As Good As A Coin Toss: Human detection of AI-generated images, videos, audio, and audiovisual stimuli

Authors: D Cooke, A Edwards, S Barkoff, K Kelly

Published: 2024

Publication: arXiv preprint arXiv:2403.16760, 2024 - arxiv.org

People struggle to distinguish AI-generated media from authentic content, with detection accuracy averaging close to chance (50%), and performance declines in certain conditions like synthetic human faces, foreign languages, or heterogeneous media modalities.

Methods: Perceptual study evaluating participants' ability to identify authentic and synthetic images, audio, video, and audiovisual stimuli under various conditions.

Key Findings: Human ability to detect AI-generated media compared to authentic media, including factors like modality, content type, foreign languages, and participant demographics.

Limitations: Does not address technological or systemic strategies for combating synthetic media threats; focused only on perceptual capabilities of individuals.

Institution: King’s College London, Center for Strategic and International Studies

Research Area: Detection of AI-Generated Multimedia, Misinformation, Human-Computer Interaction (HCI)

Discipline: Human-Computer Interaction (HCI), Artificial Intelligence, Media Studies

Sample Size: 1276 participants

Citations: 44

DOI: https://doi.org/10.48550/arXiv.2403.16760