reCAPTCHA Trains AI

A story about how reCAPTCHA evolved from digitizing books to training AI, as explained by Shaan Puri.

"The original CAPTCHA was created by a professor at Duke and Carnegie Mellon. It started with those hard-to-read letters you had to type. They were actually using this to digitize book pages. The computer wasn't good enough at translating scanned book pages to text, especially when pages were rounded, wrinkled, or fuzzy. But human eyes could easily read it.

They solved two problems at once: websites could verify users weren't bots, and they managed to transcribe millions of books in the process.

Then CAPTCHA 2 (reCAPTCHA) evolved to focus on image classification. When you click squares with traffic lights or buses, you're actually helping train self-driving cars to identify objects.

The inventor, Louis von Ahn, went on to create Duolingo with a similar dual-purpose approach. When people use Duolingo to learn languages, they're actually translating content that third-party services pay for. He basically found ways to get humans to do useful work while completing everyday tasks."