CONCEPT
Limitation Disclosure
The design principle that tools built for intelligence augmentation should make their limitations visible rather than hiding them—'I am uncertain' when uncertain, not confident wrongness dressed in polish.
Limitation disclosure is Winograd's second principle for AI collaboration design: a system designed to support rather than replace human understanding should make its limitations visible, not conceal them. The language interface's characteristic failure mode—confident wrongness dressed in polished prose—is a design problem precisely because the confidence conceals the limitation. A system that flagged its own uncertainty when uncertain, that marked pattern-matching as pattern-matching rather than presenting it as insight, that disclosed the specific conditions under which its outputs become unreliable, would be a less fluent but more honest collaborator. Winograd argued decades before language models existed that systems designed assuming they understand would fail in ways systems designed with honest awareness of limitations would not—the argument applies with greater force now, because language models' failures are harder to detect than expert systems' diagnostic errors.
In The You On AI Field Guide
The principle runs counter to the commercial incentives shaping AI development. Fluency, confidence, and seamless performance are marketable; explicit uncertainty and limitation disclosure reduce perceived capability. A system