How CAPTCHA is training AI, reading ancient texts

A LESSON IN EVERY CLICK

- Taranna Khetpal

It is essentially a test meant to distinguish bots from humans, block the former and usher in the latter.

That's why it's called the Completely Automated Public Turing test to tell Computers and Humans Apart. But because of its vast reach, and stellar text recognition capabilities, CAPTCHA software has been used in a range of other ways. Take a look.

Helping read ancient texts

For decades now, libraries, publishing houses and media companies attempting to digitise texts have run into a timeless problem: that of people gathering to squint at a word and go: What does that say?!

In some cases, manuscripts have frayed; in others, old newsprint has faded.

Using CAPTCHA, it turned out that scattered multitudes could help, by metaphorically gathering around and hazarding a guess at what the errant letters might be.

Letters and words that character-recognition software could not reliably decipher were repurposed as CAPTCHA challenges, and presented in their distorted forms. As internet users around the world typed out the letter these called to mind for them, the results often produced an aha moment, confirming what archivists had guessed. At other times, the test generated user consensus strong enough to produce high-confidence transcriptions for major archives, including that of The New York Times.