14 dec 2017: the blog is being migrated from blogger.com, therefore not all the articles are present yet....

Machine learning and black swans

Machine learning is all about prediction based on historical data. Does it take into account black swans -- unpredictable events that occur out of the blue and are completely random and thus can't be predicted? It doesn't. How accurate is machine learning in general then?

Let's take the famous turkey. For a thousand days it's been fed, but on the 1000 and 1st day it was killed, cooked and eaten. What would a machine learning algorithm make of it, is it able to predict it being killed based on the first 1000 day? Not only is it unable to, but also it'll predict, obviously, that the chance of death is infinitely small, if not zero.