The rapid advancement of artificial intelligence (AI) has made fake images so realistic that it has become more difficult than ever for ordinary people to distinguish between real and AI-generated pictures. However, a new study has found that teaching people to focus on a few specific characteristics can significantly improve their ability to identify fake images in a relatively short time.
The research was conducted by psychologist Dr. Claire Sutherland of the University of Aberdeen in Scotland and Professor Amy Dawel of the Australian National University, in collaboration with scientists from the UK, Australia, and Canada. Participants were trained to look for six key features.
Unusual Perfection:
Human faces naturally contain minor imperfections, such as a slightly drooping eyelid or an uneven smile. Experts say that if a face appears excessively flawless, it is more likely to be AI-generated.
Facial Proportions:
AI tends to create highly symmetrical and conventionally attractive faces, whereas real people naturally have subtle differences in the size and shape of features such as the nose, ears, and eyes.
Exceptional Attractiveness:
According to the study, AI often generates faces that appear unusually beautiful or attractive. Dr. Sutherland noted that AI tends to favor highly attractive facial features.
Lack of Distinctive Features:
Real people usually have unique characteristics that make them recognizable, while AI-generated faces often appear generic and similar to one another.
Limited Emotional Expression:
The study found that AI-generated faces often display weaker emotional expressions. Emotions such as happiness, anger, or sadness are generally less convincing than in real human faces.
Lack of Memorability:
Experts say AI-generated faces are often forgettable and do not leave a lasting impression, whereas real human faces are generally easier to remember.
Researchers emphasized that no single characteristic alone is definitive proof that an image is fake. Instead, identifying AI-generated images requires assessing multiple features together and forming an overall impression.
During the study, participants were first tested by viewing a series of images. They were then trained to distinguish between real and AI-generated images before taking the test again.
The results showed that participants’ accuracy improved from an average of 40 percent to nearly 80 percent, while some individuals achieved almost 100 percent accuracy.
Experts concluded that, for now, people can improve their ability to detect AI-generated images with proper training. However, as artificial intelligence continues to evolve and learn, distinguishing fake images from real ones is likely to become even more challenging in the future.
Also Read: Will AI take away jobs faster than the world can adapt?
