People can learn to spot AI faces – but the clues are no longer obvious

Deepfake faces generated via AI are increasingly realistic and contribute to a projected US$40 billion worth of fraud by 2027.
Incorrectly identifying AI faces is prevalent; even those overconfident about detecting AI are often wrong.
Software-based detectors for deepfakes exist but have flaws, including being easily fooled by simple image format changes.
With sufficient practice and exposure, individuals can improve their ability to identify AI-generated faces.
The key difference identified between AI faces and human faces is that AI faces tend to appear hyperreal, more symmetrical, and attractive but less expressive and memorable.
Training people with exposure to AI face characteristics without direct instruction effectively enhances their detection skills.
Results from both Australian and Canadian studies indicate that the training method is reliable and effective both in-person and online.
Although promising, broader verification of the training’s effectiveness across other AI models, ages, and longer periods is still needed.
To improve personal capability to spot AI faces, viewing numerous examples and focusing on key factors like distinctiveness, memorability, proportionality, symmetry, attractiveness, and expressiveness is suggested.