Researchers from Indiana University developed a training course that enabled participants to double their accuracy in recognizing AI images, with some participants achieving 100% accuracy. The study was published in the journal PNAS.
Neural networks are improving weekly, to the point where it is sometimes difficult to distinguish between real online content and machine-generated content. An expert from the Moscow Institute of Physics and Technology assessed a recent statement by American scientists that humans can be trained to recognize deepfake videos.
The basis of this method is not to look for technical flaws in the content, but rather to evaluate the overall impression of the image.
Forty-five volunteers participated in the training course. After the training, their accuracy in distinguishing between fake and real images nearly doubled, with the best participants achieving 100% accuracy. Repeated tests showed that this result was indeed related to the training, not to accumulated experience.
But Alexander Beznosikov, director of the Center for AI-Powered Systems at the Moscow Institute of Physics and Technology, cautions against considering this method a ready-made solution. The experiment used only one model, StyleGAN3, which is no longer up-to-date. Therefore, the participants may have learned to observe the characteristics of this specific generator, rather than to recognize any fake image.
He noted, "If a new model emerges tomorrow, it will be necessary to readjust this skill."
Furthermore, the researchers did not determine how long this effect lasts. This method may be useful for journalists, fact-checkers, and content moderators, but its practical value for ordinary users remains debatable.
The expert continued: "As for practical importance, this skill may not be of great use to the general public. Often today, not only are faces generated, but entire videos with the surrounding environment are created, and no studies have been conducted on them, so saying that this skill might be useful is debatable."
The expert stressed that, in general, this research should be treated as proof of principle, not a ready-made solution.
what is your opinion on this news ?do let us know in the comments on tha comment box
if you liked this news article please share on friends & family & don't forget to follow on website & social media
