Except that’s not at all how neutral networks function... They literally learn based on example input and output, aka training data, not based on hard and fast rules. So if you trained a neural network to make a fruit salad, it would never think to use a tomato unless your training data included many fruit salads…