Paper Type

Complete

Abstract

Advances in generative AI are producing images that are increasingly indistinguishable from human photography, challenging how users evaluate visual authenticity in digital environments. This study examines how individuals distinguish AI-generated from human-produced images and which cognitive processes guide these judgments when perceptual cues become unreliable. Using a controlled experiment with 130 participants, we analyze detection performance through mixed-effects modeling and Signal Detection Theory. Results suggest a perceptual tipping point: participants successfully detect AI images when visual artifacts are present, but discrimination collapses for highly realistic AI imagery (d′ ≈ 0). This failure is accompanied by a strong shift in response bias toward assuming human authorship. The findings support a two-stage heuristic model of authenticity judgment: artifact detection followed by a human-default heuristic when cues disappear. By linking sociotechnical boundary theory with cognitive credibility research, the study shows how users cognitively reconstruct the human-AI boundary under perceptual uncertainty.

Paper Number

1924

Comments

SIG HCI

Download

Share

COinS
 
Aug 15th, 12:00 AM

When Seeing Is No Longer Believing: A Two-Stage Heuristic Model of Authenticity Judgment Under Generative AI

Advances in generative AI are producing images that are increasingly indistinguishable from human photography, challenging how users evaluate visual authenticity in digital environments. This study examines how individuals distinguish AI-generated from human-produced images and which cognitive processes guide these judgments when perceptual cues become unreliable. Using a controlled experiment with 130 participants, we analyze detection performance through mixed-effects modeling and Signal Detection Theory. Results suggest a perceptual tipping point: participants successfully detect AI images when visual artifacts are present, but discrimination collapses for highly realistic AI imagery (d′ ≈ 0). This failure is accompanied by a strong shift in response bias toward assuming human authorship. The findings support a two-stage heuristic model of authenticity judgment: artifact detection followed by a human-default heuristic when cues disappear. By linking sociotechnical boundary theory with cognitive credibility research, the study shows how users cognitively reconstruct the human-AI boundary under perceptual uncertainty.

When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.