IllusionCAPTCHA, developed by Yuekang Li at the University of New South Wales, employs generative AI to create unique optical illusions from abstract prompts. Unlike traditional CAPTCHAs that often frustrate users, this innovative approach leverages the distinct ways humans process images, making it easier for people but difficult for AI. In tests, human participants identified hidden text in images 83% of the time, with performance improving to 88% using illusory images, while AI systems like ChatGPT failed to decode any of the IllusionCAPTCHA tests.
Using AI-generated optical illusions, this new sneaky test is (ironically) difficult for AI to decipher, leveraging the unique way that humans process images.
Created by Yuekang Li and the team at the University of New South Wales in Sydney, Australia, IllusionCAPTCHA uses generative AI to create illusion-like images from abstract prompts.
The study explains that the 10 people who took the test successfully identified hidden text illusions 83% of the time, rising to 88% when illusory images were used.
When ChatGPT and Gemini were tasked with deciphering the images, they failed the IllusionCAPTCHA test every time.
Collection
[
|
...
]