This is just straight up false. The activation function acts as a conditional, and is very important to making functional AI systems.
I am susprised looping/backtracking isn't used in running the model, but it is used in training it; the difference being that a human learns at the same time they do things, while an AI is trained and run in different steps.