March summary - 2017-03

03 Mars



  • Security & AI
    • The scope is wider, with this insidious trends:
      • AI can find the most efficient way to influence you, analysing your tweets, Facebook and so on ...
    • Breaking Google security with Google services! ReCaptcha broken with AI https://east-ee.com/2017/02/28/rebreakcaptcha-breaking-googles-recaptcha-v2-using-google/
    • Very informed Podcast on the subject RiskyBusiness "Machine Learning: Woot or woo?" https://risky.biz/RB449/ others episodes mention AI also.
      • Key point is that Neural Networks will find solutions that are not human understandable. Of course, you can look at what it does. You have access to all the intermediates steps and all the variables, un-like a real brain for example. But that does not mean that the operations will makes sense to you. Unless you are applying Neural Networks to trivial physical data, like the position, speed, mass of a ball, you will not end-up with an nice equation showing that the speed is proportional to the weight and not to its color for example.  You end-up blindly trusting a system that is somehow better than you. But AI fails in unexpected ways that we might learn the hard ways. (and hackers may learn faster.)
Post a Comment

Online Scientific Notebooks, Course, Visualization show and tell comics

Course, Visualization, Notebooks A long paper on the subject "The Scientific Paper Is Obsolete" (2018) https://www.theatlantic....