>>11902424>but they can't come up with something entirely new that wasn't part of that.Sure they can, the entire goal of modern ML is getting models as generalized as possible. The idea is by training them on such a wide range of things, they can actually identify completely new things that had no relation to the original training data.
>and thus always only shuffle around and value what has been fed to them This is a "troll's truism". It means nothing or everything. Humans can basically be said to do the same thing, but it's like saying "everything is just atoms moving around, bro".
>They also struggle to adapt once drastic changes happen to a previously stable system because the intend they've been designed with often is not to adapt properly BEFORE evaluating.This can happen, but the same is true of many humans. Yet there are also ML models and humans that are good at adapting to drastic changes.
>A machine also doesn't know that there might be something it doesn't knowSure it does. The entire point of ML is classifying things in a general way: to identify something it doesn't already know and hasn't previously seen.
You basically just said the same thing over and over, and none of it's true.
The difference between machine learning and human learning is that human learning contains many other aspects beyond what any existing ML does (like nonlinear regression), so it's much more generalizable. One day, ML will be able to gain those capabilities (though it may or may not be called ML anymore). They just don't exist today or in the near future.
