Jove
Visualize
Contact Us
JoVE
x logofacebook logolinkedin logoyoutube logo
ABOUT JoVE
OverviewLeadershipBlogJoVE Help Center
AUTHORS
Publishing ProcessEditorial BoardScope & PoliciesPeer ReviewFAQSubmit
LIBRARIANS
TestimonialsSubscriptionsAccessResourcesLibrary Advisory BoardFAQ
RESEARCH
JoVE JournalMethods CollectionsJoVE Encyclopedia of ExperimentsArchive
EDUCATION
JoVE CoreJoVE BusinessJoVE Science EducationJoVE Lab ManualFaculty Resource CenterFaculty Site
Terms & Conditions of Use
Privacy Policy
Policies

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by
Same author

Classifying Phonotrauma Severity from Vocal Fold Images with Soft Ordinal Regression.

Proceedings of machine learning research·2026
Same author

Using smartphone surveys to predict next-week suicide attempts.

Journal of psychopathology and clinical science·2026
Same author

Pretending Not to Know Reveals a Capacity for Model-Based Self-Simulation.

Psychological science·2026
Same author

Part-whole effects in visual number estimation.

Attention, perception & psychophysics·2026
Same author

"Core Perception": Re-imagining Precocious Reasoning as Sophisticated Perceiving.

The Behavioral and brain sciences·2025
Same author

Event-based warping: A relative distortion of time within events.

Journal of experimental psychology. General·2025

Related Experiment Video

Updated: Oct 8, 2025

Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention
06:37

Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention

Published on: December 15, 2023

4.4K

Deepfake detection by human crowds, machines, and machine-informed crowds.

Matthew Groh1, Ziv Epstein2, Chaz Firestone3

  • 1Media Lab, Massachusetts Institute of Technology, Cambridge, MA 02139; groh@mit.edu.

Proceedings of the National Academy of Sciences of the United States of America
|December 31, 2021
PubMed
Summary

Humans and AI are equally good at spotting deepfakes, but make different errors. Combining human and AI detection improves accuracy, yet flawed AI predictions can mislead humans.

Keywords:
artificial intelligencefacial recognitionforensic sciencemisinformationwisdom of crowds

More Related Videos

A Step-by-Step Implementation of DeepBehavior, Deep Learning Toolbox for Automated Behavior Analysis
05:41

A Step-by-Step Implementation of DeepBehavior, Deep Learning Toolbox for Automated Behavior Analysis

Published on: February 6, 2020

9.5K
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
03:31

Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications

Published on: December 15, 2023

676

Related Experiment Videos

Last Updated: Oct 8, 2025

Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention
06:37

Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention

Published on: December 15, 2023

4.4K
A Step-by-Step Implementation of DeepBehavior, Deep Learning Toolbox for Automated Behavior Analysis
05:41

A Step-by-Step Implementation of DeepBehavior, Deep Learning Toolbox for Automated Behavior Analysis

Published on: February 6, 2020

9.5K
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
03:31

Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications

Published on: December 15, 2023

676

Area of Science:

  • Cognitive Science
  • Computer Vision
  • Media Studies

Background:

  • The proliferation of sophisticated machine-manipulated media, or deepfakes, presents a significant societal challenge in discerning authentic video content.
  • Current deepfake detection methods rely on either human perceptual abilities or automated computer vision models.

Purpose of the Study:

  • To compare the deepfake detection performance of human observers against a leading computer vision model.
  • To investigate the synergistic effects and potential drawbacks of combining human and machine detection.
  • To identify the distinct strengths and weaknesses of humans and machines in identifying deepfakes.

Main Methods:

  • Two online studies involving 15,016 participants who were presented with authentic videos and deepfakes.
  • Direct comparison of human accuracy with a state-of-the-art deepfake detection model.
  • Analysis of performance variations based on video features and the impact of randomized interventions.
  • Examination of how model predictions influence human judgment and overall accuracy.

Main Results:

  • Human observers and the leading computer vision model demonstrated comparable accuracy in deepfake detection, albeit with differing error patterns.
  • Collaborative detection, where humans utilized model predictions, generally enhanced accuracy compared to individual methods.
  • Inaccurate model predictions sometimes reduced human participants' detection accuracy.
  • Visual manipulations targeting facial processing impaired human performance significantly more than the machine model's performance.

Conclusions:

  • Humans and AI possess complementary, yet distinct, capabilities in deepfake detection.
  • Human reliance on specialized cognitive capacities, particularly for face processing, influences their vulnerability to certain deepfake manipulations.
  • Further research is needed to optimize human-AI collaboration for robust deepfake detection in real-world scenarios.