In this first part, we explore the basic assumptions of neural networks in terms of their goal to approximately model some function in the real world, like traffic densities in a city. We state formally what it means to learn and then tie this to the notion of error minimization and gradient descent.