As long as we take "models" (what are they modeling, btw ?), optimize a set of parameters, and then apply parameters and apply parameters and model to a problem

As long as we take "models" (what are they modeling, btw ?), optimize a set of parameters, and then apply parameters and apply parameters and model to a problem, we shouldn't speak of AI. Real intelligence would have two more ingredients: Memory and a feedback loop. So instead of "training" the parameters and distributing the set, one would have to keep a single instance of an artificial intelligence and feed it the reaction to every problem it tried to solve.