This article explains Aristotle’s doctrine of the mean — virtue as the middle state between excess and deficiency — and how to use it as a calibration tool for AI regulation without collapsing into split-the-difference compromise.
What Aristotle argues (NE Book II, chapters 6–9) #
For Aristotle, every moral virtue sits between two vices: one of excess and one of deficiency (1106b–1107a). Courage is the mean between recklessness and cowardice; generosity between wastefulness and stinginess. Three refinements keep the doctrine from becoming a platitude:
- The mean is relative to us, not arithmetic. The right amount depends on the agent and the circumstances — what is moderate for one situation is excessive for another (1106a-b).
- The mean is defined by right reason — it is where the person of practical wisdom (phronimos) would locate it, given the particulars (1106b36–1107a2).
- Some actions have no mean at all. Aristotle names acts like theft and murder: they are wrong in themselves, and there is no “right amount” of them (1107a). The doctrine is not a license to moderate the unacceptable.
Practical advice follows in chapter 9: since hitting the mean is hard, steer away from the more damaging extreme, and correct for the direction you naturally drift (1109a-b).
The policy translation: a vocabulary for calibration #
AI policy debates are routinely framed as binary — innovate or regulate, adopt or ban. The doctrine of the mean replaces the binary with a calibration exercise. For each governance question, name the excess, the deficiency, and the contextual factors that locate the mean:
| Governance virtue | Deficiency (vice) | Excess (vice) |
|---|---|---|
| Transparency | Opacity: unexplained decisions, secret systems | Indiscriminate disclosure that enables gaming, breaches privacy, or drowns citizens in noise |
| Courage in adoption | Timidity: refusing beneficial tools out of vague fear | Recklessness: deploying unproven systems on vulnerable populations |
| Oversight | Negligence: no review, no monitoring | Paralysis: process so heavy nothing ships and shadow IT flourishes |
| Data collection | Flying blind: no evidence base for decisions | Surveillance: collecting because you can |
| Vendor reliance | Refusing all external expertise | Dependence that hollows out internal capacity and judgment |
Two warnings #
The mean is not the midpoint of the lobbying positions. “Relative to us” means relative to the facts — stakes, reversibility, affected populations — not relative to whoever shouted loudest. A well-funded push toward one extreme does not move the mean; it only moves the noise.
Some AI uses have no mean. Just as Aristotle held that certain acts are simply wrong, some applications (e.g., systems designed to manipulate citizens or strip them of legal agency) are not candidates for calibration. The first regulatory question is always: is this a mean-seeking case at all?
For a step-by-step method, see Applying the Mean: Calibrating Between Innovation and Precaution.
