Skip to content

No. 181Technology3 minviews 83

Published by GK Library Annex

How Does One Click on "I'm not a robot" Tell You're Human?

The checkbox isn't the whole test, just the final check of a review that looks at what happens before and after

A finger about to press a glowing empty checkbox button on a dark screen, next to a small toy robot

When you tick the "I'm not a robot" box on a website, you usually pass right away. But this box isn't a test that looks only at the click itself. Google says its reCAPTCHA decides whether you're human by looking at your activity before, during and after you click the box.

CAPTCHA is a reverse Turing test

CAPTCHA stands for "Completely Automated Public Turing test to tell Computers and Humans Apart." Researchers at Carnegie Mellon University in the US gave it that name in the early 2000s.

In the original Turing test, a human judges a machine, but in a CAPTCHA it's the other way around: a machine judges a human. It's meant to stop automated programs from signing up for accounts thousands of times or snapping up every ticket. At first, the common method was to make you read distorted letters and type them in.

Why distorted letters were pushed aside

Because the assumption that machines can't read them fell apart. In a December 2014 post, Google shared its own research showing that artificial intelligence at the time could solve even the hardest kind of distorted text with 99.8% accuracy.

It had become a test that was annoying for people and easy for machines. So instead of letters, Google built a risk analysis method that looks at the whole process of how a user interacts with a site, and the result was a screen with a single checkbox.

The box is only the final check

A common misconception is that you have to wiggle the mouse like a human before clicking to pass. Google hasn't disclosed exactly which signals it looks at; it only says it judges by combining many clues gathered before and after the single click.

If it isn't confident, an extra challenge appears asking you to pick photos containing traffic lights or crosswalks. In 2018, a version also appeared that gives only a score behind the scenes with no checkbox, so these days you often pass without clicking anything at all.

If photo challenges keep appearing

If several people share the same computer, you connect through a virtual private network (VPN), or you send many requests in a short time, the judgment can get harder and photo challenges may appear more often. It usually means there aren't enough clues, rather than that you did something wrong.

In that case, calmly solving the challenge or trying again a little later usually fixes it. But a window claiming to be a robot check that asks you to install a program or paste in a command may be fake, so don't follow it.

The checkbox isn't a test of the click alone; it's the final check of a judgment that also looks at what you did before and after.

Notes and references · Google Online Security Blog: "Introducing No CAPTCHA reCAPTCHA". Dec 3, 2014 post: AI solved distorted text at 99.8%; risk analysis checks activity before, during and after the click. Signals undisclosed; 2018 scoring and VPN notes are general

Editorial standardsReport an error

Quick quiz

In 2014, what accuracy did Google say AI reached on distorted-text CAPTCHAs?

Saved only in this browser.

On the same shelf · Technology

  1. 11.06TechnologyWhy traffic lights are red, yellow, and greenviews 56 · 3 min
  2. 11.05TechnologyBluetooth Is Named After a Danish King's Nicknameviews 60 · 3 min
  3. 11.04TechnologyWhy does a chatbot answer differently when you ask twice?views 48 · 3 min