How captchas evolved

A captcha is tied to the abilities of programs at the time it is drawn. When those abilities move, the test moves, or it stays and becomes easier to read.

The first generation bet on vision. Distorted letters, noise lines, and odd fonts were hard for the optical character recognition of the early 2000s and ordinary for a person. Distorted text is that generation, and it is still what many portals draw.

The second generation kept the human in the loop and spent the answer on a second job. reCAPTCHA's two words, from 2007, checked one known word and collected a reading of a scanned word. Later, people transcribed house numbers from street photos. The visitor still typed text. The vendor also got data.

The third generation moved off a puzzle the person always sees. A checkbox, a photo grid, or no visible test at all sits on signals from the browser: how the page loaded, how the pointer moved, what the network looks like. The vendor returns a pass or a score. There may be no image, and there may be no answer string. Checkbox and invisible captchas are this generation.

A fourth pattern never left the first one. Civic and government sites often still paint digits, letters, or a one-line sum. The software is small, the check is an equality test, and the picture is easy to explain. Those images got more noise as readers improved. They did not turn into a risk score. Why image captchas remain is about that choice.

Captcha OCR belongs to the image generation. It reads text, numeric codes, arithmetic, and Hindi-number sums. It does not replace a checkbox vendor, and it does not listen to audio captchas. When you look at a portal, name the generation first. The options you send depend on it.