On July 11, 2026, psychologist Dr. Clare Sutherland of the University of Aberdeen and her Australian colleague explored the growing challenge of distinguishing authentic images from AI-generated deepfakes. Their research aims to determine if individuals can be trained to recognize when a face is created by artificial intelligence.
Understanding AI Deepfakes
As AI technology advances, the ability to create realistic images has become increasingly sophisticated. This poses a significant problem, especially in cases of fraud, where deepfakes can deceive individuals into believing they are communicating with someone they trust. According to Prof. Amy Dawel, director of the Australian National University Emotions and Faces Lab, past attempts to identify deepfakes by spotting obvious flaws, such as an extra finger, have become less effective as AI improves.
In their study, Sutherland and Dawel investigated whether training could enhance people's ability to spot AI fakes. Participants were subjected to a pool of thousands of AI-generated faces created using the StyleGAN3 image tool, known for its high realism. The results were promising, indicating that training can significantly improve recognition accuracy.
Key Characteristics of AI-Generated Faces
Participants in the study were trained to observe six perceptual qualities that often differentiate AI-generated images from real ones:
- Symmetry: AI may struggle to replicate human quirks, such as a drooping eyelid.
- Proportionality: Deepfakes may feature exaggerated facial features, like oversized noses.
- Attractiveness: AI-generated faces often appear more conventionally attractive.
- Distinctiveness: AI faces tend to be more generic and less memorable.
- Expressiveness: AI-created images typically exhibit fewer emotional expressions.
- Memorability: They are often harder to remember due to their generic nature.
These characteristics can help individuals develop a 'gut feeling' for identifying deepfakes, rather than relying on distinct markers.
Training Improves Identification Accuracy
The researchers found that participants' accuracy increased significantly after training, with scores rising from approximately 40% to 80%. Some individuals even achieved near 100% accuracy. This improvement highlights the potential for training programs to enhance public awareness and skills in recognizing AI-generated content.
Moreover, the study revealed that increased confidence in identifying deepfakes can lead to better performance. As Sutherland noted, “If you don’t know when you’re correct or not, you can’t really do anything with that information.”
Given the rising instances of deepfake-related fraud, such as a recent incident where a deepfake impersonated a boss in a video call leading to a £25 million scam, understanding how to spot these fakes is critical.
🤖 This article was rewritten by Feed and Figures' editorial AI from a report originally published by BBC News. Facts and quotes are preserved from the original; the rewrite focuses on clarity and structure. For the unedited original, see the source link below.
