- Six framings
- 1. "AI policy is character policy for institutions."
- 2. "The question isn't for or against AI — it's where between two failures."
- 3. "Automation should serve human flourishing — and we should be able to say how."
- 4. "Rules run out. Judgment doesn't. Keep judgment in the loop."
- 5. "'The algorithm did it' is a plea of ignorance — and avoidable ignorance doesn't excuse."
- 6. "Every algorithm needs an appeal and an override — that's due process, and it's older than due process."
- Objection handling
- Related articles
Ready-to-use framings for briefings, testimony, and op-eds — each with a one-liner, a supporting point, and the source passage — plus responses to the three most common objections.
Six framings #
1. “AI policy is character policy for institutions.” #
Support: Agencies become what they repeatedly do. A one-time ethics review builds nothing; recurring practices — audits, incident reviews, red-teaming — build agencies that handle AI well by default. Source: NE II, 1103a-b.
2. “The question isn’t for or against AI — it’s where between two failures.” #
Support: Refusing beneficial tools harms citizens through delay and backlog; deploying unproven tools harms them through error. Good regulation names both failure modes and calibrates between them based on stakes and reversibility. This framing defuses the innovation-vs-safety false binary that stalls legislation. Source: NE II, 1106b–1109b.
3. “Automation should serve human flourishing — and we should be able to say how.” #
Support: Every government AI deployment should state, in one citizen-readable sentence, the human good it serves. If the sponsor can’t write that sentence, that’s the finding. Flourishing language travels across party lines better than either “innovation” or “risk.” Source: NE I, 1094a–1098a.
4. “Rules run out. Judgment doesn’t. Keep judgment in the loop.” #
Support: No rule written in advance anticipates every case — Aristotle’s core insight about practical decisions. High-stakes automated processes need a human with genuine authority to depart from the recommendation, staffed by people experienced enough to exercise it. Source: NE VI, 1140a–1142a.
5. “‘The algorithm did it’ is a plea of ignorance — and avoidable ignorance doesn’t excuse.” #
Support: If an agency could have tested, monitored, or asked and didn’t, responsibility stands. Legislate a duty to know: documentation, disaggregated monitoring, decision logs, and a named accountable official per system. Source: NE III, 1110b–1114a.
6. “Every algorithm needs an appeal and an override — that’s due process, and it’s older than due process.” #
Support: Law speaks in universals and fails in particulars; equity corrects it. Automated systems are universals at machine speed. Notice, reasons, human appeal, and remedy aren’t anti-technology — they’re the 2,300-year-old fix for exactly this failure mode. Source: NE V, 1137a–1138a.
Objection handling #
“This is too abstract for legislation.” Every framing above lands on a concrete lever: purpose clauses, calibration criteria, judgment-preservation requirements, duty-to-know standards, appeal rights. See Model Policy Language for draft text.
“Whose virtues? This smuggles in one worldview.” The operative concepts — proportionate calibration, responsibility for avoidable ignorance, rectification of wrongs, judgment where rules fail — are shared across ethical and legal traditions. Aristotle supplies the architecture, not a sectarian content. And the criterion-of-merit point (1131a) actually requires the value choices to be made in public deliberation rather than smuggled in by anyone.
“We need enforceable rules, not philosophy.” Agreed — and this framework tells you which rules to write and why they hold together. Rules without a rationale fragment under lobbying; rules grounded in a coherent account of responsibility and justice are defensible in committee, in court, and in public.
