If AI is meant to be so intelligent, why can’t it identify a set of traffic lights? Well… it can. Artificial intelligence (AI) is so powerful today that most CAPTCHA images can be easily solved.
A CAPTCHA (Completely Automated Public Turing test to tell Computers and Humans Apart) is often an image or distorted text that you need to identify or interpret to prove you’re human.
Originally, the idea of image-based CAPTCHAs, named reCAPTCHA, was also to help train AIs to perform text recognition better when digitising books.
Invented by Luis von Ahn (co-founder of Duolingo), poorly scanned words were shown to humans as a CAPTCHA and by identifying them, we taught the AIs.
We no longer need to train AIs this way – they’re more than able to cope. Research reported in July 2023 showed that most can solve CAPTCHA images with 96 per cent accuracy, compared to humans who range from 50–86 per cent.
The AIs are even adept at mimicking humans to fool the bot detectors, by copying our poor accuracy, for example, or even our mouse movements as we figure out which boxes to click.
Yes, today’s reCAPTCHAs are remarkably advanced security systems behind the scenes.
Even the ‘I’m not a robot’ box has multiple levels of encryption as it measures a surprising amount of data about you: your time zone, IP address, screen size, browser and plugins, key presses, mouse clicks, browsing history and things we may not know about.
But will AIs soon be able to fool these too? Yes.
This article is an answer to the question (asked by Steve Hunt, via email) 'If AI is meant to be so intelligent, why can’t it identify a set of traffic lights?'
To submit your questions, email us at questions@sciencefocus.com, or message our Facebook, X, or Instagram pages (don't forget to include your name and location).
Check out our ultimate fun facts page for more mind-blowing science.
Read more:
