Artificial intelligence can now create fake human faces that look almost identical to real people.
These realistic images, often called deepfakes, are becoming increasingly common online and are being used in scams, identity fraud, and misinformation.
However, a new study from the Australian National University (ANU) suggests that people can learn to recognize these fake faces with surprisingly short training sessions.
The research, published in the journal PNAS, found that even brief training significantly improved people’s ability to tell the difference between real human faces and AI-generated ones.
According to the researchers, many current tips for spotting fake images focus on looking for obvious mistakes, such as extra fingers, strange earrings, or distorted backgrounds.
While these clues were useful when AI image generators were less advanced, today’s technology has become much better.
Modern AI can often produce images with very few visible errors, making these old methods less reliable.
Instead of searching for obvious flaws, the ANU team developed a new training method that teaches people to notice more subtle features that often appear in AI-generated faces.
The training focuses on six important facial qualities: distinctiveness, memorability, proportionality, symmetry, attractiveness, and expressiveness.
Researchers found that AI-generated faces often appear unusually symmetrical, perfectly proportioned, and highly attractive.
Surprisingly, many people naturally assume these qualities mean a face is real. In reality, these characteristics can actually be signs that the image was created by artificial intelligence.
After completing the training, participants became much better at identifying AI-generated faces. Some of the best-performing participants were able to detect fake faces with almost perfect accuracy.
The researchers were encouraged by how quickly people improved.
The training sessions were led by ANU Honors student Tanya George, who said that even relatively short lessons produced noticeable improvements. She explained that AI image technology is advancing rapidly, and many people do not realize just how convincing fake faces have become. Learning to recognize them could help people stay safer when using social media, online dating platforms, business websites, and other digital services.
To make sure the findings were reliable, another research team at the University of Victoria in Canada repeated the study with a different group of participants. They found almost identical results, showing that the training works in different countries and with different groups of people.
Because the lessons can be delivered online, the researchers believe the program could be offered to large numbers of people at very little cost.
The researchers also believe that human judgment will remain important even as AI detection software continues to improve. Computer programs designed to detect deepfakes are not always accurate, and their decision-making processes can be difficult to understand. Training people to recognize AI-generated faces provides a transparent and practical way to reduce the risk of fraud.
The team is now working to make the training even shorter and to find out how long the benefits last. They also plan to test whether the same techniques work on newer types of AI-generated faces as image-generation technology continues to evolve.
