do neural networks make sensory more and more abstract and then gives a defined output from that abstraction?

Lol. Neural network documentation is so overcomplicated. Let's imagine you want to make a computer that can play chess. You want to get from the computer not knowing how to play to playing very well but you don't know how to get there. What you can do is simply record the moves people make and record which moves result in a win. Then the computer can figure out the probability of winning based on how frequently a set of moves results in a win. Now we can use algorithms to simulate the people knowing all the possible moves and let the computer pick random moves instead. This is the basic idea of Neural Networks. You have many threads/processes dedicated to running simulations and working out probabilities of success which have been called neurons. You can use either people to input the data (like was used for ocr) or just generate it randomly both are neurons. 

Where this stuff gets interesting is say you have a patient with certain symptoms you could train an AI to read those symptoms and give probable diagnoses based on previous patients with similar symptoms etc.

Edited December 23, 20187 yr by fiveworlds