The paper suggests that software could build a model through black-box analysis of a system. That is to say, there would be a program to create the model of the system without the programmer having to necessarily interact in order for the model to be built. The only real input the programmer would need to provide would be updates to the model itself. This makes the role of testing suitable for educated programmers as they will actually be engineers of the models instead of analysts of the results of a scripted test.