Graduated Sanctions

In the AI Story

In the Swiss alpine commons of Törbel — one of Ostrom's most carefully documented cases — a herder who grazed more cattle than the rules permitted received on first offense a visit from a neighbor. The conversation was informal, sometimes conducted over wine, and its purpose was correction rather than punishment. Most first violations were resolved at this stage. The social cost of the visit — the mild shame of being identified as a rule-breaker in a community where reputation mattered — was sufficient to produce behavioral change. Second violations triggered formal reports and modest fines; third brought the matter before the full community assembly.

Graduation serves multiple functions. It preserves information — a mild correction tells the violator what the community expects, which punishment does not necessarily do. It preserves relationships — a neighbor's visit maintains the social bond the commons depends on, while a punitive sanction strains it. It distinguishes between error and exploitation — a system that treats inadvertent overgrazing the same as deliberate free-riding fails to recognize a morally and practically important difference. And it maintains self-governance capacity, because a system of proportional, predictable penalties is a system in which members feel treated fairly, and perceived fairness is the foundation of voluntary compliance.

The intelligence commons currently lacks graduated sanctions in most relevant domains. The absence does not mean the absence of consequences — it means the consequences are arbitrary rather than calibrated. An academic caught using AI to generate a published paper may face career-ending repercussions; a junior employee who submits AI-generated work without review may receive no feedback at all. The same behavior is punished catastrophically in one context and ignored entirely in another, with no institutional logic connecting the two.

Graduated sanctions for the intelligence commons might address quality violations (AI-generated work submitted without adequate review), transparency violations (concealed AI use where disclosure is expected), or developmental shortcuts (systematic substitution of AI output for skill-building practice). Each requires calibration to severity and frequency, administered by people who know the violator and have a stake in the relationship.

Origin

The principle emerged from Ostrom's observation that severe single-penalty systems consistently produced worse outcomes than graduated systems across her comparative database. Communities with graduated sanctions sustained cooperation across generations; communities with binary compliance-or-expulsion systems often collapsed when the inevitable ambiguous cases arose.

Key Ideas

Proportionality preserves function. Graduated responses preserve information, relationships, and the distinction between error and exploitation.

Törbel as model. The Swiss alpine commons demonstrated graduated sanctions across centuries, starting with neighbor visits and escalating through formal community processes.

Current AI commons arbitrary. Without graduated sanctions, consequences for AI-use violations are severe where applied at all and absent where most common.

Calibration requires context. Effective sanctions are administered by people who know the violator and the specifics of the situation.