>>4668964You're imagining an AI as receiving no help from human beings. This is how you'd want "strong AI" to work, where you'd lead the robot into a room, sit it down in front of the test, and say "figure it out". But the type of AI that CAPTCHA is meant to guard against isn't that type of AI, because it doesn't have to be.
Those tricks are irrelevant because they can all be programmed around ahead of time. Each letter of the CAPTCHA appears independently? No sweat, the program solves for each letter independently. There are wrong answers that flash up quickly? No sweat, the program ignores answers that show up for less time than it calculates a human eye would need to detect an answer. From an AI perspective, the only difficult thing in the scenario you're describing is still "determine if this collection of pixels on screen is a letter, and if so which one". Everything else is just busywork that can be solved without even using a NN.
