All these websites have the verification captcha to prove that I am not a robot. Why cant robots read that? It doesnt seem that difficult when all these advanced robots are being produced.
In: Technology
It’s a matter of objects recognition. It’s global problem of computer science, and by now we have very slow, time-consuming and partial solutions. Small change in images make them unrecognizable for AI but easy for humans.
Even though I know that a modern robot can literally be any computer of any size I still can’t help but picture an 80s style white robot sat in an office chair with a huge gaming pc set up whenever I see the phrase “I am not a robot”
It’s feature recognition of pixel encoding. Pixels make different rectangular Tetris shapes called HAR-like features. These shapes and features have different visual filters attached them. Similar to filters people put on Instagram filters. Eventually patterns are recognized from these Tetris shapes. If enough of these shapes are recognized by training a recursive neural network, eventually finding enough features will be considered detection. These captchas take pictures with not enough HAR-like features to allow a computer to perform object recognition. A human will still be able to perform a classification.
There’s a pretty good chance that a bot can read the CAPTCHA, but CAPTCHAs are more sophisticated than that now. The most advanced ones like [reCAPTCHA v3](https://developers.google.com/recaptcha/docs/v3) don’t just rely on what your choices are on what parts of the picture have traffic light. While the details are obviously a secret, they look at a number of factors based on your behavior on the website such as what pages you’re looking at, how long you spend on them, how quickly you can click the traffic lights, browsing history, etc. (these examples of course are speculative). Then it uses all that information to try to determine whether you’re acting like a human.
For example, a bot might navigate to a website about dogs from a site that’s completely unrelated, have a suspicious IP address, select a link at the bottom of a long page really quickly, and select the traffic lights in the CAPTCHA faster than a human could. The website knows what the typical behavior of legitimate traffic is. For example, maybe a most of the visitors to the website are from Google, Wikipedia, and other major dog related websites, most have IP addresses in English speaking countries, and most people spend 15-45 seconds reading the page before clicking that link on the bottom.
The CAPTCHA is probably looking at all the data it can and looking for anomalous behavior with a machine learning algorithm, sort of like how banks monitor credit card activity and flags it when it sees unusual behavior. Your answer to which parts of the image have a traffic light is only part of the process.
Latest Answers