Microsoft’s AI for Good lab conducted an experiment involving more than 12,500 participants from around the world who rated 287,000 images.
Research showedthat the overall success rate of recognizing images created by artificial intelligence from real images is only 62%. This means that people have only a moderate ability to discriminate «creativity» AI, and the success rate «is only slightly higher than a coin flip», i.e., than simple chance.
Participants had the easiest time recognizing fake images of people, but faced significant difficulties when it came to natural or cityscapes, where the success rate dropped to 59-61%. As part of the study, participants had to play a «Real or Not» quiz where they were shown images created by AI that they could probably actually come across on the Internet. During the quiz, people avoided choosing unclear images or only selected misleading ones. The researchers note that artificial intelligence is constantly improving, so future models will be able to create even more convincing images.
The researchers also used a proprietary AI detection tool — it was able to achieve over 95% success in all categories. This shows that automated recognition is much more reliable than human recognition, but it is not perfect.
The authors of the study note that humans may have had an easier time detecting AI-generated face images due to the human innate ability to recognize faces well. The study found that generative adversarial networks (GANs) and the inpainting method were quite good at fooling users because they created images in the aesthetics of amateur photography, as opposed to «studio» Midjourney and DALL-E 3.
Inpainting is a technique that replaces a small element of a real image with a generated one. Microsoft has noted that this makes it extremely difficult to detect fakes and poses a significant risk of deliberate misinformation.
Based on the research, the company calls for the introduction of transparency tools such as watermarks and reliable AI detection tools to eliminate Risks of spreading disinformation. To help people become aware of these dangers, Microsoft has previously launched a campaign to combat disinformation created by AI.
Sources: Windows Central, Neowin
Spelling error report
The following text will be sent to our editors: