---
id: "claim-human-over-trust-ai"
type: "claim"
source_timestamps: ["§ Behavioral Change"]
tags: ["human-computer-interaction", "quality-assurance"]
related: ["concept-gen-ai-hallucinations", "entity-mit"]
confidence: "high"
testable: true
speakers: ["Tom Davenport", "John J. Sviokla"]
sources: ["spine"]
sourceVaultSlug: "hbr-seg-spine"
originDay: 1
articleStem: "hbr-cl-95-6-disciplines-genai"
sourceUrl: "https://hbr.org/2024/07/the-6-disciplines-companies-need-to-get-the-most-out-of-gen-ai"
sourceTitle: "The 6 Disciplines Companies Need to Get the Most Out of Gen AI"
---
# Knowledge workers default to trusting AI outputs without editing

**Claim:** Despite the known risk of hallucinations, humans do not naturally incline toward reviewing AI output. **An [[entity-mit-d1|MIT]] study found that 68% of participants chose not to edit the output of a language model** during a knowledge-work creation task.

**Confidence: high · Testable: yes.** This is the empirical justification for treating output review as a *discipline* rather than assuming it happens — see [[concept-gen-ai-hallucinations]] and [[concept-behavioral-change-gen-ai]].

Enrichment validation (directionally correct, details need care): there is strong evidence users over-trust LLM outputs and accept them with minimal editing. Adjacent evidence: Noy & Zhang (Stanford), Mollick & Mollick (Wharton), and HCI research on **automation bias**. **However**, the *specific* "68%" figure and its "MIT" attribution may not be reported in exactly that form in public summaries — treat the number as an **approximate representation** rather than a fully verified statistic pending the precise paper.

**Counter-perspective:** in high-stakes expert domains (e.g., physicians), studies show *under-trust* / cautious use — professionals heavily review and sometimes ignore AI suggestions. The "no editing" default may be stronger in general office tasks than in high-stakes work (algorithm aversion vs. automation bias).