A short history of captchas

Captchas have a short public history. The pieces are a search engine, a university project, a book-scanning trick, and then a shift away from text.

1997, before the name

Around 1997, AltaVista wanted to stop programs from submitting URLs into its search index. Andrei Broder and colleagues asked the visitor to read distorted characters and type them. The server knew the string it had drawn. A replay of yesterday's string failed, because the picture was new. Nobody called this a CAPTCHA yet. It is the pattern every later text captcha copies: draw, hide the answer on the server, check what comes back.

2000, the name

In 2000, Luis von Ahn, Manuel Blum, Nicholas Hopper, and John Langford at Carnegie Mellon University gave the test its name and a definition. A CAPTCHA had to be generated and graded by a computer, and it had to be a standing problem for computer vision or a related field. Yahoo used an early version so scripts would stop opening mail accounts. A 2003 paper, "CAPTCHA: Using Hard AI Problems for Security," wrote the idea down for the research community.

2007, two words

In 2007 von Ahn and colleagues launched reCAPTCHA. The visitor typed two words. One word was the security check, with a known answer. The other came from a scanned book the computer could not read. Human effort on the login form was reused to digitize print. Google acquired reCAPTCHA in 2009. A later version of the same idea asked people to read house numbers from Street View photographs.

After text became easy to read

As models learned to read warped letters, text stopped being a hard problem in the sense of that 2003 paper. In 2014 reCAPTCHA added the "I am not a robot" checkbox, and showed a photo grid only when it was unsure. In 2018 a later version could return a risk score without a puzzle on every request. Other vendors offered their own photo grids. Many government and civic portals kept the older picture of digits or a sum, because it is easy to draw and easy to check.

That split is the one that matters for this API. Image captchas are still pictures. Checkbox captchas are a different system. The evolution note is the same story told as a change in what "hard for a computer" means.