People are more likely to trust white faces in generated photos than in real photos - study

DALL-E 3/Bing Create.

A team of researchers from Australia, the UK and the Netherlands have found that people are more likely to mistake artificial intelligence-generated white faces for real white faces than photos of real people.

Here's What We Know

The experiment involved 124 people. They were each shown 100 real and generated images of white faces. The participants were then asked to determine how real these faces appeared to them.

The results showed that 66% of the AI images were assessed as human, while for real photos this figure was 51%. No such difference was found for faces of other races.

A second experiment with 610 participants found that the main factors that make AI faces look real are their symmetry, familiarity and less memorability.

According to the researchers, the results point to the importance of combating biases in training image generation algorithms. Such a situation could increase discrimination based on race.

Source: The Guardian