Skip to content
AI + Quantum Ethics Standards

Quantum Standard: Optimization Ethics & Decision Delegation

1. Purpose

Prevent harmful delegation of consequential decisions to quantum-accelerated optimization systems, especially where objectives can encode unfairness, coercion, or destabilizing incentives.

2. Applicability

  • Applies to quantum (and hybrid) optimization used in resource allocation, logistics, finance, security, hiring, admissions, healthcare, and other high-impact domains.
  • Strongest requirements apply to Tier 2–3.

3. Ethical Mapping

  • A2 Human Dignity & Moral Agency
  • A1 Unity & Social Cohesion
  • A5 Proportionality & Moderation

4. Requirements (Normative)

Q-O-1 (Objective Disclosure). For Tier 2–3, operators MUST document the optimization objective(s), constraints, and any proxy variables that materially shape outcomes.

Q-O-2 (No Autonomous Moral Delegation). Systems MUST NOT be used to autonomously select or recommend actions that constitute moral/legal determinations about persons (e.g., guilt, punishment, coercive targeting) without a defined human decision authority and contestability pathway.

Q-O-3 (Constraint Safety). Tier 2–3 operators MUST implement constraints and guardrails that prevent:

  • prohibited discrimination
  • unsafe operational outputs (e.g., violating safety margins)
  • objectives that incentivize deception or coercion

Q-O-4 (Sensitivity & Stability Testing). Operators MUST evaluate sensitivity of outcomes to:

  • objective weighting changes
  • constraint perturbations
  • data uncertainty

For Tier 2–3, results MUST be recorded and reviewed before deployment.

Q-O-5 (Human Oversight and Stop Conditions). Tier 2–3 workflows MUST include:

  • an identified decision owner responsible for approvals
  • stop conditions when outputs exceed risk thresholds
  • periodic review of whether objectives remain socially legitimate

5. Compliance Evidence

  • objective/constraint documentation and code references
  • prohibited-use policy with enforcement controls
  • sensitivity test reports and decision memos
  • oversight logs showing approvals and stop-condition triggers

6. Rationale (Non-normative)

Optimization often hides normative choices inside weights and constraints. Making those choices explicit is necessary for fairness, consent, and accountability.

7. Failure Modes & Abuse Cases

  • encoding inequity via proxy variables
  • “objective laundering” (claiming neutrality while embedding values)
  • brittle solutions that fail under minor perturbations

8. Change Log

  • v0.1: Initial draft.