The following is a demo of the JSE captcha bot prevention widget. This is a client-side only demo but by testing it out you’ll be helping train our AI models which will be used to differentiate real users to automated bots.
https://jsecoin.com/misc/captcha/
The in-script games are modular and can be switched in and out. These little mini games are quick easy to code if we need more diversity to the challenges. They are really there not to test the user but to collect click/touch data which can then be used along with browser variables and IP data to train the machine learning modules.
The server-side code will be built using Tensorflow models based around the client data. These can be updated regularly as more data becomes available forever increasing the accuracy of results. Client-side data passed in will be anonymous and will not contain any sensitive or personal information.
For webmasters this will be a free tool which they can use to protect forms, logins and backend services. We decided against running any mining while the script operates as most of the time a captcha is completed just before leaving a page or submitting a form and the benefits in computational power are negligible.
So what is the benefit to the project and webmasters/visitors?
The project benefits by collecting anonymous IP based data on visitor interaction. Eventually the ability to detect human vs bots will play a big role in preventing 51% attacks for a client-side consensus algorithm.
Additionally we can leverage our existing and new AI models to provide enterprise grade API’s for bot detection based on non-interaction data. This can benefit e-commerce, ad-tech and commercial sites and will form part of our efforts to create additional demand on exchanges.
For the webmaster and visitors we offer an improved captcha experience. By using a range of “mini-games” we feel there is a UX benefit to the current standard of clicking on bridges/traffic lights etc.
Once in production we will be using this on our internal platform login/registration pages.
