After years of not working in statistics I've decided to learn something new, so I've purchased a Coursera MOOC course on that new kid on the block - Machine Le

After years of not working in statistics I've decided to learn something new, so I've purchased a Coursera MOOC course on that new kid on the block - Machine Learning (now rebranded as AI I think). To my surprise (I wanted to learn something new) all the "old stuff" was there, ML turned out to be some variant of statistical inference with all its typical methods applied (like logistic regression, clustering, etc.). The new things were: deep neural networks ("normal" NN but with more layers) and total lack of thinking in terms of model - ML keeps adding more and more variables to the model, as much as data we have, no matter if it make sense or not. If the "model" is over-fitted, then, well, let's add some additional quadratic term to the equation, it seems to be helping... Maybe... Sometimes. So, yes, AI is just old stuff, rebranded and fed with much, much more data. From what I see AI hasn't changed much over the last years. AI seems to be pretty effective in everything that is machine generated data set, as it is easily quantified and has simple representation - That's why it is so good in tracking people - there is a huge amount of data coming from people devices, they are nice and clean. It is also good in picture analysis - again data are pretty clean (pixels) and easily quantified, what is even more important, so it easy to "train" AI. When it comes to something more complicated like, for instance, purchase recommendation we get what we get: you bought a fridge, good, you will see fridge adds for next half a year.