In simple words: Loss measures the mistake; backpropagation shows how weights should change.
Measure the gap
A loss function gives a number for how far the prediction is from the known answer. Different tasks use different loss functions. A large loss means the prediction needs more correction.
Send useful feedback backward
Backpropagation calculates how much each adjustable weight contributed to the loss. It sends that information backward through the network. The optimizer then uses it to change weights. Backpropagation is a calculation, not a person blaming individual neurons.
See the flow
A model predicts a sale chance of 0.2 for a customer who did buy. The loss is high, so training adjusts weights that influenced the low score.
Remember: Loss measures the mistake; backpropagation shows how weights should change.