Captcha technology has been with us for more than a decade when we interact on the web. This method makes it possible to avoid bots and, incidentally, train text or image recognition systems. Now a new and unique version has appeared.
What is a Yoko? In Vice they tell how recently people who wanted to use Discord ran into a challenge. A new Captcha asked them to identify images that contained a “Yoko”. When showing the reference image, an object appeared that was a kind of mix between a yo-yo and a snail. If that object does not sound familiar to you, it is normal. Does not exist.
Created by AI. It turns out that the hCaptcha system used in these “authentication puzzles” – which we already knew – had used an object created by a generative artificial intelligence model. The uniqueness is not only that: the rest of the images that appeared in the matrix of this hCaptcha had also been generated by AI, so in essence we were faced with a set of images of objects that did not really exist.
Discord explains itself. In statements to that medium, those responsible for Discord explained that “the technology that generates these messages is owned by our external partner and Discord does not directly determine what is presented to users.” The hCaptcha system generates hundreds of millions of users, and they acknowledged that the appearance of the object called “Yoko” had been part of a “brief test that was seen by a small number of users”, but the strangeness of that object caused some to share that experience on networks like Twitter.
How hCaptcha works. The hCaptcha system was born in 2017, but it has evolved over the years and, according to its creators, it is now fully focused on offering a security platform. Their requests, they say, are generated for clients seeking high-quality responses from human users “for their machine learning needs.”
A Captcha that trains AI. All this means that the captchas that are displayed help train machine learning systems and antagonistic generative networks. The appearance of these types of AI-generated image matrices—some of them of non-existent objects—is expected to occur with increasing frequency.
Reviews. However, criticism of this type of image is numerous among users who simply have not been able to “get it right” with the system because the AI itself must wait for a response that human beings do not provide. A Reddit user complained about how in a Captcha that asked us to identify a robot, it was not clear which of those images represented it (as you can see in the image.
Image | Xataka with Midjourney
In Xataka | The Turing test no longer works: seven alternative tests to verify that an AI is not a human
