Loss function describes how far the results in your network produced is from the expected result. It indicates the magnitude error of your model on its prediction.
You can take that error and 'backpropagate' it through your model, adjusting its weights and making it closer to the truth.
Hope this answer helps you! Studying Machine Learning Tutorials would also give you a brief and valuable explanation of the problem.
If you want to know more about Machine Learning then watch this video: