A rough explanation of "phase transition" in regards to insight can be described as follows: At the beginning, each component is relatively weak, so the progress of each component contributes very little to the overall progress. However, as the components develop, other components become more useful, meaning the overall gradient (the speed of learning) increases non-linearly, which is the true mechanism of insight (phase transition).