Answers are grounded in this paper and cite their sources.
Mandatory human review is the principal safety valve written into proposals for regulating artificial-intelligence pesticide advisory systems in smallholder agriculture, yet the trigger that activates review is almost always specified as a fixed confidence threshold rather than as a governance instrument calibrated to the political economy of the regulator. It treats the override rate — the share of advisory outputs that a calibrated trigger routes to human review — as the object of institutional choice, and derives the conditions under which a trigger chosen by a captured regulator falls below the rate required to protect smallholders. The analysis is theoretical and methodological: no observational data are used or claimed. The bound is expressed as a function of the capture-contingent cost the regulator attaches to overriding industry-preferred advisory outputs, the smallholder safety loss from an unreviewed erroneous recommendation, and the base rate of high-consequence advisories. A trigger calibrated to minimise expected regulatory cost under capture is systematically more permissive than the safety-preserving trigger, and that the gap widens as the capture weight rises. It proposes an override-rate floor as a governance override-rate bound, specifies the audit information needed to enforce it, and identifies the conditions under which the bound is binding. The contribution is a calibration discipline that makes the political vulnerability of the trigger explicit and auditable rather than implicit in a technical threshold.
No takes yet. Share an insight, caveat, or question.
Aïcha Aït Moussa (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: