame
Primary question	Does this system meet standards?	Who authorized this system to decide?
Success metric	Compliance rate	Democratic legitimacy
Expertise role	Technical experts set and enforce rules	Multiple knowledge types contribute
Stakeholder involvement	Consultation	Co-governance
Democratic values	Constraints on optimization	The objective itself

The Information Architecture Connection

Format shapes cognition. How governance frameworks structure the governance problem shapes what solutions become thinkable.

Optimization framing produces:

• Risk taxonomies focused on measurable harms
• Compliance checklists
• Technical audit requirements
• Expert-driven assessment

Authorization framing produces:

• Participation mechanisms
• Legitimacy assessments
• Stakeholder deliberation processes
• Coalition-building requirements

Both can be rigorous. Both can fail. But they fail in different ways and miss different things.

When Each Applies

Optimization works when:

• Harms are individual and measurable
• Technical expertise captures relevant risks
• Democratic authorization can be assumed (established public mandate)
• Speed matters more than participation

Authorization required when:

• Harms are systemic and structural
• Technical metrics miss experiential knowledge
• Legitimacy is contested
• AI systems redistribute power in ways that affect who decides

Related: 02-atom—format-shapes-cognition, 05-atom—ai-redistributes-epistemic-authority, 05-molecule—democratic-risk-taxonomy