Related Experiment Video
Updated: Jul 1, 2026
06:37
Artificial Intelligence-Based System for Detecting Attention Levels in Students
Published on: December 15, 2023
Training humans to detect AI-generated faces
Amy Dawel1, Tanya George1, Eric Y Mah2
1School of Medicine and Psychology, The Australian National University, Canberra, ACT 2600, Australia.
Summary
Training people to recognize subtle global impressions of AI faces significantly improved deepfake detection accuracy. This new method offers a more robust defense against AI-generated content than artifact-based detection.
Area of Science:
- Computer Vision
- Cognitive Psychology
- Information Security
Background:
- Deepfake technology generates realistic AI faces, threatening information integrity.
- Current deepfake detection methods (algorithms, human artifact training) have limitations.
- AI and human faces elicit distinct perceptual impressions.
Purpose of the Study:
- To develop and evaluate a novel deepfake detection method.
- To train participants to identify AI faces based on global perceptual impressions.
- To assess the effectiveness and scalability of this training approach.
Main Methods:
- Participants were trained to focus on global facial impressions distinguishing AI from human faces.
- A pre-post design with untrained test faces measured detection accuracy improvement.
- Test-retest and online replication studies validated the findings.
Main Results:
- Participant accuracy in detecting AI faces nearly doubled after training.
- High performers achieved near-perfect detection rates.
- Training enhanced participants' metacognitive insight and confidence calibration.
Conclusions:
- Training individuals to perceive global features offers a promising, durable deepfake detection strategy.
- This approach addresses inherent biases in generative AI, unlike artifact-specific methods.
- The method is scalable and effective, providing a new defense against AI-generated disinformation.
