They can. Depending on the type of captcha depends how easy it is for a computer to solve. Ones involving character recognition computers have become reasonably good at.
The current style used by Google which is selecting all pictures of a certain type is substantially easier for humans than computers.
Humans can fairly easily tell the difference between a set of stairs and a zebra crossing, or a train and a bus. That’s quite a hard task for a computer. Humans are good at using contextual information. I.e. There are train tracks thus the big thing sitting on it is a train 🚆. Buses are not too dissimilar visually, just compare these emotes 🚍
Modern Captchas take many forms. The most common kind tracks the movement of your mouse to check irregularities. For example, if a bot moves the mouse to click on an image, it will likely move in a straight line and stop instantly on the image. A human doing the same thing will have imperfections (slower, less straight line, course correction, etc.)
Older captchas only cared about the result (can you successfully type what is shown on screen or can you successfully identify which pictures are of cars) as computers were bad at those problems when captchas were first invented. Now, with machine learning, they’re much better than humans at those problems.
Captchas can still be beaten fairly simply, but it requires a little more effort on the programmer to do so.
Captchas are less about definitively proving you are a human and more about blocking the most obvious bots while getting you to do free labor.
Many of these captchas have you train a model for computer vision, seemingly for self-driving cars: identify the traffic lights, cars, bicycles, etc.
What I don’t understand is how it knows the right answer already … and how we can tell it when it is incorrect. One time a captcha asked me to identify taxis. It included an image of a yellow pickup truck with some black details. It was not a taxi! But I couldn’t move on until I selected it.
Latest Answers