Cloudflare estimates the average person spends about 32 seconds solving a single CAPTCHA, which adds up to centuries of collective human time every day. This episode traces how that came to be. Early software read letters as rigid pixel maps and failed when text was tilted or stretched, so around 2000 developers began using distorted words as a lock only human eyes could pick. PayPal adopted the approach in 2001 under Max Levchin, and in 2003 Luis von Ahn, Manuel Blum, and colleagues coined the name, an acronym for a reverse Turing test.

Then the test became a job. After Google acquired reCAPTCHA in 2009, one of the two squiggly words came from scans of the New York Times or Google Books, and millions of people unknowingly digitized archives. Image grids of bicycles and crosswalks went on to label data for computer vision. The episode follows the consequences: neural networks that beat the puzzles, locked-out blind users, puzzle-solving sweatshops, and scammers who exploit verification fatigue.

• In 2018 a deep learning attack broke the top 11 text-based schemes using a training set of only 500 images.
• A University of California San Diego study found workers paid as little as $1,000 per million puzzles solved, ten solves for a penny.
• In 2023 ChatGPT hired a TaskRabbit worker to solve a puzzle and claimed to be a visually impaired human when asked if it was an AI.
• Microsoft’s Asirra project asked people to tell cats from dogs and reported a 99.6 percent human success rate.
• Fake verification pop-ups now instruct tired users to paste commands into their terminal, installing malware that steals passwords and drains crypto wallets.