You’re probably familiar with those annoying CAPTCHAs that constantly ask you to click on all the traffic lights, trucks, or crosswalks to prove you’re not a robot. It’s the very reason why Daft Punk never had a Gmail account, by the way.
These CAPTCHAs, supposedly our last line of defense against a machine invasion, have just been thoroughly trounced by artificial intelligence. Surprised? (Not really.)
Indeed, a new study has shown that AI is now capable of solving these famous puzzles with a 100% success rate!
But before we go any further, let’s briefly explain CAPTCHAs for those who may have been living under a rock for the past few years: CAPTCHAs (which stands for “Completely Automated Public Turing test to tell Computers and Humans Apart”—quite a mouthful) are visual tests designed to differentiate humans from bots on the internet. Google popularized a widespread version called reCAPTCHA, which you’ve likely encountered while browsing the web. And if you’ve ever spent 10 minutes trying to figure out if that blurry pixel is part of a bicycle tire or not, you know how frustrating it can be.
But let’s get back to our artificial sheep, or rather, our artificial intelligences, the new electric sheep. Researchers from the Swiss Federal Institute of Technology in Zurich have developed a system based on advanced deep learning models, notably the famous YOLO (You Only Look Once), capable of solving 100% of reCAPTCHA v2 CAPTCHAs.
For the more tech-savvy among you, they used a fine-tuned version of YOLOv8 trained on no less than 14,000 labeled traffic images.
But how on earth did they achieve this feat?
Here are the ingredients of their devilish recipe:
- A finely-tuned object recognition model: Their AI can identify the requested elements in images with surgical precision, whether they’re traffic lights, cars, or crosswalks, etc.
- A well-oiled attack strategy: To bypass bot detection mechanisms, the researchers used a VPN to change the IP address for each attempt, simulated realistic mouse movements, and even added fake browsing data to appear more “human.
- A tailored approach for each type of CAPTCHA: The team developed specific techniques to solve the different variants of reCAPTCHA v2, adapting to every situation.
The result?
A 100% success rate, even surpassing human performance! In fact, the study showed that their AI solved CAPTCHAs more quickly and with fewer errors than human participants. This raises serious questions about the effectiveness of these tests in distinguishing humans from machines. It seems we’ll need to come up with something new. But don’t worry, it’s not as though robots will be taking over your online accounts anytime soon (or will they?). The researchers conducted this study ethically and, of course, shared their findings with Google and the scientific community. The goal is to improve online security, not to create an army of malicious bots ready to spam every forum in the world (there are probably only three left online anyway!).
So, what does the future hold for CAPTCHAs?
Google has already begun rolling out reCAPTCHA v3, an “invisible” version that analyzes user behavior instead of asking them to solve visual puzzles. Other avenues are being explored as well, such as using more abstract challenges or tests based on contextual understanding. And perhaps one day, to prove you’re human, you’ll need to schedule an appointment with your proctologist, who will provide a certificate to be submitted to Google or Meta.
In the meantime, the next time you’re faced with an especially tricky CAPTCHA, just remember: somewhere, an AI has probably already solved it faster than you.
