Proof
Print a document, especially AI output, so every paragraph gets an active judgement in ink.
Variants
- default
Data fields
The fields a spec's page can set for this component (see switchback show proof --json):
sourceblocks
Grounding
Helps: when an AI draft needs checking point by point: a number and a margin on every paragraph invite an active judgement on each, and paper gives slightly better comprehension of expository text
Backfires: when marking turns into rubber-stamping every paragraph, or when the job is catching surface typos, which one study found easier on screen
- CPeople reviewing automated or AI output accept its errors at elevated rates, both acting on wrong outputs (commission) and missing unflagged problems (omission). Interventions that force active evaluation reduce this overreliance.Parasuraman & Manzey (2010), Lyell & Coiera (2016), Goddard, Roudsari & Wyatt (2012), Buçinca, Malaya & Gajos (2021), Mosier et al. (1998)
- BReading on paper yields better comprehension than reading the same text on a screen, especially for expository text read under time limits.Clinton (2019), Delgado et al. (2018), Ackerman & Goldsmith (2011), Calvo-Ferrer (2022)
- DA fixed, shared set of marks, each mapped to one editing action, is used reliably by writers and editors to communicate corrections on paper.Christiansen (2009), AMA Manual of Style