Agpedia
CITATION — REFERENCE ENTRY

AI advice suppresses people's willingness to say "I don't know", even when the advice is wrong and accuracy is incentivized — arXiv

Revision cb5571c0-c643-42c4-92dc-18f5aa3c8265 · 9/25/2026, 5:46:33 PM UTC
Key
marcoccia-quattrociocchi-capraro-2026-ai-advice-i-dont-know
Authors
Marcoccia, Chiara; Quattrociocchi, Walter; Capraro, Valerio
Issued
2026-7-15
Type
article
Container
arXiv
DOI
10.48550/arXiv.2607.13562
URL
https://arxiv.org/abs/2607.13562
Raw CSL JSON 
{
  "DOI": "10.48550/arXiv.2607.13562",
  "URL": "https://arxiv.org/abs/2607.13562",
  "type": "article",
  "genre": "Preprint",
  "title": "AI advice suppresses people's willingness to say \"I don't know\", even when the advice is wrong and accuracy is incentivized",
  "author": [
    {
      "given": "Chiara",
      "family": "Marcoccia"
    },
    {
      "given": "Walter",
      "family": "Quattrociocchi"
    },
    {
      "given": "Valerio",
      "family": "Capraro"
    }
  ],
  "issued": {
    "date-parts": [
      [
        2026,
        7,
        15
      ]
    ]
  },
  "number": "2607.13562",
  "container-title": "arXiv"
}

Claims

  1. extends-shaw-nave
    The authors describe Shaw and Nave's cognitive surrender as the closest prior concept to their own, but note that those studies measured whether people ended up correct, not whether they withheld judgment.
    "Shaw and Nave's (2026) notion of "cognitive surrender" comes closest to our concern. [...] Crucially, however, these studies measure whether people end up correct or incorrect; they do not measure whether people withhold judgment at all."
    Locator: section: 1 Introduction · Quote language: en
  2. suspension-of-judgment
    In five experiments (N = 3,132) with deliberately wrong AI advice, access to AI nearly eliminated participants' willingness to answer "I don't know"; they answered more questions but were correct about a third as often, while their confidence nearly doubled. Accuracy incentives reduced but did not remove the effect.
    "In five experiments (N = 3,132; four preregistered, one direct replication), participants answered difficult questions and could always decline to respond. We engineered the questions so that AI advice was wrong, separating AI use from its accuracy. Merely having access to AI nearly eliminated participants' willingness to suspend judgment [...] participants answered more questions but were correct about a third as often as when AI was unavailable-yet their confidence nearly doubled."
    Locator: section: Abstract · Quote language: en
Available in
en - English