Are you a robot?

You've clicked it a thousand times: a little box that says "I'm not a robot," or a grid of blurry photos asking you to find the traffic lights. It's mildly annoying, faintly absurd — and one of the most quietly clever systems on the internet. This course is about what's really going on, and it ends somewhere surprising: the puzzles you solve have been training the very AI that can now solve them too.

Start with the basic problem. A huge fraction of all web traffic isn't people — it's automated scripts (bots). Most are harmless crawlers, but many are not, and a website needs a fast way to tell who's a person.

Scroll to watch the doorman sort an incoming crowd.

What the bots are after

Why go to all this trouble? Because at scale, bots do real damage:

• Spam — posting junk comments, messages and reviews by the million.
• Fake accounts — signing up thousands of throwaway identities.
• Scraping — vacuuming up content, prices or personal data en masse.
• Scalping — buying out concert tickets or sneakers in seconds to resell.
• Credential stuffing — trying leaked passwords across sites automatically.

A human doing any of these one at a time is a nuisance; a script doing them a million times a minute is a catastrophe. The defense is a test that's cheap for a person but expensive (or impossible) for a script.

CAPTCHA: a reverse Turing test

That's exactly what a CAPTCHA is — the name is a tongue-in-cheek acronym: Completely Automated Public Turing test to tell Computers and Humans Apart.

Where we're headed

So we need a puzzle in the doorway that humans ace and machines flunk. The very first idea was beautifully simple and stared you in the face for a decade: take some text and bend it until only a human can read it. Next: read this squiggle.