A new study reveals that humans would hardly identify images generated by artificial intelligence.
Between the absence of limits of certain image generators such as Grok and their direct accessibility from our smartphones, identifying an image generated by an AI becomes more and more difficult. A recent study conducted by researchers from Microsoft AI for Good Labrelayed by Microsoft Centralreveals that the capacity to identify images by humans is “ barely higher than that of a battery or face game ”.
A rapid evolution
To serve as a basis for this study, researchers used the quiz “real or not?” »Developed by Microsoft, accessible online. With 287,000 image assessments took place and more than 12,500 people, the researchers’ conclusions are final. Participants would only getan overall success rate of 62 %.
The researchers point the finger at the rapid evolution of the generative AI “Displaying even more realistic results”adding that their results “Probably overestimate the ability of people today to distinguish the images generated by the AI of real images”.
Targeted identification
Study participants would identify human faces 65 %. This result drops to 59 % for photos of nature. Our brain would be programmed to easily recognize human faces, which would explain these results, indicates Microsoft Central.
The study specifies that AI detection tools are already more reliable than humans to identify IA images. For example, researchers say they have developed their own AI detector, from which The success rate would exceed 95 %.
Always accessible online, the quiz “Real or not?” »Allows users to test their identification capacity.
