Saved in:
Bibliographic Details
Main Authors: Postiglione, Marco, Gortner, Isabel, Subrahmanian, V. S.
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2603.14658
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911518524702720
author Postiglione, Marco
Gortner, Isabel
Subrahmanian, V. S.
author_facet Postiglione, Marco
Gortner, Isabel
Subrahmanian, V. S.
contents Deepfake detection is widely framed as a machine learning problem, yet how humans and AI detectors compare under realistic conditions remains poorly understood. We evaluate 200 human participants and 95 state-of-the-art AI detectors across two datasets: DF40, a standard benchmark, and CharadesDF, a novel dataset of videos of everyday activities. CharadesDF was recorded using mobile phones leading to low/moderate quality videos compared to the more professionally captured DF40. Humans outperform AI detectors on both datasets, with the gap widening in the case of CharadesDF where AI accuracy collapses to near chance (0.537) while humans maintain robust performance (0.784). Human and AI errors are complementary: humans miss high-quality deepfakes while AI detectors flag authentic videos as fake, and hybrid human-AI ensembles reduce high-confidence errors. These findings suggest that effective real-world deepfake detection, especially in non-professionally produced videos, requires human-AI collaboration rather than AI algorithms alone.
format Preprint
id arxiv_https___arxiv_org_abs_2603_14658
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Human-AI Ensembles Improve Deepfake Detection in Low-to-Medium Quality Videos
Postiglione, Marco
Gortner, Isabel
Subrahmanian, V. S.
Computer Vision and Pattern Recognition
Artificial Intelligence
Deepfake detection is widely framed as a machine learning problem, yet how humans and AI detectors compare under realistic conditions remains poorly understood. We evaluate 200 human participants and 95 state-of-the-art AI detectors across two datasets: DF40, a standard benchmark, and CharadesDF, a novel dataset of videos of everyday activities. CharadesDF was recorded using mobile phones leading to low/moderate quality videos compared to the more professionally captured DF40. Humans outperform AI detectors on both datasets, with the gap widening in the case of CharadesDF where AI accuracy collapses to near chance (0.537) while humans maintain robust performance (0.784). Human and AI errors are complementary: humans miss high-quality deepfakes while AI detectors flag authentic videos as fake, and hybrid human-AI ensembles reduce high-confidence errors. These findings suggest that effective real-world deepfake detection, especially in non-professionally produced videos, requires human-AI collaboration rather than AI algorithms alone.
title Human-AI Ensembles Improve Deepfake Detection in Low-to-Medium Quality Videos
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2603.14658