When AI and Humans Produce Partial Truths: Examining Acceptability of Perceived Error and Perceived Associated Harms

Area of Science:

• Information Science
• Human-Computer Interaction
• Communication Studies

Background:

• Misinformation research often overlooks messages containing both true and false elements.
• Real-world information frequently blends accurate and inaccurate content, necessitating research with greater external validity.

Purpose of the Study:

• To investigate user perceptions of acceptable error in mixed-information messages.
• To examine how source type (human-only, AI-only, or combined) influences these perceptions.
• To understand the interplay between acceptable error, perceived harms, and engagement intentions.

Main Methods:

• Experimental design to assess reactions to messages with varying truthfulness and source types.
• Utilized a framework of uncertainty attributes of truth claims.
• Measured acceptability of perceived error, associated harms, and intentions to engage or intervene.

Main Results:

• Source type significantly impacts the acceptability of errors in mixed-information messages.
• Perceived error and harms interact to predict engagement and intervention intentions.
• Generative artificial intelligence (AI) sources may be perceived differently regarding error tolerance compared to human sources.

Conclusions:

• Understanding perceptions of error in mixed-information is crucial for combating its spread.
• Source attribution (human vs. AI) is a key factor in evaluating information credibility and acceptability.
• Interventions targeting misinformation should consider the nuanced nature of content and its perceived origin.