News

For most, deepfakes are more real and more trustworthy than authentic faces.

A study carried out by researchers at the universities of Lancaster, in the United Kingdom, and California, in the United States, showed that most people have difficulty identifying real faces from synthetic faces, generated by algorithms, also known as deepfakes.

To reach that conclusion, researchers Sophie J. Nightingale and Hany Farid conducted three experiments to determine if and how people differentiated real faces from artificially-synthesized faces.

“Our assessment of the photorealism of AI-synthesized faces indicates that synthesis engines have gone through the sinister valley and are capable of creating faces that are indistinguishable—and more reliable—than real faces,” wrote Farid and Nightingale.

In one of the experiments, 315 participants evaluated faces to determine if they were algorithmically generated or if they belonged to real people. The accuracy rate was 48.2%. In a second experiment, 219 participants received feedback on previous assumptions as they progressed through the assessment; this nomination raised the score up to 59%.

Image: PNAS

In a third and final experiment, the researchers asked 223 participants to rate faces, this time on a perceived reliability scale — quick judgments to determine whether a person is trustworthy based on face alone — which could correlate with shape. of them detect fake and authentic faces. Thus, most considered synthetic faces as more reliable.

At the PNAS newspaper, in which the study was published, the researchers say this is due to the fact that fake faces represent a composite of the average of faces. White faces, both male and female, were the most likely to be misclassified; with male faces scoring less accurately. “We hypothesized that white faces are more difficult to classify because they are overrepresented in the StyleGAN2 training dataset and therefore are more realistic,” the researchers wrote.

Machine learning-based technology first emerged in 2017. At the time, there were many attempts to detect deepfakes and efforts by private companies and government budgets created to combat them. Today the deepfakes are created for commercial use.

According to the researchers, “current techniques are not efficient or accurate enough to cope with a barrage of daily uploads,” which makes it increasingly difficult for our ability to discern the fake from the real.

Via vice

Related Articles

Leave a Reply

Your email address will not be published. Required fields are marked *