New research is examining whether humans can find integrated humans – the need to develop preventive technologies before it is too late.
Until recently, if we asked you if you could distinguish between a real face and a non-existent synthesized face, you would say that you would definitely succeed. After all, humans can recognize humans, can’t they? Thus, a new study published shows that we are now moving towards speculation – we are not really excelling at it.
You better throw a coin
Hundreds of people who took part in the experiment were asked to look at pictures of their faces individually and determine if it was a real person’s picture or a picture of a person integrated with GAN technology, or their full name, the Generative Advertising Network. . These are networks that rely on data containing millions of different items to create a new item in the template. For example, millions of face paintings can enter such a system, which will create a completely new face based on the structure of the face, eye color, and the shape of the mouth learned from existing images.
In the first experiment, 315 participants were asked to sort 128 random images from a database of 800 half-original and half-fabricated cover art. The average identification was only about 48% on average – which means that the next time you discuss whether an image is real or not, you have a better chance of winning a coin toss. In this experiment, the researchers found that the most difficult faces to identify were those of white men and women, who believe that this is due to the fact that GAN networks are increasingly represented in trained databases. To give them more realistic faces.
In the second experiment, 219 new participants were asked to craft the same classification, this time they were trained and received feedback after each selection, which allowed them to improve the results to approximately 59% correct identification. However, researchers say there has been no improvement in detection capabilities over time. In the third experiment, the researchers decided to experiment with something completely different, and rated the same person’s credibility rating as shown in Figures 1 to 7 (the latter is definitely a higher rating). The average rating of the real face was 4.48 while that of the synthetic face was 4.82. Although relatively small (approximately 7.7 percent), researchers define this difference as “significant.”
Although the researchers tried to test it, the researchers found no correlation between the degree of reliability and the type of face presented, although it was noted that women were more reliable than men. The researchers also concluded that a smiling face increases your chances of being trusted. In the conclusion of the experiment, the researchers argued that developers of related technologies should consider whether the advantages outweigh the disadvantages, while at the same time building protection systems against synthetic media and the damages involved. May we all have success.
The full study is available to read here. Here are the results of the challenge at the beginning of the article (original R, synthesized S, along with identification percentages). By the way, if you want to try too – you can always try on the site This.
Tags for article:
Problem solver. Incurable bacon specialist. Falls down a lot. Coffee maven. Communicator.