But perhaps one of the most worrying aspects of deep learning is that it can be easily fooled. For example, it is possible to make minor changes to an image, which are imperceptible to the human eye, to fool the model into giving a completely different output.
