I have an even simpler one. Machines can’t learn. Machine learning is a term made up of two words that by their very definitions when combined in that order result in a logical contradiction, and a thing which is logically impossible, a learning machine. To put it more simply, if a machine could learn it would no longer be a machine. What you have explained is one aspect of modern computing that attempts to mimic certain theoretical aspects of how a human learns. I say theoretical because there are many competing theories of learning each of which has some experimental support, none of which is settled upon as fact. Since we do not even know how a human learns or even non human animals (at least not in the fine detail that would be required to replicate that system in a computer), there is no way we could create machines that learn. Setting aside the logical problem (a thing one cannot do since logic is a system of universal laws that cannot be violated), how does one create/replicate something if one does not even know what they are creating/replicating? To sum up, all we need do to create a learning machine is violate a logical law, and invent something to do a thing that we do not understand. I have a feeling this could take a few more years of research. lol!