It DOES make a loop, it DOES have feedback, and couldn't work without it. Read the text. Look at how those resistors from the outputs are tied into networks that feed back into the input. I spent an hour finding a perfect, concise example of what you were talking about, and you barely looked at it.
Stop kidding yourself that the feedback in Perceptrons and N-nets used in "small corrections for training process" is somehow incidental or unimportant, if you couldn't train it, it wouldn't work.
To solve NP problems, systems like these are going to need to maintain a superposition of state through the computing elements. Thermal interactions destroy this, they don't stabilize it.