This article explains Aristotle’s analysis of voluntary action and moral responsibility (Book III) and applies it to the most common evasion in AI governance: “the algorithm did it.”
What Aristotle argues (NE Book III, chapters 1–5) #
Aristotle asks when praise and blame are appropriate. His answer: actions are involuntary only when they result from external force or from non-culpable ignorance (1109b–1110a). Everything else is voluntary, and the agent answers for it.
Three refinements are directly relevant to AI:
- Mixed actions are still ours. A captain who jettisons cargo in a storm acts under pressure, but still chooses — the action is voluntary, though we may judge it leniently (1110a). Deploying a flawed system “because the deadline was political” is a mixed action, not an excuse.
- Culpable ignorance does not excuse. Aristotle distinguishes acting in ignorance from acting by reason of ignorance. If the ignorance itself was avoidable — the drunk person, the person who never bothered to learn what the law requires — responsibility stands, and lawgivers rightly punish (1110b–1111a, 1113b–1114a).
- We are responsible for our dispositions. Character is built by our own repeated choices, so we cannot plead our settled habits as an external force (1114a-b).
The policy translation: closing the accountability gap #
“The model is a black box, so no one is responsible” is, in Aristotelian terms, a plea of ignorance. The framework asks one question: was the ignorance avoidable? If an agency could have tested, documented, monitored, or asked — and did not — the ignorance is culpable, and responsibility attaches with full force.
This yields concrete design requirements:
- A duty to know. Agencies deploying consequential AI must be required to understand system behavior to the degree reasonably achievable: documented training data provenance, tested failure modes, monitored performance across populations. Making ignorance expensive makes it rare.
- Named responsibility. Every consequential system needs a named accountable official — a person, not a committee — whose sign-off certifies the duty to know was met. Aristotle’s framework attaches responsibility to choosers; diffusion across a vendor, an agency, and a help desk is how everyone escapes.
- A responsibility chain, not a responsibility gap. Vendors answer for what they represented; procuring officials for what they verified; operators for how they used the system; supervisors for the pressure they applied. Mixed actions attract lenient judgment, not immunity.
- No credit for engineered ignorance. Contracts or trade-secret claims that prevent an agency from knowing how its own systems decide are agreements to remain culpably ignorant — procurement rules should treat them as disqualifying.
