The growing trend of AI-generated videos and deepfakes has led to the spread of misinformation and a growing lack of trust in online content. Methods are being devised to detect deepfakes and distinguish between real and AI videos.
Now, a study conducted at the University of Reading has figured out a five-minute training that can enable a person to easily identify faces rendered by AI. The study and its findings were revealed by studyfinds.org.
Trick to identify AI-generated pictures
It included 664 participants who are working for security agencies in a capacity that requires them to study videos. They were shown faces created by StyleGAN3, regarded as the most advanced face-generation tool. Mixed along with these pictures were photographs of real people.
The article in Study Finds says that without training, the accuracy in identifying fake faces was just 42 percent. Then, they were given the training to detect AI-generated faces. The training was essentially informing these people about the minute flaws that AI leaves while rendering faces. Some of these are “oddly rendered hair, unusual numbers of teeth, or strange transitions where facial features meet the background.”
The participants then went through 10 practice sessions to judge their identifying powers. Following this, the accuracy of the participants improved substantially. Those in the group who were already better at recognizing fakes, increased their accuracy from 54 to 64 percent. Even those who were not part of this group became more adept at identifying AI-generated faces.
So, a bit of training, focusing on identifying subtle shortcomings in AI renderings, can go a long way in improving the ability to distinguish fakes from real pictures. However, with AI getting better, this advantage may not last long.
Other ways to detect AI
As per BBC, there are other clues as well in an AI-generated image that you can use to identify if it's fake.
The AI images usually get small details wrong. So, a closer analysis of the picture, zooming in on the small objects and components of the picture, can help you detect giveaways.
While tiny flaws can reveal a fake, sometimes, the exact opposite could give away the reality. Yes, if a picture looks too smooth, too picturesque, to the point where such perfection is almost too good to be true, you might be looking at a fake. Do pay attention to the background in such pictures.
Lastly, the simple method of reverse search is very useful to find out if an image is real or fake. Just take a picture and put it through a search engine. The original source of the photo will reveal whether it is fake or real.
Note to readers: This article is for informational purposes only and not a substitute for professional medical advice. Always seek the advice of your doctor with any questions about a medical condition.