"If you just have a single problem to solve, then fine, go ahead and use a neural network. But if you want to do science and understand how to choose architectures, or how to go to a new problem, you have to understand what different architectures can and cannot do"

Marvin Minsky is suggesting that if you are just trying to fix a single issue, then a neural network is a suitable option. Nevertheless, if you want to gain a deeper understanding of how to select the best architecture for a problem, or how to approach a new issue, then you require to understand the capabilities and limitations of different architectures. This suggests that a neural network is not constantly the very best service, and that it is essential to understand the different architectures readily available in order to make the best choice. Minsky is stressing the value of comprehending the various architectures in order to make educated choices when approaching a problem.