[Omnirizon]

this is true. in machine learning we say that we use a data set to `train' a machine to `learn' to predict the class of an instance given a set of attributes. However, the machine isn't trained or learned in any colloquial sense of the word, for it doesn't even know what the data is or what it means.

Consider a robot with algorithms to follow a path and avoid obstacles. First, an algorithm doesn't path or avoid, rather a concept does. The robot doesn't have this concept, rather the programmer does. The robot doesn't know what a path or obstacle is, it is simply following an algorithm. Thus the AI really doesn't know what a path or obstacle is, even though it can follow a path and avoid an obstacle.