---
type: "synthesis"
tags: ["trust-calibration", "over-reliance", "automation-bias", "tension"]
articles: ["a037", "a040", "a079"]
id: "cross-trust-calibration-dilemma"
sources: ["adoption"]
sourceVaultSlug: "hbr-seg-adoption"
originDay: 9
articleStem: "hbr-seg-adoption"
sourceUrl: "(unified vault: 11 sources)"
sourceTitle: "HBR — People Ⅲ-A · Adoption / trust / literacy / psych-safety"
---
The corpus contains two *opposite* trust pathologies, and a careful reader must hold both — the goal is calibration, not maximization.

**Under-trust / justified skepticism.** A040's [[concept-agentic-ai-skepticism]] documents an 89% collapse in trust toward AI that *acts* (vs. 31% for AI that recommends). This resistance is often rational and is echoed by algorithm-aversion literature.

**Over-trust / blind acceptance.** A037's [[concept-willful-ignorance-in-ai]] shows people *want the answer but refuse the reasoning* — they accept AI output without interrogating it, especially when money or morals are at stake. A079's [[concept-human-ai-oversight-paradox]] shows humans grow overconfident, cognitively offload, and *reduce* oversight exactly where it should increase. A079 adds that [[claim-sustained-ai-use-undermines-confidence|sustained use erodes the confidence to challenge AI even when the human knows it is wrong]].

**The shared remedy** is to institutionalize challenge. A037: [[action-encourage-second-guessing]] and reward critical engagement. A079: [[action-celebrate-error-catching]] — publicly reward people who catch AI errors, framing it as good judgment, not insubordination. A037's warning is the corpus's darkest line: 'the biggest risk of AI... is training people to stop asking why' (see the source note).

Synthesis: A040 fears workers who *won't* trust the tool; A037/A079 fear workers who *stop thinking*. The unifying prescription — make questioning AI a rewarded, safe, visible behavior — is itself a psychological-safety intervention (see [[cross-psychological-safety-backbone]]) and is entangled with the black box, since [[concept-attribution-uncertainty]] blocks the sense-making that would recalibrate trust.