Only 40% get it right: Microsoft's test that tests whether you can distinguish a real photo from one made with AI

Everyone knows perfectly well that artificial intelligence has already become a part of our lives since ChatGPT was launched globally at the end of 2022. Many users use this technology to program code, correct texts, plan vacations, search for information, promote study or create realistic images, however, in line with the latter mentioned, the results can generate confusion among Internet users because sometimes they question what is real and what is created by an AI.

To clear up the mystery, some apps like TikTok, Instagram, Facebook, and YouTube have taken the initiative to enable a label to identify content generated by artificial intelligence. Additionally, there are tools like 'AI or Not' that identify whether an image was created by said technology . However, until now, there were few tests to detect AI-generated photographs, so a group of Microsoft researchers has created a 'game' that allows you to view 15 images to determine which ones are fake and which ones are real.

As you can see in the image below, 'Real or Not' displays a series of photographs in a carousel-like fashion. It also includes two buttons to indicate whether the photos are real or fake . Once the 15 photographs have been analyzed, the tool provides a percentage of the number of correct images, taking into account the number of correct matches as the photos are ranked.

Test results surprise Microsoft

Surprisingly, an analysis by Microsoft researchers reveals that "analysis of approximately 287,000 image evaluations by more than 12,500 participants worldwide revealed an overall success rate of just 62% , indicating modest ability slightly above chance."

Those from Redmond point out that participants identified human portraits with high accuracy , but had significant difficulty with natural and urban landscapes. Therefore, Microsoft warns that "these results highlight the inherent challenge humans face in distinguishing AI-generated visual content, particularly images without obvious artifacts or stylistic cues."

The analysis also emphasizes the need to add more transparency tools , such as watermarks, and robust AI detection tools to reduce the risks of misinformation stemming from AI-generated content.

Sign up for our newsletter and get the latest technology news straight to your inbox.