GiPOCO CAPTCHA

Completly Automated Public Turing-test to tell Computers and Humans Apart by Gipoco

The story

This project arose from my reading of an article published by a researcher in Computer Science about the state of the art of CAPTCHAs. The purpose was to practically demonstrate that it would have been possible to develop software capable of bypassing the most secure CAPTCHAs at the time of writing the article.

Upon reading the article, I became so motivated that I immediately began conceptualizing a novel CAPTCHA design that would outsmart all the techniques proposed by the researcher. Just a few days after starting this challenge, I already had a proof of concept that was able to stump any automated system.

The next step was to reach out to the researcher and share my work with him. He was very kind and replied promptly. After a brief exchange of informations, he confirmed that at the time his method was not able to outperform my technology.

The solution

I started from the observation that one of the most difficult types of QI tests is the spatial mapping. I thought that by bringing the concept of spatial mapping to 3D, it would be possible to achieve a level of complexity beyond the reach of any Optical Character Recognition system.

I've developed a prototype that uses ray tracing to render a scene in just a few milliseconds, starting from a randomized model generated at each run based on a series of assets (background, 3d objects and alfanumeric characters). From the single scene provided to the candidate, an algorithm generates a sentence in natural language with instructions on how to select only some of the characters that appear in the scene. ("add the letter on the front wheel to the one on the handlebar and... ")

Contacts

After a few years of inactivity on this project, I decided to go back to develop it again, since the currently available open-source multi-modal artificial intelligences are opening up new research avenues, making the complexity of my solution no longer unachievable.

If you're interested in computer vision, in artificial intelligence or in the simulation of a real human user, feel free to contact me via email at my email address info (at) gipoco.com to exchange advices and suggestions on the subjects.